HomeVideos

Claude AI Full Course (2026): Build Your AI Marketing Team with Claude Skills & Agents | One Shot

Now Playing

Claude AI Full Course (2026): Build Your AI Marketing Team with Claude Skills & Agents | One Shot

Transcript

3801 segments

0:00

you've been using Claude like a smarter

0:02

search engine. You know, you probably

0:03

type a question, read an answer, copy

0:05

some of it. You're only getting about

0:07

10% of what it can actually do. So,

0:10

here's the thing, right? That's not a

0:12

prompting problem. It's not even right

0:14

about writing better prompts. It's about

0:17

having the wrong picture what the tool

0:19

can actually do for you. Now, most

0:22

marketing teams that I talk to probably

0:24

use plot the same way. They open a tab.

0:26

They type something like, "Write me a

0:29

campaign brief for our summer sale."

0:31

They read what comes back to them and

0:34

then spend the next hour turning that

0:36

text into probably a usable document.

0:38

Maybe a deck, maybe a tracker, a report

0:42

that manager can send to a client. Now,

0:46

you know, there's something about the

0:47

entire workflow that gets me going,

0:51

right? Because I know that the output

0:52

isn't bad. Claude's actually pretty good

0:55

at writing, but the process itself is

0:57

completely wrong because you've added a

1:00

step rather than removing one. So, what

1:03

I want to show you today is a different

1:05

way of thinking about this. So, not a

1:07

different prompting technique, but

1:09

rather a different entire mental model.

1:14

So, Claude isn't a writing assistant.

1:16

It's actually a workflow engine. And

1:19

once that clicks, every task currently

1:23

that eats up your time starts to look

1:26

like something you could hand off

1:28

completely. Not just partially but

1:30

completely. And this is the video and

1:33

this is the first video in a series that

1:34

I am calling flawed for marketing. Each

1:38

episode here covers a specific role. for

1:41

content writers, social media managers,

1:43

marketing managers, performance

1:45

marketers, brand managers and shows

1:48

exactly what changes when you stop using

1:51

Claude like a chatbot and start using it

1:54

like the workflow tool it actually is

1:56

meant to be. Today's video is the

1:59

foundation. If the reframe lands right,

2:02

everything else in the series will make

2:04

a lot more sense to you. So, let's get

2:07

into it. Okay, now let me show you

2:10

something concrete first because I think

2:12

abstract expla explanations of this

2:14

problem miss the point entirely. So

2:18

here's a task that most marketing

2:19

managers at least do once a week. You've

2:22

got a campaign brief, right? Let's say

2:24

it's for a brand called Orenda and this

2:27

is a midsize direct to consumer a D2C

2:30

skincare brand and they've got a Diwali

2:33

campaign coming up. Now, the brief that

2:35

they've given you is about two pages.

2:38

It's got the campaign objective, the

2:40

target audience, the key messages, the

2:42

budget, the timeline, the channels that

2:45

they're trying to activate, you know,

2:46

the standard stuff in the industry. Now,

2:49

your job is to turn that into a

2:51

stakeholder deck, something that your

2:53

brand team and the agency can actually

2:56

work from in the kickoff meeting. Now,

3:00

the chatbot version of this task would

3:02

probably go something like this.

3:05

Okay, so the chatbot version of this

3:07

task would probably go something like

3:10

this. So you would open claude, right?

3:13

You type something like here's my

3:17

campaign brief. So you have your entire

3:21

brief here, right? And you would say

3:25

turn this into a presentation outline.

3:30

So then you would just simply say that

3:33

hey you know uh this is my brief and I

3:36

want to create a presentation out of it

3:38

and then you'd probably mention for what

3:41

so let's say for a stakeholder kickoff

3:44

deck. Now you would want to also attach

3:47

the brief so I'm going to quickly do

3:50

that. So, I have the brief in the format

3:53

of

3:55

uh let's say a PDF. Right now, I'm going

3:59

to quickly attach this back into this.

4:03

So, let's say I'm going to do that and

4:06

then hit send. Now, I'm just going to

4:09

quickly wait for this to run.

4:16

So, I have the entire outline right

4:18

here. Now, what you get back is a text

4:21

outline. This is pretty good, right?

4:23

Actually, it's got a great logical

4:25

structure. So, it tells you, you know,

4:28

this is this is going to be on the

4:29

cover. That's the opportunity. This is

4:32

who we're talking to, you know, what

4:34

we're trying to say and the strategy at

4:37

glance, right? And it covers everything.

4:39

So, the campaign overview, audience,

4:42

messaging hierarchy, channel plan,

4:44

timeline, the thinking is absolutely

4:46

sound, right? But then what what do you

4:48

do with this? You probably just copy

4:50

this entire thing into PowerPoint or

4:52

Google Slides and you start building

4:54

slides manually. So you're formatting

4:56

basically text boxes. You're picking

4:58

fonts, adjusting spacing, adding a logo,

5:02

making sure that the title on each slide

5:06

are consistent. And probably let's say

5:08

an hour later, maybe 90 minutes later,

5:10

you've got even a longer brief that you

5:12

just saw right here is not something

5:16

that's broken, right? It works. And if

5:19

someone has been doing it this way for

5:22

years before actually Claude existed,

5:24

they'd say that Claude saved them time

5:27

because it actually did. It gave them

5:29

the structure. But here's the honest

5:31

question that I want to ask you. Did

5:33

Claw do the job for you or did it just

5:36

do one step of the job while you did the

5:39

rest? Well, the job was to not produce

5:42

an outline like this, but rather the

5:45

entire kickoff deck. What Claude gave

5:49

you was just an outline in the chat

5:51

window and you yourself had to sit

5:54

manually and turn it into a deck. So,

5:56

you're the person who did most of the

5:59

actual work and that's the chatbot

6:01

model, right? So claw drafts, you

6:03

produce, and it's genuinely useful, but

6:06

it's only 20% of what's available.

6:09

Here's the shift that I want to make,

6:10

right? Think about the difference

6:11

between two types of people you might

6:13

bring into this project. First person,

6:15

you hand them a brief, they write you

6:18

something, you take it from there,

6:19

right? And the second person, you just

6:21

hand them a brief. They come back with

6:23

the finished deliverable completely

6:24

formatted, structured, and ready to

6:26

send. Both are useful, but they're

6:28

completely different to work with. The

6:30

first one saves you the thinking time

6:32

while the second one saves you the

6:35

production time which is usually where

6:38

most of the hours actually go. Now cloud

6:41

for you can be either of those people.

6:44

Most marketing teams are just focusing

6:46

on using it as the first one. The teams

6:49

getting real leverage from it are using

6:52

it like the second. So what makes the

6:56

difference right? It's something called

6:58

a skill. So let's understand what

7:01

exactly is a skill. So a clot skill is a

7:07

preconfigured workflow. So it's not a

7:09

prompt, right? Uh you must understand

7:12

that a prompt is an instruction that you

7:14

type every time, right? A skill is

7:16

something that you only set once.

7:21

So once you have set it, you basically

7:24

end up using the skill multiple times.

7:27

So it's a trained workflow. So it goes

7:29

from let's say point A up till point B

7:33

then it goes to point C right and then

7:38

maybe it uses C to update B and so on

7:42

and so forth. So it's a it's a complete

7:44

workflow right for a specific job. And

7:48

here you usually have a trigger. So

7:51

something that invokes the skill. So

7:53

let's say an input right. So the trigger

7:56

determines when you need it, why you

7:59

need it, etc. So claude here uses the

8:03

skill and then knows the entire job. It

8:08

knows the output format. So it knows

8:10

what kind of a format

8:13

uh it needs to generate based on the

8:18

input that it receives. So the trigger

8:22

acts

8:23

as the point where you provide the input

8:28

and then it skips has its own number of

8:31

steps. So it goes from A to B to C to

8:35

produce an output that is helpful for

8:38

you. Right? So you give it the input and

8:41

then you directly get the output or the

8:43

finished thing. Now think about what

8:45

that means for this Diwali deck example.

8:48

Right? So instead of writing a prompt

8:51

something like this and asking claw to

8:55

give you the presentation outline, what

8:58

it does is that you upload the brief,

9:00

you trigger the PowerPoint X skill. So

9:05

this skill could be something that's

9:07

already inbuilt and it could also be

9:09

something that you create. And

9:10

throughout this series, we're going to

9:13

do both of them. So in today's video we

9:16

will actually invoke a skill and in the

9:18

future ones we'll see how to write them

9:20

on our own as well right. So here we're

9:23

going to invoke the PBTX skill and plot

9:26

doesn't give you the text right. So for

9:29

at the end it'll actually give you a

9:31

PowerPoint file you know structured

9:33

slides headings content blocks and

9:36

probably a format that can directly open

9:38

in PowerPoint or Google Slides and

9:42

that's the difference right so you have

9:44

the same input but instead of actually

9:47

using a prompt

9:50

what you end up doing is that you use a

9:53

skill and then an output is generated

9:57

from that skill. And now imagine this,

10:00

okay, this is not just limited to

10:02

presentations.

10:04

There are skills for content writers,

10:06

for social media managers, for

10:07

performance marketers working with PDF

10:10

reports, for brand managers who are

10:12

building documentation. Each one is

10:14

trained for a very specific role and a

10:16

specific kind of deliverable.

10:19

And we will cover each of these in

10:21

dedicated videos in the series. But

10:23

today I actually wanted to show you the

10:24

comparison right with the same brief. So

10:28

the brief that we're going to be using

10:29

is the same brief that is here and we'll

10:33

see how it differs from the chatbot

10:37

version and the skill version. So the

10:40

different isn't abstract cuz you can see

10:42

it right here.

10:44

All right. So this is the Orenda

10:46

skincare brief that let's say you

10:48

received right. So this is a Diwali

10:50

campaign brief. uh it's about three

10:52

pages long and this is the kind of

10:54

document that actually lands in a

10:56

marketing manager's inbox right and it

10:59

needs to be created let's say into a

11:01

stakeholder deck by the end of the day

11:03

now before we had seen how to do this

11:07

using the chatbot version which is

11:09

something that you'd probably do every

11:12

day and how most marketers also function

11:15

right so you go here you attach the file

11:18

maybe you know you give it the brief a

11:21

little prompt and then what comes back

11:24

is an outline, right? And this is a

11:26

structured outline for the kickoff

11:28

there.

11:30

Now, um so you know, I'll probably be

11:35

honest, this is what most marketing

11:36

managers are doing. The the output is

11:39

pretty good and claw going to also give

11:41

you back something that's logical,

11:43

right? So, this is solid. you've gotten

11:46

an executive summary and each and every

11:48

outline, a complete outline and each and

11:51

every aspect that you actually require,

11:53

right? And uh but

11:56

this is the whole point, right? You're

11:58

right now just looking at

12:01

text in a chat window. To get a deck,

12:04

you have to open PowerPoint, start a new

12:06

presentation, build this slide by side,

12:08

you know, copying one by one and making

12:11

it look like something that the client

12:13

would actually like. And you know, I've

12:16

timed myself actually doing something

12:17

like this last week, and it almost gets

12:21

me 2 and 1/2 hours to go from this to a

12:26

usable 12 slide to, you know, 15 slide

12:30

deck with just working, you know,

12:34

quickly and having no design tweets. So,

12:38

let's do the same skill with the

12:41

PowerPoint presentation skill, right? Uh

12:45

let's figure this out how to write here.

12:49

But instead of actually writing a prompt

12:51

directly, I am going to invoke a skill.

12:54

So how am I going to do that? I'm going

12:57

to trigger the PPTX skill and get it

12:59

done. So here's how I'm going to do it.

13:02

I'm just going to quickly type out and

13:05

uh say use the PPTX skill create a

13:09

stakeholder kickoff deck from this

13:12

campaign brief.

13:14

So now I have attached it right let's

13:18

say uh instead of highq maybe let's use

13:21

sonnet the model for this particular

13:25

uh task and now let it run and you know

13:31

let's just wait for it to run

13:36

okay so what's happening right now is

13:38

very different from what had happened

13:40

before right so claude isn't generating

13:43

an actual outline for me rather it is

13:46

creating a complete PPTX so a complete

13:50

PowerPoint presentation for me. So it's

13:53

making the actual file. So you can see

13:55

here it says creating the kickoff tag

13:57

and if you go down you see that it's

13:59

writing some kind of code and you know

14:02

it's installing stuff and it's creating

14:05

the outline for uh not just the outline

14:09

but it's also creating the actual actual

14:12

campaign deck that I require. So you can

14:15

see things like font size, you can see

14:17

things like shapes etc. And now it just

14:19

says generate the PBTX. So it looks

14:22

solid now. visual things. It's adding

14:26

some images into it. It's finding some

14:30

other scripts and then it's building it.

14:34

Okay. So, this is what came out, right?

14:36

Um, my tool had actually hit the limit.

14:40

So, I had to press continue so that you

14:42

know it can continue the uh entire

14:45

process. But now I have this right here

14:49

in front of me. This is my complete deck

14:52

that I need to send. So, instead of just

14:54

having, you know, uh the document

14:58

outline and me having to put all of it

15:00

into this, now I've got my title slide,

15:03

my campaign name, my campaign dates, a

15:06

summary of what's going to be on today's

15:08

agenda, I've got the objectives, I've

15:11

got the entire overview, what's the

15:13

budget, what's the campaign window, you

15:16

know, when does it go live, each and

15:18

everything. And all of this probably I

15:21

got it under 3 and 1/2 or maybe 4

15:24

minutes. You see the difference? I am

15:27

saving hours here. This is a 12.12

15:32

slide deck and it can be opened in

15:35

PowerPoint. It can also be added to your

15:37

Google Drive. The formatting is

15:39

absolutely clean. Each and everything is

15:43

from my PDF that I had uploaded.

15:46

Everything is readable. There are some

15:49

touch points that they have mentioned.

15:51

There is allocation with intent that it

15:54

also shows. It tells me what we're going

15:57

to produce the kind of critical

16:00

milestones you know for the campaign to

16:02

be live for 21 days to own Diwali.

16:07

Now from just uploading the brief right

16:10

to having this file I probably just took

16:15

5 minutes. Five. That's all I needed.

16:19

Five. Now, think about this this way,

16:21

right? The only difference is whether

16:24

Claude

16:25

gave me text to work with versus it

16:28

actually gave me the finished thing. And

16:30

that's the skill model. And that's what

16:32

I mean when I say Claude is a workflow

16:37

engine and not a chatbot. It's not doing

16:40

one step of the job and handing it back.

16:43

It's actually completing the job for

16:45

you.

16:47

So the question now at hand is where

16:50

does this apply in your week? Here's the

16:53

way that I like to think about it,

16:54

right? Take any task where you currently

16:56

use CLA to generate the text and then

16:58

you spend time turning that text into a

17:01

deliverable. That gap between Claude's

17:03

output and what you actually need to

17:05

send, that's where a skill lives. Okay.

17:09

So for most marketing teams, these tasks

17:11

actually fall into let's say three

17:14

buckets. The first bucket that we have

17:18

at hand is called content creation. So

17:24

this is the bucket where you're writing

17:27

copy script, social posts, emails. So

17:30

you generate something in claw, then you

17:32

spend time cleaning it up, removing the

17:34

air tone, formatting it for the

17:36

platform. The humanizer skill actually

17:39

handles that cleanup for you. the

17:41

caption generator, the real script

17:42

generator, those handle the output

17:45

format directly for you. The second

17:49

bucket that is at hand is the bucket

17:52

that is related to data and reporting.

17:55

So let's say you're getting PDFs from

17:57

agencies, right? You're pulling numbers

18:00

uh out of reports, rebuilding

18:03

spreadsheets manually. The PDF reading

18:06

skill extracts and structures that data.

18:10

The Excel skill will basically build a

18:14

spreadsheet for you. And then there

18:17

exists the third bucket which is

18:19

presentations and documents. Now this is

18:23

something that I just showed you here.

18:27

So presentations and documents creates

18:31

your decks, your brand guides, your

18:32

client reports, everything. So a PBTX

18:35

skill, a DOCX skill, a PDF output skill,

18:39

all of it can get covered within this

18:44

little framework that I just gave you.

18:46

So if your team does any of those

18:48

things, right, and most marketing teams

18:50

probably do all three of them, there's a

18:52

skill that can change that workflow. So

18:56

here's a practical thing that you can

18:58

actually do um you know this week. So

19:03

let's say think about any one task

19:06

right. So I'm not asking you to think

19:08

about multiple tasks. All I'm asking you

19:11

to think about is that one task that

19:13

takes most of the time to go from let's

19:15

say the claude's output to something you

19:18

can actually use. And that's going to be

19:20

a starting point and that's also going

19:22

to be the first skill that you can start

19:26

learning properly. And it's also going

19:28

to be oh it's also going to be worth the

19:31

first skill that you learn here. Now

19:36

I've linked the relevant episodes in the

19:38

description below. Each one covers a

19:41

specific role and goes deep on that

19:45

exact workflow. So find your role and

19:48

probably you can start right there.

19:50

Here's the oneline version I keep coming

19:52

back to. Right? A skill is basically a

19:54

trained junior in your team for one

19:57

specific job. So you don't manage them.

19:59

You just give them the brief and they do

20:01

it. So that framing matters because it

20:04

changes how you use the tool. You

20:06

wouldn't hire a junior team member and

20:08

then stand over their shoulder the

20:09

entire time rewriting everything that

20:11

they do. You would give them the brief,

20:13

let it run, and then review the output.

20:16

And that's exactly how you use a skill.

20:19

I've put down all the differences for

20:21

you. So it's easy for you to understand

20:24

in a tabulated format. And let's quickly

20:26

run through it. So here's the difference

20:28

between a prompt and a skill. So what

20:32

does a prompt do? In a prompt, you have

20:34

to write the instructions every single

20:36

time. So every time you want to do

20:39

something, you have to write it again.

20:41

You describe what good looks like. So it

20:44

varies. In a skill, on the other hand,

20:47

the instructions are preset. usually or

20:50

when you develop your own skills, you

20:52

set the instructions yourself so you

20:54

know what you want every single time.

20:57

Then in a skill, the quality standard is

21:00

usually built in. So this is especially

21:03

in relation to let's say skills that

21:07

make PowerPoint presentations or

21:09

documents or Excel files. These are

21:12

inbuilt skills. So in inbuilt skills,

21:14

the quality standard already exists. And

21:17

when you're creating a skill, you can

21:18

also set the quality standard. Now, a

21:21

prompt, the output is usually text in a

21:24

chat window. And um in a skill, the

21:28

output is usually a fixed

21:31

a finished file or a deliverable. These

21:35

deliverables are known as artifacts in

21:39

the clawed ecosystem. So these artifacts

21:42

can be HTML pages

21:46

right or SVG files depending on what

21:50

your output is and how you describe it.

21:52

Um sometimes prompts can also lead to

21:56

the creation of a deliverable but it

21:59

depends on what and how you exactly

22:02

define the output. In a prompt, you have

22:05

to do the production work yourself

22:08

because there is no end output. It's

22:13

just text that is given to you. So

22:15

translating it into a PowerPoint,

22:18

um you have to do all of that production

22:20

work yourself. In a skill, the

22:22

production is part of the skill, right?

22:25

Prompts are inconsistent. Basically, it

22:28

depends on how well you prompt it. A

22:31

skill is consistent because once you've

22:33

set the standard, it runs the same way

22:35

every single time.

22:38

So the last row is actually the one that

22:41

matters the most at a team level and

22:43

we'll come back to it later in this

22:46

video.

22:47

All right, let's get on to the practical

22:50

aspect. So we understood the

22:51

differences. Now let's get to the

22:54

practical part. So I've opened Claude,

22:57

right? This is what the interface looks

23:00

like. You might be familiar with the

23:01

chat interface. So it says good

23:04

afternoon chhatanya that's my name and

23:07

here you have different options to

23:09

either write something to create

23:10

something to create code bring something

23:13

from my calendar and bring something for

23:15

Gmail. Now these two are connectors.

23:18

Basically what a connector is is it

23:21

connects with your claw system

23:24

uh to give access to any kind of

23:27

external apps. In the claw interface you

23:30

also have different models that are

23:32

accessible for you. So hiq gives you the

23:36

fastest answers. Sonnet gives you the

23:38

most efficient ones for everyday tasks

23:41

and opus 4.5 is the one which is the

23:44

heaviest model. So this is something for

23:46

let's say developing a website. I would

23:49

probably use the opus 4.7. Now um I'm

23:53

going to quickly switch to highQ 4.5 cuz

23:56

most of the formatting stuff uh can be

23:59

done by IQ 4.5 and it's very low when it

24:02

comes to token usage. Now what is a

24:04

token? A token is essentially the

24:08

currency that Claude uses to generate

24:12

answers for you. So you would want to

24:15

limit your tokens because you don't want

24:17

to exhaust through them quickly. Now up

24:21

here in this interface when you go here

24:25

which says add files, connectors and

24:27

more here you can see something called

24:30

skills. Now I have created a lot of

24:32

skills. I have imported a lot of skills

24:34

that I have found on the internet. These

24:37

are skills that I use on a daily basis.

24:39

Whether it's an AI weekly digest or it's

24:43

a canvas which creates designs and PDFs,

24:46

whether it's content writing, content

24:48

research, whether it's creating a new

24:50

skill in itself. Now, the skill creator

24:52

is an inbuilt skill that comes include

24:55

which lets you create skills for

24:58

specific use cases and that's what we're

25:01

going to also use in this video to

25:04

create a new skill.

25:07

Now u there are certain skills like

25:10

PPTX,

25:13

XLX

25:15

and PDF creator that are inbuilt skills

25:19

which you cannot access from your

25:22

dashboard here. So if you go to manage

25:24

skills, this is what opens. So these are

25:26

all my personal skills if you see right

25:29

and all of them uh humanizer is a very

25:32

famous one that's available on the

25:34

internet. I recommend you use this uh so

25:36

that you can remove signs of AI

25:38

generated writing from your text. Now um

25:43

you can also see skill creator is a

25:45

skill where you can see the method that

25:50

Claude uses to create a skill for you.

25:54

Similarly here you can see the actual

25:56

markdown file or the code. So you can

26:00

toggle between the code and the preview.

26:03

This is the preview mode where you can

26:05

see what exactly it does. So it gives

26:08

you context on how to create a skill,

26:10

how to write the markdown file, you

26:13

know, what are the writing patterns to

26:16

follow, what is the output format, what

26:19

are the examples for the output format,

26:21

the kind of writing styles that are

26:23

required in creating a skill, everything

26:26

comes from here. So what are the use

26:29

cases to create new skills? Modify and

26:32

improve the existing ones. Measure skill

26:35

performance. Use it when users actually

26:38

want to create a skill. Edit and

26:41

optimize an existing skill. Run

26:44

evaluations to test a skill and

26:46

benchmark skill performance with

26:48

variance analysis or optimize a skills

26:50

description for better triggering

26:53

accuracy. This description is the part

26:55

which actually helps you to decide or

26:59

even cla to decide when to use a skill

27:02

in chat. And this is probably one of the

27:06

most important ones, right? And if you

27:10

see the skill creator, it says added by

27:12

who? Added by anthropic. And the trigger

27:14

is usually the slash command. That is

27:16

you do slash skill creator or

27:19

automatically whenever it's relevant.

27:21

Now if I go to some other one like

27:24

caption generator is something that I

27:25

did. So it says added by you when it was

27:29

last updated and how it's triggered. Now

27:32

let's get on to creating and utilizing a

27:36

skill to make a PowerPoint. Okay folks.

27:39

So what we're going to do right now is

27:42

utilize the inbuilt skill to make the

27:48

PowerPoint presentation using the orinda

27:50

brief that I covered in the last

27:52

episode. Right? So I'm going to quickly

27:54

write help me make a stakeholder

27:58

presentation

28:00

for

28:03

proposing or for this brief.

28:08

Very simple. And uh let's see if H HighQ

28:12

does it or I'll have to maybe change the

28:14

mode to set. Now it's going to read the

28:17

campaign brief. It's already done that.

28:20

The next step what it's going to do is

28:23

read the presentation. So it's reading

28:26

the PowerPoint XKL. So I want you to

28:28

understand this. See, it's reading the

28:31

PowerPoint XKL. Now it's installing the

28:35

required scripts for the presentation

28:39

right and then it's running the command.

28:41

The running the command is the command

28:43

that it's running to make the

28:45

presentation itself.

28:48

Now while that's running this is the

28:50

moment where most people's instinct is

28:52

probably adding more instructions. Um

28:55

you have to be absolutely patient with

28:57

this. It takes time to run but it

28:59

delivers one of the best results. So if

29:02

you have the urgency to write something

29:05

like make sure the slides are clean,

29:07

keep it under 12 slides, use a

29:08

professional tone, you actually don't

29:10

need to. The skill already has those

29:13

standards built in. And that's usually a

29:16

sign you're trying to override. If

29:21

uh you tell it not to do it, it'll

29:25

basically override the skill if you

29:28

prompt heavily inside the skill. And

29:30

that's a sign usually where you're

29:33

trying to override it rather than using

29:35

the skill. So take a seat back and just

29:38

relax. And if you see it's running

29:41

commands, it's editing the slide. It's

29:43

verifying the presentation. It's doing

29:46

everything. Now it's giving me the

29:50

entire presenter guide along with the

29:53

deck.

29:54

Now if you see it's presentation

29:57

generation, converting presentation to

29:59

PDF for image interaction, converting

30:02

PDF to individual slides. So it's doing

30:05

each and everything. See, it's doing

30:07

visual inspection.

30:10

It's verifying the final presentation

30:12

slide. And now it's creating the

30:15

presentation. This is a markdown file

30:18

using which I think it'll create a

30:21

presentation uh for this. And now the

30:26

PowerPoint is here for you. So if you

30:30

see here the entire PowerPoint is

30:34

available along with a presenter's

30:37

guide. So a presenter's guide is

30:39

basically telling us about the talking

30:42

points of what decks this deck is not

30:45

and how to present it. And this is the

30:48

stakeholder Diwali deck that will be

30:53

utilized

30:54

for

30:57

yeah that will be utilized for this. Now

31:00

I used highq so the kind of colors that

31:04

they use and everything is a bit um you

31:07

know basic but let's see what happens

31:10

when you upgrade the model with the

31:12

skill. All right folks so now I'm going

31:14

to try and upgrade the model to sonnet

31:16

4.6 six and I'm going to quickly add the

31:19

file the orenda brief right here and I'm

31:24

going to simply write create a

31:28

presentation

31:31

for

31:34

take

31:36

holders

31:38

based on this brief. Let's see how this

31:43

model performs. Um

31:46

see again it says it's reading the PPTX

31:50

skill. So this is the PPTX skill

31:53

actually. Let me quickly show you this.

31:56

Uh by the time it is going through

31:58

everything and creating the presentation

32:01

so it's setting up dependencies and

32:02

doing all of that. So in here you see

32:05

the description. It says you can use

32:07

this skill anytime a pro point X file is

32:10

involved in any way. Input and output

32:13

both. Uh then it says what to make,

32:16

whether to make, how to read and analyze

32:18

the content, how to edit and create from

32:19

a template or create from scratch. And

32:22

then it gives extensive guidelines of

32:24

what it should do. It also says what not

32:28

to do. So what kind of mistakes need to

32:30

be avoid uh avoided in creating the

32:34

actual PowerPoint. It talks about

32:36

textbox padding. It talks about do not

32:39

default to cream or beige backgrounds.

32:42

think about these different colors or

32:44

users brand palette. Now, this is also

32:47

something that you can set inside um

32:51

your claude which I have not yet but I

32:54

will show you how to set that up in

32:57

upcoming videos as well. So, how to set

32:59

your own brand palette

33:02

and then there is other things about

33:06

visual the different sub aents even for

33:08

two three slides. Then it gives code and

33:11

it even talks about converting to

33:13

images. Now you cannot save this skill.

33:16

If you see I will tell you why. Because

33:19

the skill name is already in use. So you

33:22

already need to have another thing. Why?

33:24

Because this is an inbuilt skill

33:28

include. So there's no way that you can

33:30

redo it. Let's check where our

33:32

presentation is. Okay. So it is at the

33:36

part where the Java script so JS uh

33:40

script is being written. So you can't

33:42

see the file yet for very obvious

33:44

reasons. But let's also understand what

33:47

language is the skill written in. The

33:50

skill is written in

33:53

a markdown language. So this is

33:56

markdown. This is another pro uh I

33:59

wouldn't say programming language but

34:01

another document writing language. All

34:04

right. So this is completely created and

34:08

it also hit its limit. Uh this happens

34:12

more often than not because if there are

34:15

any issues it also tries to resolve the

34:17

issues. So there were two issues and

34:19

then the tool usage limit hits. So you

34:21

have to basically click continue so that

34:24

everything gets

34:26

delivered to you. So it's doing

34:28

everything. So it's it's it's literally

34:32

trying

34:33

to create

34:36

a well rendered and complete file that

34:39

is with the correct output. So now it's

34:42

rechecking everything ensuring that

34:45

everything that's given to me it's

34:47

perfect. And now it says that all the

34:49

slides are clean. So it's copying to

34:50

output and there you have it the entire

34:55

Orenda Diwali campaign created for us.

35:01

Brilliant. And you see the moment that

35:03

we actually upgrade

35:05

our

35:08

uh model, the kind of campaign, the kind

35:11

of designing also goes better. So

35:13

designing can even go better if I change

35:17

to opus 3 or even opus 4.6 and uh yeah

35:23

it'll basically create a brilliant

35:26

PowerPoint like this. Now we saw this in

35:28

the last video right? So this is very

35:29

similar to that. But now let me take you

35:34

through the next section which is

35:36

actually understanding what kind of

35:39

skills to invoke when you are working in

35:46

a different role inside your team. So

35:49

the question I get most often at this

35:51

point is how do I know which skill to

35:54

use for a given task? And there are two

35:57

ways to come at it, right? And both of

35:59

them are valid. So you can either choose

36:01

by your role. So say I'm a content

36:02

writer and what applies to me or you can

36:05

choose by output you need. So I need a

36:08

spreadsheet. So what skill produces

36:10

that? So let me walk you through uh

36:12

through that both of them in the next

36:16

segment.

36:19

Now let's first understand what skill to

36:22

use when according to the different

36:25

roles. Right. So, a content writers,

36:29

your main skills are going to be

36:31

something like a humanizer and the dog

36:33

X. So, humanizer strips the AI tone out

36:37

of copy and makes it sound like an

36:40

actual person wrote it. On the other

36:43

hand, the dox skill takes your content

36:47

and makes it into a clearly formatted

36:49

document that can be sent ahead to

36:52

either clients or to the various

36:55

stakeholders that require this copy,

37:00

right? Whether it's for a website or if

37:04

it's for a real or if it's a blog,

37:07

whatever it is, you can send it ahead in

37:09

a properly formatted document.

37:12

For social media managers, caption

37:15

generator is a very simple skill that

37:19

you can create using the claude skills

37:22

create especially the skill uh creators

37:25

the skill creator skill. Wow, that's a

37:27

mouthful. But the skill creator skill

37:30

and it takes a script or a topic as an

37:34

input and it makes ready to post

37:38

captions according to the different

37:40

platforms that you need to post it

37:42

across. A real script generator is

37:45

something that I made to help me write

37:48

and produce short form video scripts in

37:50

a specific length and format. So

37:52

whatever suits me, that's the way I

37:55

write it. Both of these get full

37:58

episodes in this series. So don't worry

38:01

about it. We'll cover both of them

38:03

extensively.

38:04

Now for marketing managers, right, what

38:08

is it that is your biggest headache?

38:10

That's creating decks and making

38:11

trackers and budget sheets. So there is

38:15

a PowerPoint X deck uh creator that we

38:18

just saw in action. Similarly, there is

38:21

an Excel file creator and reader. So it

38:25

handles your data. It creates structure

38:28

for you and gets back a working

38:31

spreadsheet in the right format and this

38:33

is important if you're somebody who

38:35

handles data analysis for your marketing

38:39

team on a daily basis. Now a performance

38:41

marketer again XLS the Excel file is

38:44

something useful because a lot of your

38:46

workflow also involves PDFs. So reading

38:49

PDF reports from different agencies or

38:51

platforms, right? Whether it's invoices

38:53

and the PDF reading skill is

38:55

specifically designed to extract and

38:59

restructure this kind of data. And once

39:01

you pair it with the Excel skill, you

39:05

have replaced the copy paste grind from

39:07

all the agency reports. Now

39:11

the part that is important is the brand

39:13

manager skill. So you can have docx for

39:16

guidelines that you would want to create

39:18

to tone documents to create briefs. You

39:21

have PDFs to make client ready reports

39:24

and deliverables and you can create your

39:26

own brand voice skill that can help

39:31

decide the formatting of all the

39:33

documents that you have to send ahead

39:36

and all the creatives that need to be

39:38

put out. So according to role these are

39:41

the skills that we think are going to be

39:43

useful for you. Now let's go to output

39:47

type.

39:50

All right by output type there are a lot

39:55

of things that skills can do for you. So

39:56

if you know what you need to produce but

39:58

you're less sure which skill hand

40:00

handles it, you can work backwards. So

40:03

for example, if you want to create a

40:05

PowerPoint,

40:06

you can use the PowerPoint skill. To

40:10

create a document, you can create a DOCX

40:13

skill, right? And if you want to read

40:18

Excel files, you have this, right? If

40:22

you want to understand a PDF that is

40:26

read it or write a PDF you have the PDF

40:29

reading and also the PDF writing skill.

40:33

Now all of this is there if let's say

40:35

you want to humanize data right that is

40:40

humanize your writing you have the

40:43

humanizer skill. If you need a script

40:46

then you have the real script generator.

40:48

If you need a platform

40:50

uh specific caption generator, you have

40:52

that. Now, all of this is available, but

40:56

there's one honest caveat here. That is

41:00

some tasks may span across skills. So,

41:04

what does that mean? It basically means

41:07

that instead of one job being fit into a

41:11

particular box of a skill and pretending

41:14

that every job fits neatly in one box,

41:17

we are trying to understand that all of

41:20

these can work across each other. So let

41:23

me give you a quick example. Let's say a

41:24

campaign report comes in and it's in a

41:26

PDF format, right? And then it needs to

41:30

be written in the form of a docx, right?

41:33

That your team can edit. So it is going

41:35

to touch two skills. One PDF reading and

41:39

two dockets to create the actual dock.

41:42

We cover those multi-kill chains in the

41:44

later episodes. But for now, if you're

41:46

just getting started, pick one skill,

41:48

one output type, and get the pattern

41:50

down. Then you can start chaining

41:53

better. All right. Before I get into the

41:56

demo and the practical side of things, I

41:58

want to actually spend 2 minutes naming

42:00

the problem specifically because most

42:02

people know AI copy sounds off, right?

42:04

But they can't always tell why. And

42:07

knowing why makes everything that

42:09

follows make so much more sense. So

42:12

there are four patterns that show up in

42:14

almost every piece of AI generated copy.

42:17

Once you can name them, you'll probably

42:19

spot them in 3 seconds. So if you look

42:21

towards here the first is vocabulary. So

42:25

there are set of words and phrases that

42:27

AI uses constantly and almost no human

42:31

writer does whether it's writing

42:33

something like delve into it is worth

42:36

noting X underscores Y in today's

42:41

rapidly evolving landscape right instead

42:44

of writing in today's landscape or

42:47

something like pivotal moment. If you

42:49

see any of these, you're probably

42:50

looking at an AI copy. So, beware of

42:54

that. Next is sentence rhythm. So, what

42:59

exactly is sentence rhythm? Well, the

43:02

easiest way to understand this is

43:03

actually this graphic that was given by

43:06

Gary Pvost in his 100 ways to improve

43:09

your writing. This is an excerpt for

43:10

that from that. And if you see here, if

43:13

you read a paragraph of AI writing and

43:16

count the sentence lengths, right?

43:18

They're almost identical, whether it's 8

43:20

words, nine words, 10 words, 8 words.

43:24

Well, you can see it something like

43:26

this. So, this is how the excerpt goes,

43:28

right? This sentence has five words.

43:30

Here are five more words. Five word

43:32

sentences are fine, but several together

43:34

become monotonous. Listening to what is

43:36

happening. The writing is getting

43:38

boring. The sound of it drones. It's

43:40

like a stuck record. The ear demands

43:42

some variety. You see how monotonous

43:45

that can get, right? But you know how

43:47

actually humans write? Human writing

43:50

varies. A short sharp sentence maybe

43:52

followed by a longer one that takes its

43:54

time getting somewhere. Maybe changes in

43:57

directions partway through lands

44:01

differently than you expected. Right?

44:03

And AI writing is metronomic. So that's

44:06

what makes it feel very flat even when

44:08

the information is right. If you see

44:10

here uh what Gary Pos says is now listen

44:14

I vary the sentence length and I create

44:16

music

44:18

the writing sings it has a pleasant

44:21

rhythm a little a harmony I use short

44:24

sentences and I use sentences of medium

44:27

length and sometimes when I am certain

44:29

the reader is rested I will engage him

44:31

with a sentence of considerable length a

44:34

sentence that burns with energy and

44:36

builds with all the imp impetus of a

44:39

crescendo, the role of the drums, the

44:42

crash of the symbols, sounds that say,

44:45

"Listen to this. It is important." So,

44:47

write with a combination of short,

44:50

medium, and long sentences. Create a

44:52

sound that pleases the reader's ears.

44:55

Don't just write words, write music.

44:59

Well, it's a brilliant piece of writing.

45:01

Um, but I think you've understood what

45:03

sentence rhythm is from this. Now, the

45:05

third aspect that AI covers is actually

45:07

inflation, right? So what does that

45:09

mean? AI has a specific habit of ending

45:11

paragraphs by stating that whatever it

45:14

just said is important. So this

45:17

underscores the significance of X for

45:20

brands seeking to Y. The sentence adds

45:23

no new information. It just points all

45:25

the previous sent it just points at the

45:28

previous sentence and says that that

45:31

sentence mattered. See humans don't need

45:33

to write like that, right? We trust the

45:34

reader to decide what mattered and what

45:37

did not matter. And the fourth rule in

45:41

this entire fiasco is actually the rule

45:44

of three. Now AI defaults to distinct

45:48

things in groups of three, right? So

45:50

whether it's apple, bananas, oranges,

45:52

cars, bikes and trucks, eating,

45:53

drinking, dancing, everything, right? So

45:56

without variation, it just goes always

46:00

in a rule of three. You'll notice it

46:02

when you're reading a piece and every

46:03

single point comes across in a trio. So

46:07

that's usually actually a finding its

46:09

comfortable pattern rather than a writer

46:11

making a structural choice. Um now that

46:15

we understand the background right the

46:17

humanizer skill by a blader is actually

46:19

trained specifically to catch all four

46:21

of these. It's not a style preference

46:23

tweak rather it's a systematic pass on a

46:26

documented list of AI writing patterns.

46:29

It's a moment I'll show you uh where and

46:34

how to do it. So in a moment I'll show

46:37

you exactly what it does to a real piece

46:39

of copy. So let's get into it now.

46:43

All right. So you can see here I'm going

46:45

to use the same brand as Orenda last

46:49

time. Uh the one that we've used across

46:52

our series and I'm going to ask clots.

46:54

This is standard chat interface to write

46:57

a blog post. So same brand and I'm going

47:00

to ask to write an article that's aimed

47:03

at marketing managers at D2C. So if you

47:05

see that the audience that's marked here

47:07

is marketing managers at D2C brands. The

47:09

tone that I want to pick is

47:10

authoritative and it should be titled

47:13

five reasons why D2C skincare brands are

47:16

losing customers in their 30s and what

47:18

to fix. Let it run. This will uh rightly

47:23

do it. So right now it's picking up my

47:25

content writer skill. So this is a skill

47:28

that I had developed to write content

47:32

right uh to write exactly you know what

47:35

I want and how I want it. So it's

47:38

writing it from that perspective. Now

47:43

uh let it's quickly generated the post

47:46

and you can see it here. So it's given

47:49

five exact points to fix. So your

47:52

product scale claims don't scale with

47:55

maturity. Your pricing strategy assumes

47:57

loyalty that doesn't exist. You're not

48:00

building this. Your content strategy

48:03

treats aging like a taboo. You're

48:05

chasing Tik Tok trends instead of

48:07

building a habit. And it quotes a real

48:11

problem. Right? All of this again is

48:14

written using the content writer skill

48:16

that I have. But this is still not

48:20

humanized. So how will I use the

48:23

humanizer? Right? So the humanizer skill

48:26

is actually not going to change the

48:27

structure of the entire blog that's

48:31

given to me. It's going to only change

48:36

the language. So let me do one thing

48:40

quickly. Invoke the humanizer skill and

48:44

say humanize

48:48

this

48:50

blog for me.

48:52

So now it's going to humanize this blog

48:55

and it's going to understand what all is

48:59

required to be rewritten. And now let's

49:04

quickly see the difference between the

49:06

two of them and we'll understand this by

49:11

placing them on a word doc. So I'll

49:14

quickly do that and then I will explain

49:16

it to you. Right? So first it's taken

49:19

the entire part of the rewrite and then

49:22

it says what obviously makes the below

49:26

so obviously generated. So words like

49:29

this isn't random, it's systemic. The

49:31

artificial triplet rhythm random,

49:34

systemic, expensive that feels

49:36

constructed rather than discovered. It's

49:38

costing you serious money. It's a vague

49:41

threat to maintain engagement. Here's

49:43

where you're falling, right? a bulleted

49:45

list setup and fix it feels like formula

49:48

like

49:50

the significance aspect that we talked

49:52

about right the overestimation the

49:54

entire value proposition breaks down

49:56

overestimated significance even heavy

50:00

use of m dashes short punchy so

50:03

basically it tells you what all is the

50:06

problem and how to fix it and then it

50:09

fixes it for you so it starts off by

50:12

saying your acquisition engine works

50:14

You're pulling in 22 year olds with

50:16

Instagram ads about dewy skin and 0.5%

50:20

retinol, but somewhere between year 1

50:22

and year 3, they're gone. The customers

50:25

who found you at 25 don't come back at

50:27

30. And unlike losing them to a

50:29

competitor, this one's quiet. They just

50:32

stop showing up. It's predictable and

50:34

it's expensive. The D2C model was built

50:37

for fast product cycles. Trends come and

50:40

go. People swap formulas constantly.

50:42

novelty drives purchases. Then your

50:44

customer turns 30 and the whole setup

50:46

stops working. She's not chasing the

50:49

next viral serum. She wants something

50:50

that actually works and it won't deliver

50:53

it. She'll find someone who will. So

50:56

once you don't deliver it, she'll find

50:58

someone who will. And then after that,

51:00

it gives an entire thing that's already

51:01

humanized and it'll tell you what kind

51:03

of changes it made as well. So that's

51:05

how you use the humanizer skill. And

51:08

this was just a run through of how you

51:12

can use the humanizer skill. Now let's

51:16

turn this copy into something that

51:19

someone can actually open. So let's uh

51:23

get on with the next segment that is

51:25

using the docx skill.

51:32

Okay, back to this now. Now we're going

51:36

to make this into a dog. So I'm going to

51:37

quickly write create this into a

51:40

properly formatted doc for me and I'm

51:43

going to save this here. And now it's

51:47

going to basically take this entire

51:49

thing. It's going to read the docx skill

51:51

to understand word document formatting

51:53

best practices and create the word

51:55

document for me. Now uh there is you

52:00

know maybe some kind of formatting notes

52:02

that you would want to add. I have not

52:05

added those yet at this stage. Actually,

52:09

I should have already done that. So, how

52:11

about we go back here and write some

52:14

formatting notes. So, let's say blog

52:18

post format

52:22

H1 for

52:24

the article

52:27

title.

52:30

Article title. Let's say uh H2 for each

52:36

of the five

52:40

five

52:42

section headings,

52:46

body text

52:48

for everything

52:50

else

52:52

and full

52:54

article

52:57

content.

52:59

Okay. Uh there are some typos in this

53:01

but please bear with me. But now it's

53:04

going to create an entire thing that's

53:06

properly formatted with the blog post

53:09

using the docx library. It's going to

53:11

use each and everything that is required

53:13

and it's also going to take reference of

53:16

the little uh you know notes that you

53:19

have added. So I've added a formatting a

53:22

brief formatting note right and that's

53:24

all it needs. So let's wait till it's

53:28

generated and let's see how it turns

53:31

out.

53:34

All right folks, you see this? This is

53:37

what it is. Now I have the entire blog

53:40

in a docx file that's properly formatted

53:44

with each and everything. So this is in

53:47

H1, the five D2C brands. The five

53:51

sections as you can see is H2

53:54

heading. You can see the document

53:55

outline on the left if you're in Word.

53:58

And the body text is clean throughout.

54:01

There's consistent spacing. There's no

54:03

inherited, you know, formatting issues

54:05

from pasting into a blank document. Now,

54:10

if I send this to a client or a brand

54:12

team today, right, this is ready. This

54:14

is the final read through. Maybe one or

54:15

two tweaks and then directly goes out.

54:17

So from a brief that was here that is I

54:22

want to just write a blog post all the

54:25

way up till here what we've done is

54:28

create an output that is uh that is an a

54:32

dox file right so now how do I apply

54:38

this this to my ecosystem is I'm going

54:42

to help you create actually a content

54:45

writer skill. Now, if you see here, I

54:48

already have a skill that's a content

54:51

writer. So, this skill drafts fulllength

54:55

marketing and business content

54:56

specifically for posts, articles, etc.

54:59

Now, let's say I did not have that

55:02

skill, right? So, how can I acquire this

55:05

skill? Well, there are two ways to go

55:07

about it. One is to search for it online

55:10

and the second is to develop it

55:12

yourself. So let's say I use something

55:14

like a skill creator and I'm going to

55:16

ask

55:18

it to write content uh for let's say a

55:21

particular task. So I want to create

55:24

let's say

55:26

help

55:28

me create a skill

55:31

that

55:33

writes

55:34

blogs

55:36

on

55:38

let's say makeup products that's the

55:39

industries I work in right on makeup

55:43

products

55:45

as a D2C

55:48

C brand

55:50

okay so now it's going to involve poke

55:53

the skill creator skill and it's going

55:55

to help me make a skill for this

55:58

particular thing. So let's say writing

56:00

makeup product blogs let me first upload

56:04

and understand the best approach. Now

56:07

they already have this and now it's

56:09

going to ask me some clarifying

56:10

questions, right? So let's quickly

56:12

answer the clarifying questions. The

56:14

first question that it's going to ask me

56:17

is what is my primary use, right?

56:19

Whether it's to drive SEO traffic,

56:21

whether it's to build brand authority or

56:24

product education. Let's say I am going

56:26

to look at product education, right? Um

56:32

how much product research or context

56:34

should the skill assume Claude has?

56:36

Should it product from your actual

56:37

product data or work generally

56:40

generically for any model product you

56:42

name? Let's say I will go here work

56:46

generically and I want it to be medium

56:48

form. Right? So now I have given it all

56:51

the necessary inputs and the parameters

56:53

that it requires to build the skill and

56:56

now it's going to ask me some more

56:59

questions about my brand voice. So let's

57:02

say I want to make this inclusive and

57:04

empowering and I want to also make this

57:08

more stepby step right and uh the

57:13

product page copy should be more of an

57:17

educational blog of how to actually use

57:19

the product rather than make it

57:21

outrightly uh you know

57:26

outrightly in your face to buy the

57:28

product and very CLC. Now, what's my

57:30

preferred blog structure? Well, should

57:32

it be lean, direct or include contact

57:36

sections? Let's say I want a hook and

57:41

yeah, as simple as that. And yes, always

57:44

include SEO. And what should be the

57:47

assumption that is this?

57:50

Yeah. So, I've just like picked up some

57:54

things and now it's going to draft my B

57:56

D B D B D B D B D B D B D B D B D B D

57:56

B2C blog skills. So, product blog skill

58:00

like this. And you can see what it's

58:02

actually doing, right? If I open that up

58:04

and while this is going on here, uh let

58:08

me quickly

58:10

get back to this once it's done.

58:16

Okay. So if you see it's also doing some

58:19

test prompts on the skill. So it's taken

58:22

some test cases. It's verifying if the

58:25

skill is actually readable and it's

58:27

taken a test prompt uh as a foundation

58:31

and now it's actually writing it of how

58:35

to apply foundation on your face. So it

58:39

says skin blurring foundation a

58:41

step-by-step guide. This is the meta

58:42

description. This is the target keyword.

58:45

you know what it does actually etc. And

58:48

now it's also doing a skill summary and

58:51

quality assessment of whether the blog

58:53

that it wrote is actually good or not

58:56

and if it does something that's

58:59

important um in terms of SEO or not. And

59:03

once it's done it's actually basically

59:05

testing out all the different products

59:08

that you know uh can be used. So it's

59:11

also testing it out on a different

59:13

product type. But this is how you make a

59:17

skill that's particularly for let's say

59:21

your

59:22

use case. So here I have taken the use

59:25

case of makeup products. The one before

59:26

that I taken the use case of skin care.

59:31

You can use it for any segment that

59:34

you're in. And this is how to make this

59:37

more concrete rather than just something

59:39

you watch right. So I want you to find

59:41

uh uh find a use case in your particular

59:45

brand where you have to do a lot of

59:46

writing. So you find that one piece of

59:50

AI generated copy that you've already

59:52

edited this week. Something you ran

59:53

through claude and then manually cleaned

59:55

it up. You find the original draft and

59:58

the end draft and then you can run that

60:02

original through the humanizer and

60:04

compare the output of what you actually

60:07

understood uh what you edited yourself

60:10

and what you needed

60:14

the humanizer to edit for you and

60:16

compare both of them. So meanwhile I've

60:20

seen the makeup blog editor is ready and

60:24

it's given each and everything checklist

60:26

when to use it when not to use it right

60:29

the kind of iterations required why it

60:32

feels off you know the quality checklist

60:35

etc everything is ready now you can

60:37

simply save the skills I'm not going to

60:39

save the skill because obviously I am

60:40

not you know using it for any particular

60:44

um product cuz I don't have a makeup

60:46

company but if I were to I would

60:48

probably look at this practical skill.

60:51

Okay, before the demo, I just want to

60:53

spend about 2 minutes on why generating

60:56

captions in a plate cla chat isn't going

60:59

to solve this. So when you ask Claude,

61:02

right, with no skill, just in a standard

61:04

chat to write you an Instagram caption

61:07

and maybe also look at a LinkedIn

61:09

caption for the same topic, what you get

61:12

back actually is just the same copy but

61:16

applicable

61:18

in just two different rappers. So

61:20

instead of having entirely different

61:22

things that the same text can focus on,

61:26

um it just basically says

61:29

let me change the language and hope that

61:32

it works for both the platforms. Right?

61:34

So it uses the same sentence structure,

61:36

the same tone. It's roughly the same

61:38

length with a few words trimmed here and

61:40

there. and Claude doesn't know by

61:42

default that Instagram and LinkedIn are

61:44

actually read completely differently

61:46

because we all know that Instagram is a

61:48

platform where everything is a bit

61:51

casual but on LinkedIn you have to be

61:54

much more professional in nature right

61:58

so that's the difference so Instagram is

62:00

probably for somebody who's just

62:02

scrolling right the thumb is moving and

62:04

you have about half a seconds to stop it

62:06

the caption that works there would

62:08

probably be something that's thought,

62:10

right? And that's very specific, right?

62:15

A statement that creates maybe a jolt of

62:17

recognition or curiosity before they

62:19

even decide whether they want to keep

62:20

reading or not. And then it earns the

62:23

rest of the post. Now on LinkedIn is the

62:26

other way around. The person reading it

62:29

has paused because they actually want to

62:30

read the post. They're in a professional

62:32

mindset. they're reading it for the

62:34

relevance part and they'll give you two

62:36

or three sentences before they decide

62:38

whether they want to continue reading or

62:41

whether they want to discard it. Now the

62:44

pose that works here probably opens well

62:47

with an insight and not the hook. So it

62:50

assumes slightly more patience and

62:53

slightly more context. So it's the same

62:56

idea with completely different opening

62:58

moves and that's the difference. So

63:00

which platform content writers develop

63:03

an instinct for over the years is what

63:07

the caption generator skill is going to

63:08

be trained on. So it just doesn't

63:11

reformat the copy. It actually rewrites

63:13

the reading context of it. And we're

63:15

going to do that for Instagram LinkedIn

63:18

X because this has again a character

63:21

limit of about 270 characters, right? if

63:24

you don't have the premium and more uh

63:27

if you do have the premium version and

63:29

YouTube let's say you are posting one

63:32

video across Instagram, YouTube,

63:34

LinkedIn and X how is the caption going

63:37

to vary then that's also something that

63:39

we're going to look at. So let right so

63:43

let me show this to you and let's get

63:45

right into it.

63:48

All right, folks. So, let's begin with

63:51

actually creating a

63:55

skill

63:57

for a real script generator. Now, this

64:02

is going to be a real script generator

64:06

that can also work for YouTube shots.

64:10

So, basically I have two aims. One

64:13

taking an idea. So it takes an idea and

64:18

generates

64:20

two scripts.

64:24

One I'm going to say YouTube shorts

64:29

and one for let's say Instagram res.

64:34

Right?

64:36

The focus

64:38

should be on

64:44

generating

64:47

scripts

64:50

in a hook

64:53

content

64:55

CTA style. So I've given it a style as

64:58

well. I have a hook, then I have the

65:00

content, then I have the CTA style

65:03

for Orenda

65:06

Skinare. Okay, I am attaching

65:10

a brief

65:12

of a campaign

65:14

that we did some time ago.

65:19

Please take a look and then ask me

65:25

clarifying

65:27

questions

65:29

for the same. So now it's going to ask

65:33

me clarifying questions as well. I'm

65:35

going to quickly change to sonnet so

65:36

that it does better thinking and then

65:40

we're going to go here give it more

65:42

options and within 5 minutes

65:45

approximately

65:46

I am going to

65:51

oh s me I forgot to actually attach the

65:54

orenda brief so it's going to say seems

65:57

like you haven't uploaded the file yeah

66:00

that's all right so we have uploaded the

66:03

file Now I've added the Orenda skincare

66:06

brief as well. It's a solid Diwali

66:09

campaign. So now it's going to ask me

66:11

some questions. Uh so you see that the

66:15

questions it's asking me are very

66:17

specific. So instead of asking these

66:20

questions like this, I'll be like ask me

66:23

in MCQs

66:26

where I can select because I'm lazy. I

66:29

do not want to type anymore. I can even

66:31

ask it to ask me in an MCQ format where

66:36

I can just directly select. So now it's

66:39

going to ask me in my questions in a MCQ

66:44

format that's an interactive artifact

66:46

that comes out of it. So

66:51

what's my brand voice? So while this is

66:53

generating

66:55

uh let's just wait while it's going on.

67:02

Okay, so it's asking me questions now

67:05

about so it's generated all the

67:07

questions, right? And let's say let's

67:10

quickly pick these out. So what is

67:12

Renda's brand voice? Let's say it is

67:15

more

67:17

aspirational and premium kind of script

67:19

length I'm targeting. YouTube shots is

67:21

going to be 30 seconds. Instagram res

67:23

are going to be 15 seconds. What CTA uh

67:28

drives viewer? What should the CDA drive

67:30

to view? Um, I'm going to say shop for

67:32

links and maybe also visit our website,

67:36

right? And save the video because that

67:39

generates engagement. Let's say I am

67:42

going for CTA differ between. Yes, they

67:46

do differ per platform. Which product

67:48

should the script cover? Uh

67:52

let's say skill asks user each time name

67:55

products mentioned

67:57

in this script.

68:00

Uh maybe you know let's make that answer

68:03

hook style should the skill default to

68:07

question based bold statement generate

68:09

all three and maybe let me pick one

68:11

should YouTube shorts and Instagram res

68:14

feel totally different yes

68:18

that is true any claims that are off

68:21

limits no competitor mentions let's say

68:24

I want to say stuff like anti-aging cuz

68:27

that's too much in the market and also I

68:29

want to make no medical claims, right?

68:31

Um cuz legality problems. Now all of

68:34

that gets spilled into it and now the

68:37

skill creator skill is going to

68:39

understand how to build a proper skill

68:41

and it's going to build me a real script

68:43

skill. Now I did this specifically for

68:46

Orenda using a campaign brief that I had

68:48

used in my previous videos. But what you

68:51

can do is you can make one for your own

68:56

brand, the brand that you are looking at

68:58

or you could make it campaign specific

69:00

by saying that every time I add a

69:03

campaign brief or I attach a campaign

69:05

brief with the idea of I want to make a

69:08

real hey give me invoke this skill that

69:13

gives me an overall line and structure

69:16

of what can be done. So now it is

69:19

creating that real script generator. So

69:22

if you see it's writing the entire

69:24

markdown here

69:26

and I'm going to quickly wait for it to

69:29

be done so that I can show this to you

69:31

and then I'll save this as well and I

69:34

will invoke it and show you how exactly

69:38

it works. So let's take a moment till

69:41

it's done and uh then I'll quickly show

69:45

you. See now it's going to create a

69:48

brand voice reference file as well cuz

69:51

we have given the brand voice here to be

69:53

aspirational and premium. So it's going

69:55

to say everything of that sort and if

69:58

you see it's actually picking from the

70:03

from the campaign that I had put because

70:06

that campaign was about Diwali. So if

70:08

you see Diwali ready bridal season of

70:11

celebration the night before everything

70:13

it's also using like festival and

70:15

occasional language if festivities are

70:18

there right so the campaign positioning

70:20

skin that celebrates extends across all

70:23

festivals and occasion this is not a

70:25

Diwali lens so it's giving an entire

70:28

thing now it's a 500 line so it's going

70:31

to do a test prompt and it's given me a

70:35

test prompt as well check the line count

70:37

of the skill which is fine and now it's

70:41

asking me to simulate something and it's

70:44

given

70:46

an entire thing. So it's doing it. It's

70:50

packaged the skill and it's given it to

70:53

me. So now the only thing that's left to

70:55

do for me is to save this skill. It's

71:00

also telling me here's what all there

71:02

is. And let's say I go save the skill.

71:06

So now this skill is saved. And if you

71:08

go to my manage skills, I have a lot of

71:11

skills uh from what I actually do. But

71:15

here you can see Orenda real script here

71:18

is here. The entire skill lives here.

71:22

And you can see revisions whatever you

71:26

know different things of collecting

71:27

inputs is here. So now let's quickly go

71:30

back to the panel and see how we can

71:34

invoke the skill and utilize it.

71:39

All right folks, so let's quickly invoke

71:43

the skill. If I put a slash, you see all

71:46

my skills here, but the one I'm trying

71:48

to show you is the Orenda real script.

71:51

And now I'm just going to invoke it. I'm

71:53

not even going to give it any context.

71:55

Just call on that skill. And now it's

71:57

going to read the real script skill and

72:00

it's going to ask me some questions. So,

72:04

so let me collect inputs as specified.

72:07

So, it's now going to ask me questions

72:09

and it's going to take inputs. So, here

72:12

we go. Let's say what's the idea or

72:15

topic for this script. Um, let's see. I

72:18

want to go for a daily skincare ritual.

72:20

And maybe this is, you know, here is the

72:24

point where I can tell the product. So,

72:26

let's say I want to keep it general,

72:27

right? The hook should be a relatable

72:30

scenario. And let's see what it writes

72:33

for us. Most probably is going to write

72:36

something that is going to be pretty

72:39

good. So, here are the two scripts for

72:42

the YouTube shorts. That's going to be

72:44

30 seconds.

72:46

Some mornings you rush through it,

72:48

cleanser, something something out the

72:50

door. But a ritual, even five minutes,

72:52

works differently when the ingredients

72:54

are doing what they're supposed to.

72:56

Saffron and rose do not need 12 steps.

73:00

They need consistency. Your skin

73:02

responds to what you give it every

73:03

single day, not just a right before

73:05

something matters. So build yours.

73:08

Explore the full range at the link in

73:10

bio. Right. The next one is a 15-second

73:14

2 minutes every morning. The same ones

73:16

every time. That is what your skin

73:19

remembers not the right not not the one

73:22

night you went all out the ritual save

73:25

this for the morning you decide to start

73:28

save right so that's the kind of CTA now

73:32

let's say I

73:37

want to tweak this to be a little bit

73:39

more better how would I do that so on

73:42

GitHub exists a humanizer skill that's

73:45

what I'm going to use and something that

73:46

I referred in my last video. This is

73:48

Blader's humanizer. And I'm going to say

73:51

humanize

73:54

these

73:56

so that they feel like

74:01

it's butter

74:03

flowing through.

74:06

So what does butter flowing through here

74:08

mean? Basically, I want it to glide

74:11

smoothly the way that I'm speaking. So

74:14

that's why butterflowing um this is uh

74:18

sometimes you write things like this so

74:21

that it appears more conversational. So

74:26

all right hook most B most most mornings

74:28

it's barely a routine cleanser and

74:30

whatever close it's an out the thing

74:33

about saffron and rose is they don't ask

74:35

for more time they ask for the same time

74:38

every day 5 minutes done with the

74:40

intention beats 30 done in a hurry once

74:42

a week your skin is quieter than you

74:44

think it notices the pattern before you

74:46

do so it's actually telling me what it

74:49

is and now it's given me the script

74:52

right they need At the same time 5

74:55

minutes done daily does more than 30

74:57

months. These ingredients work through

74:59

repetition not the intensity your skin

75:01

catches. So now you have you see the

75:05

natural breathing area and the breathing

75:07

space. So now you understand how you can

75:10

generate read scripts. This is again

75:13

I've done it very very I would say

75:16

watered down version of it. The one that

75:19

we actually use at Upskll Nexus is my

75:22

real script generator. This is very

75:26

extensive. So I'm going to show you a

75:29

little bit about how extensive you can

75:32

develop this to be. So now this is going

75:35

to ask me the domains that I want the

75:39

information that I require that it

75:42

requires to generate a skill. And if you

75:45

see it has six kind of questions.

75:48

that I have input. So you can make it as

75:52

extensive as you would like. Now this

75:55

was the real script generator. The next

75:56

one that we're going to cover is the

75:58

caption generator. So let me quickly

76:01

walk through that one as well.

76:07

All right folks. So the next one is

76:10

going to be a caption generator. So now

76:13

what we're going to do is that we had

76:16

created our real script right um the

76:19

real script that we created was very

76:22

dumbed down. So I will create or utilize

76:25

a real script that I already have

76:29

generated. So let me just quickly go to

76:31

Orena's real script generator and say

76:34

write me a 100word

76:38

real script on turmeric and its uses for

76:45

skin care.

76:48

Right? So now I'm going to get a real

76:50

script out of it and this is the

76:52

workflow that I want to use. So let's

76:55

say I have my real script at hand.

76:57

Right? Then I want my skill that I'm

77:01

going to develop for the caption

77:02

generator to take this real script and

77:04

write captions based on the real script

77:07

that is provided. So the first step is

77:10

writing the real script uh or generating

77:12

the real script in itself. So let's

77:14

quickly do that. And um

77:21

now they're going to quickly do it. So

77:24

it's going to ask me some questions

77:26

about my hook style. and etc. So

77:35

let's say I want to say something like 2

77:36

days before a wedding and your skin

77:38

looks exhausted and this is going to

77:40

give me a 100 word one. So this is again

77:45

72 words 34 words. So if let's say uh

77:50

let's just for the sake of time let's

77:53

quickly just utilize this one. I am not

77:58

going to show you how to basically fix

78:02

this. I just showed you a workflow and I

78:04

want you to implement this for your

78:08

particular

78:09

let's say use case. So now I'm going to

78:12

write all right so I'm going to dictate

78:16

based on a real script that is provided

78:18

to you. I want you to create

78:23

captions for LinkedIn X,

78:27

Instagram, and YouTube. Paste

78:32

on Orenda Skinare.

78:37

All right. So, sometimes it does not get

78:40

the accent, which is fine. So, I'm just

78:42

going to quickly correct some of this

78:47

script that is provided to you. So I

78:48

want you to create captions for now I am

78:52

going to show you how this will do it

78:57

uh when it does not have see right now I

79:01

already have a caption generator skill

79:03

right so it's actually just invoking

79:05

that skill and utilizing it so I have

79:08

already specified so I can get straight

79:11

into the generation now I had this

79:14

already inbuilt but I am going to now

79:17

show you the process of building this

79:19

skill in itself. Right? So let's get on

79:23

with it.

79:28

Okay, let's write help me create a

79:33

caption

79:34

generator skill

79:38

that

79:40

takes in

79:42

a script

79:44

and

79:46

makes

79:49

captions

79:52

out of it. Here

79:55

is your test case. So, I'm going to give

79:59

it the test case

80:01

and

80:03

platforms

80:05

X, LinkedIn,

80:09

and let's say YouTube

80:13

and Instagram.

80:16

Okay. So, now I'm going to quickly wait.

80:20

It's going to read and it's going to

80:21

generate one, right? It's going to maybe

80:24

ask me a couple of questions the same

80:26

way that it was before. And because I

80:29

already have an existing caption

80:31

generator.

80:33

So now it's going to evaluate it, right?

80:38

And going to write one. So the core

80:40

message extractions, it's done. And now

80:42

it's written one based on the one that I

80:44

already have. So what does this tell

80:46

you? If you already have a caption

80:48

generator skill like I do, it is just

80:51

going to refer to that. Again, I

80:53

deliberately decided to show you whether

80:56

or not you require a skill every time.

80:59

But this is what it says. Now, a few

81:03

honest notes on the skill itself before

81:05

you decide whether to update it or not.

81:07

The existing skill is solid, right? But

81:10

the brand voice is generic because it

81:12

actually works for upscale nexus and not

81:15

for Oenda. Right? The ask platform step

81:17

conflicts with your workflow. And there

81:20

is no guidance or CTA. Right? So all of

81:24

this improvements. No, I do not want to

81:27

improve this. So I will just say I am

81:30

happy with the current output or I can

81:32

just leave it at that. Right? Because I

81:34

don't want to actually update it. But

81:35

you see that you can create a caption

81:37

generator skill like this. But you know

81:40

what's an additional thing that I did

81:42

not initially think of that I would like

81:44

to cover in this video is actually

81:47

turning a real script into a carousole.

81:52

Now there might be times that you have a

81:54

lot of real scripts lying in right and

81:56

you want to turn it into a carousole. So

81:58

how do you do it? So I'm going to

82:00

quickly write help me create a skill

82:04

that takes a real script

82:09

in and converts it

82:15

into a carousel.

82:18

So now this is what I'm going to do.

82:21

Let's see how this is going to be. So

82:24

now it's going to read the skill creator

82:26

skill and mayhaps

82:29

take

82:32

everything that is given to me and it's

82:34

going to also read my carousel content

82:36

generator skill. So I have two skills uh

82:39

one for the real script and the other

82:42

for carousel. So it's comparing both of

82:45

them, right? And it's also seeing that

82:47

what this particular skill needs to do

82:49

that's going to be distinct from my

82:51

existing carousle content creator skill.

82:54

Now you see the way that I have decided

82:57

to structure my entire social media

82:59

content aspect is to do with the

83:01

modality that I'm trying to you know um

83:05

use to put my ideas in front. So you

83:09

have to think of it in that way and

83:13

have that workflow in your head and how

83:16

you're going to make it. So let's

83:17

quickly just pause and take a moment

83:20

till it's generating and then we'll get

83:22

right back.

83:26

All right folks. So now we have the

83:30

entire real to carol skill MD broken

83:33

down. So what it does is it takes the

83:36

core insight uh from let's say the real

83:39

script that is linear into a kazol which

83:42

is a self-contained unit. The skill just

83:44

doesn't copy paste. It actually does

83:45

structural decomposition. It breaks

83:49

the script into parts and then maps

83:52

those slight boundaries. Okay. So now

83:55

I'm going to save the script cuz this is

83:56

a skill that I can also utilize. So how

84:00

about this? Let's actually invoke this

84:02

and try this out with the brief that we

84:06

have. So this is the thing and let's say

84:10

real to carousol enter. So now I'm going

84:13

to quickly see what's going to happen

84:16

and most probably the real to carazol

84:18

skill is going to actually build the

84:21

carousol. Now I did not specify the

84:24

output in which uh the carousol needs to

84:27

come out but usually the carousol comes

84:31

out in the format

84:34

of either content. So what it's done is

84:38

it's actually taken the content and it's

84:40

given me what kind of hues my designer

84:43

requires to take into consideration and

84:46

what kind of things I need to give to

84:48

Gemini to create using nano banana model

84:52

images. So it's given me prompts for

84:54

both and that's brilliant. One I like

84:58

for LinkedIn because the LinkedIn

84:59

version is actually in the form of a

85:00

PDF. On the other hand, the Instagram

85:03

version actually is for uh let's say in

85:08

the form of an HTML or in the form of

85:10

PNG images. Now it gives each and

85:13

everything right. So if you see this is

85:17

what it is and once so one thing I like

85:21

to do is find the factual claims cuz

85:24

sometimes when you're claiming something

85:25

AI can hinate, right? If you see here

85:28

it's written AI can make mistakes.

85:31

Please double check responses. So what

85:33

does that even mean? It basically means

85:36

that Claude can provide misleading or

85:40

incorrect responses sometimes. So you

85:42

should always double check it. And the

85:44

way I do that is that I tell Claude in

85:47

itself to fact check whatever it says by

85:51

pasting the final version back and it

85:54

fact checks each and everything that

85:56

it's saying. All right. So this is how

85:59

you can create a carousole a real to

86:03

carousole skill and a carousole with

86:05

existing content. Now this thing is not

86:09

just for converting reals to carousols.

86:13

You can actually do carousol to real.

86:16

You can do full-fledged blogs to podcast

86:19

scripts to whatever content is in your

86:23

head and whatever content you would like

86:24

to translate. Okay, before I get on with

86:27

it, I want to show you a little tool

86:29

which is called Whisper Flow. So,

86:31

instead of uh typing the entire prompt,

86:33

I'm going to show you how you can

86:36

dictate the entire prompt. Right? So, we

86:38

have seen this skill the PowerPoint one

86:40

in episode one with the campaign kickoff

86:42

tag. This time it's a debrief. So, I'm

86:45

going to take the same brand that is

86:46

Oenda and I'm going to create a debrief

86:49

tech with uh different structure from

86:52

the kickoff, right? It has leads with

86:55

results. uh moves through what worked

86:57

and what didn't and it closes with

86:59

recommendations. So, let me show you

87:01

what the re brief actually looks like.

87:04

So, I'm going to quickly come here and

87:07

I'm going to I can either click this or

87:09

I can hold function. I'm going to click

87:11

this quickly and then I'm going to

87:12

dictate it. Right? So, build a campaign

87:16

brief uh a campaign debrief on Orenda

87:21

Skincare. It's the Diwali 2025 campaign

87:24

debrief. The audience is the internal

87:28

leadership team. The campaign period was

87:31

between 15th of October to 15th of

87:35

November 2025. The results are that the

87:38

revenue was up by 41% in comparison to

87:42

last year's October 2024. That means it

87:45

beat the target of 35%. New customer

87:50

acquisition was around let's say 2,340

87:55

which beat the target of 2,000. The ro

88:00

was about 4.6 times against the target

88:03

of 4.2

88:05

and email open rate was about 31%

88:09

against the target of 28. What worked in

88:12

this campaign was Instagram reels

88:16

outperformed

88:17

static posts by three times on reach.

88:20

Influencer partnerships drove about 33%

88:25

of the new customer acquisitions. Um the

88:28

gift set messaging resonated strongly

88:32

with the gifting buyer segment. What did

88:35

not work? Well, Google display ads

88:38

underdelivered. So the spend of about

88:41

18% of the total budget uh for 9% of the

88:46

revenue was utilized for Google display

88:48

ads. The email list fatigue was visible

88:52

by week three. So open rates had dropped

88:56

by six points on the third week and the

88:59

recommendations from my side is that

89:02

increase the real production budget for

89:05

the next campaign, reduce the display

89:06

ads uh allocation to about 10% and

89:11

redirect it to the influencers.

89:14

Also set a cap on the email sequence at

89:18

about two cents per campaign and format

89:22

this entire PPT in a 10 slide debrief

89:25

deck. So show me the results first, then

89:28

the channel breakdown, then what exactly

89:31

worked, what did not, and then tell it

89:34

the recommendations and a final summary

89:37

slide.

89:39

All right. So, Whisper Flow is going to

89:42

do its magic and whatever I just

89:44

dictated, it's going to prop up on my

89:47

screen. There we go. Everything right.

89:52

So, I'm going to change this to either

89:54

highq or sonnet. Let's do highq only. I

89:57

do not want to, you know, uh unnecessary

90:02

do it. So, see claude wants to you use

90:07

generate from gamma. Now this is

90:11

something uh generate from gamma that

90:13

it's using because I have connected my

90:16

gamma using a connector. So let's say I

90:19

deny this for now. Okay. So this is not

90:23

I don't want to use gamma.

90:27

Let me see how quickly it gets done. All

90:31

right.

90:33

Okay. It's going to pick up gamma

90:35

itself.

90:37

GMA might not be available in your Let

90:39

me create this as a polished PowerPoint

90:41

deck. Okay, it's working now folks. Uh

90:45

there was a little bit of a problem with

90:46

Garma, but GMA is another PowerPoint

90:50

tool that you can use to create your

90:54

campaign decks or to create your

90:56

presentations.

90:58

So let's see. See the campaign debrief

91:01

deck has one particular job. It just

91:02

needs to tell the leadership what

91:04

happened, why it happened, and what to

91:05

do differently in a format they can read

91:08

in under 10 minutes. And this skill is

91:11

basically turning it into a deck.

91:20

Okay. The skill's job is basically to

91:22

turn this into a particular format where

91:28

they're making an entire polished 10

91:31

slide campaign brief using PPTx

91:35

genjs which is PPTX generation

91:37

JavaScript. Now it's doing it. It's

91:40

creating everything. It's removing the

91:42

buggy file. It's creating the entire

91:45

deck. While it's doing it, um I also

91:48

want to quickly address that all of this

91:53

has been covered in a previous video. So

91:55

if you have not seen it before, well,

91:58

I'll place it in the icard above. So

92:01

quickly go see the first two videos so

92:03

that you can understand this better,

92:06

especially the PBTX skill. Now the

92:08

PowerPoint presentation is created. I'm

92:10

going to quickly refresh this so that I

92:14

can see it. Uh sometimes it's a bit

92:17

slow. Why is this blank?

92:21

Let me see.

92:24

There you go. It takes a little bit of

92:27

time. So I used highq. So the formatting

92:30

is a little bit off. But if I turn this

92:32

to sonnet, it'll be much better. Uh, in

92:36

my personal experience, set does do way

92:41

way way better when it comes to creating

92:46

your PBTs. I just used HighQ because I

92:49

wanted to show you how to do this. The

92:51

Sonnet version has been covered in the

92:53

previous video. So, I would say that you

92:55

should go and check them out. Well, see

92:58

this is a tense like debrief deck and we

93:00

just saw this. The structure is very

93:02

logical is the exact structure that I

93:05

had mentioned here. So it goes from

93:07

title to key results to revenue

93:08

performance so on and so forth. Right?

93:11

But let's see now how to use the Excel

93:15

skill because this we have seen before.

93:17

Right? What we need to now venture out

93:19

into is the new skill. So let's get on

93:23

with that one.

93:26

Right folks let's see how the Excel one

93:29

works. So before I do that number one

93:32

I'm going to change this to set because

93:34

this uh is going to be analysisdriven

93:38

work. So before we get on to this right

93:41

if I want data to analyze I need to

93:44

first start by getting the data in the

93:48

form of a sample marketing data for me.

93:52

Now uh for the sake of brevity I had

93:54

already created this but I'm also going

93:56

to give you a prompt of how you can do

93:59

this yourself. So I'm going to quickly

94:01

record. So Claude allows you to record

94:03

uh your voice here as well to dictate.

94:06

However, I prefer whisper flow because

94:08

it also is an AIdriven

94:12

uh I would say dictating tool that edits

94:15

out the ums and okay. So I'm going to

94:19

press function this time and I'm going

94:20

to show you how it's done.

94:24

All righty. So,

94:27

generate sample marketing data for me.

94:30

Uh, it should have two tables. First,

94:33

campaign performance with columns for

94:36

month channel impressions, clicks,

94:39

spend, and conversions. Make it 6 months

94:43

across four different channels. Paid

94:45

search, paid social, display, and email.

94:51

Second table should be a lead funnel

94:53

data with columns for month, source,

94:57

leads generated, MQLs,

95:00

SQLs, and closed deals. Make it

95:03

realistic but messy. Some inconsistent

95:06

formatting, a few missing values, mixed

95:10

capitalization. I want to clean it up.

95:15

All right. So, here you can see this,

95:19

right? So it's uh written should have

95:21

two tables. You know how it's formatted

95:23

this. So Claude won't be able to do

95:25

that. So which is exactly why I like it.

95:28

So now I'm going to quickly take a pause

95:30

here and I'm you see it's reading the

95:33

Excel skill to create the marketing data

95:36

Excel file as well. This can happen and

95:39

I had covered this in my previous videos

95:40

where I showed you how to do this

95:43

exactly by extracting data from reports

95:46

which are PDF documents into Excel

95:49

files. Right? And let me uh quickly take

95:54

a cut here and then you can see

95:57

how it will come out. So let's wait for

96:01

this to get processed.

96:05

All right, folks. It ran the command and

96:07

it's given this um data called marketing

96:10

data. It has the campaign performance as

96:13

well as the leadfunnel raw data. The

96:15

formatting is a bit off. So now what I

96:17

want to do is I'm going to quickly

96:19

download this and save this in my

96:22

downloads as marketing data. And now I'm

96:26

going to show you how to clean this up.

96:29

Right? So let's now clean this data, fix

96:34

the inconsistencies and flag missing

96:36

values. Okay, for that I'm going to

96:39

quickly connect this right here. I'm

96:42

going to add the file and I'm going to

96:44

show you how to do this exactly.

96:48

Clean this data, fix the

96:50

inconsistencies, fill or flag the

96:52

missing values, standardize the

96:54

formatting, and build me a professional

96:58

Excel file with two sheets, one for

97:01

campaign performance and one for

97:03

leadfunnel. Use live Excel formulas and

97:06

not hard-coded values. It's very

97:09

important that you use live Excel

97:11

formula so that whenever you change the

97:16

uh specific aspect specific numbers in

97:19

your sheet it does not generate

97:22

something that's different from earlier

97:26

ones right um so I'm going to quickly

97:29

wait for this so they if you see it's uh

97:32

you can't see it here but the Excel file

97:35

I've attached it you had a glimpse of it

97:37

for now it's reading the XLS skill this

97:41

is the skill right I last time explained

97:44

what all of this is right and if you see

97:47

this is a proprietary skill which

97:49

basically means that enthropic owns this

97:52

skill so I'm going to quickly wait for

97:56

this to run and then I will show you the

97:59

difference between the truth.

98:07

All right, folks. It is here. If you see

98:10

the campaign performance completely

98:12

formatted, we saw in the last one how it

98:15

was different and the lead funnel also

98:19

completely formatted with the missing

98:21

numbers reported in terms of the

98:24

impressions, in terms of the CTR. If

98:27

there are any missing numbers, they're

98:29

completely here. And the entire channel

98:32

summary for each of them is here with

98:35

the CTR, CBR, CPC, right? All of the

98:39

important ones that you require. This is

98:42

the campaign performance and the lead

98:44

funnel. It is it has calculated the

98:48

leads that were marked in yellow. These

98:52

were the ones that were not there. If

98:54

you see see the yellow call yellow cells

98:58

are estimated missing values. So it has

99:00

done that and if you see the formulas

99:03

are all dynamic. So it's right here. And

99:07

now that I have done this I have my

99:09

clean data. Now in this right I have

99:14

shown you the difference. Let's do the

99:17

analysis aspect in this as well. So

99:23

let me quickly say uh what the analysis

99:27

aspect is and let's get on with it. So

99:31

the analysis aspect right okay I'm going

99:34

to deletate this

99:36

add a third sheet called analysis

99:40

calculate the following metrics from the

99:42

data CTR by channel cost per click cost

99:46

per lead MQL to SQL conversion rate SQL

99:51

to close rate and overall funnel

99:53

conversion rate color code the

99:55

performance green for quartile amber for

99:59

mid red for bottom and also include a

100:02

summary row.

100:06

Right, the lead funnel already has all

100:08

of this, but let's see. Um, it's going

100:12

to create a third analysis sheet. And

100:16

well, I'm going to take a step back.

100:19

It's going to inspect the workbook

100:21

structure first. And now it's going to

100:24

build the analysis sheet. It's running

100:26

the command. And it should be here in

100:29

the next four to five

100:34

seconds actually. Um four to five

100:36

minutes. Sorry, not second.

100:39

So I'm going to quickly wait for this to

100:40

run and I will see you again.

100:44

Okay folks, I actually want to bring

100:47

your attention to something here. It

100:49

says that it's computed 330 formulas

100:53

with zero errors. Now, if you see here,

100:55

it actually also recalculates and checks

100:58

for errors. There's an entire script

101:00

that sees all of that and it also fixes

101:04

the path, runs it over, it creates the

101:07

entire thing using scripts. Now, what's

101:12

a script? Well, a script is nothing but

101:15

a set of code that the machine runs in

101:18

order to execute let's say XYZ step for

101:22

you within the skill. Now if you see the

101:27

analysis section has all four of them

101:29

they're live linked. Live linked

101:31

basically means if you change numbers in

101:32

a part it automatically gets changed in

101:35

part B. And all of this is here. See one

101:38

structural note per flagging display

101:41

doesn't map into leadfunnel data at all.

101:44

So it's CPL and funnel metrics are not

101:47

applicable. If you ever get display

101:49

attribution those cells will need

101:52

formula. So see it's giving you a

101:54

warning also if something is missing

101:57

right and that's the part. So if I go to

101:59

analysis here, oops. If I go to analysis

102:02

here, you can completely see the

102:04

dashboard right of

102:08

the entire thing including the

102:11

scorecard,

102:12

right? It's all here. It's all

102:15

available. You can completely see it.

102:18

And you've seen the difference from the

102:21

original one where we only had campaign

102:23

performance and leadfunnel to now also

102:26

having analysis. Well, now let me

102:29

quickly show you a little thing that I

102:32

wanted to actually add here and this is

102:35

my little surprise in the end. So, um,

102:39

now I know what happened over the last 6

102:42

months, right? But the question is, what

102:44

should I do differently and how should I

102:46

do differently if I wanted to impress,

102:49

you know, my higherups? Here's how I

102:52

would do this. All right. So, I'm going

102:55

to quickly show you a little trick in my

102:57

book that I use to impress my higherups

103:01

and the officials, which is to build an

103:04

interactive HTML dashboard. All right.

103:07

So,

103:09

uh using this campaign and funnel data,

103:12

build me an interactive HTML dashboard.

103:15

I want sliders or input fields for

103:18

monthly ad spend per channel, target

103:22

CTR, and target funnel conversion rate.

103:26

As I adjust them, I want to see

103:29

projected impressions,

103:31

leads, MQLs, SQLs, closed deals,

103:37

projected revenue. Assume the average

103:40

deal size is of about 50,000 and blended

103:45

ros

103:46

all updating in real time and make it

103:49

clean and professional.

103:52

Okay, now I've given this and let's see

103:55

it's going to build me an HTML

103:58

interactive dashboard. So I'm going to

104:01

quickly take a pause here. Give it a

104:03

moment to generate it. Um, it does take

104:06

a little bit longer than usual and I

104:09

will come back to you.

104:15

All right, folks. It's done it. So,

104:19

based on what I just asked it, it's done

104:23

this entire dashboard, right? Um, you

104:27

might be confused that oh, this is

104:28

inside the claw interface. That's not

104:31

true. You can actually download the file

104:33

from here and it gives you a marketing

104:36

performance dashboard HTML but it does

104:40

show you the output sometimes in here.

104:43

So here's how things change here, right?

104:47

So let's say you see the sliders. Let's

104:50

say I increase my paid ad spend to about

104:54

19,000.

104:56

Reduce my paid social. Make display ads

105:00

just 1,000 emails to be 1,000. So you

105:04

see with how as you interact the numbers

105:08

below are changing right and let's say

105:11

if my target CTR then uh once I change

105:15

this right my lead to MQL rate how

105:18

that's going to be there my MQL to SQL

105:22

rate how that's going to fluctuate and

105:26

the SQL to close rate right and what my

105:28

total spend was it also gives me the

105:31

lead funnels and so on and so forth If I

105:34

do change these numbers and the

105:36

projected revenue, it says where it's

105:38

coming from. So right now I had uh put

105:43

the others as zero. But the moment like

105:46

I add this here and I change it to a

105:51

little bit higher based on my older

105:53

projections. You can see how this

105:56

changes right where the projected

105:58

revenue is going to come from based on

106:00

our past data and is doing the entire

106:04

thing. Now

106:08

my question is that what if I tell it

106:11

that based on what

106:15

I you know just modeled if I wanted some

106:20

recommendations

106:21

right um let me ask it for

106:24

recommendations give me three budget

106:29

reallocation

106:32

recommendations ations

106:36

uh let's say recommendations

106:38

in plain English right I don't want all

106:41

these buzzy fancy words so let's see if

106:45

it can avoid the buzzy fancy words and

106:48

then see this right looking at the

106:52

current default state of the model it's

106:55

xyz xyz here are the three

106:56

recommendations move budget out of

106:58

display into paid search all right email

107:02

is massive massively underinvested

107:03

relative to its output. So, invest more

107:06

in emails and don't touch paid social

107:08

until you fix your SQL to close date.

107:14

So, you see from what I just modeled,

107:16

it's given me three insights that I

107:19

should focus on. And this is uh an added

107:24

little spice that I wanted to show you.

107:27

And now, let's get on with just wrapping

107:31

this up. All right, folks. So, let me

107:34

show you the report first. This is what

107:37

a campaign report would look like. Let

107:41

me go to continuous scroll so it's

107:43

easier for you to see, right? So, it's

107:45

prepared by Meridian Media. And this is

107:49

what you receive right in your inbox.

107:51

Probably something like this. Now, this

107:53

is a dummy report obviously, but you can

107:56

see how these numbers are added in here.

108:02

uh what exactly is the performance

108:04

analysis? What are the key observations,

108:06

the strategic recommendations

108:08

and the conclusion, right? So you have

108:11

this and the table on page three here is

108:16

what you actually require. So this part

108:20

the entire table here, this is the one

108:23

that you actually require. So let me

108:26

show you how to do this and how to do

108:29

this fast.

108:34

Okay, this is my clawed interface. Uh, I

108:37

think you are already aware of what the

108:40

interface actually looks like. Now, this

108:44

is I've attached the PDF here from add

108:46

files or photos. And now I'm going to

108:50

quickly use the PDF reading skill. This

108:53

is an inbuilt skill, right? uh is that

108:56

it's going to read the entire skill and

109:01

will uh

109:04

read it, identify what's the context,

109:06

what's the data, we'll pull out the

109:07

structured information. The difference

109:10

is that it's not going to get distracted

109:12

by the agency's framing and it'll find

109:14

the numbers. Okay. So, let me uh write a

109:20

quick one right here. So what I am going

109:24

to do is that I am going to ask it

109:30

to give me the data from here. Okay. So

109:37

PDF reading skill please

109:42

extract all the important

109:48

information

109:51

from here. Now I'm using highQ for this.

109:56

Uh so you know it's checking the

109:58

uploaded files. It's going to extract

110:01

all the critical data from the PDF. So

110:05

it's already see all of this is

110:08

extracted for you. You don't have to do

110:10

much work. Now this might seem fine

110:13

right um you are extracting the data.

110:17

This is all right right it checked the

110:20

uploaded files and then it extracted all

110:24

the critical data. Now the real question

110:27

is that you want this right extracted

110:32

data and use this output to feed it

110:37

directly to the Excel skill. Now it's

110:40

the first time uh in this series that

110:42

you're probably going to see one skills

110:45

output become another skills direct

110:48

output. So keep watching this.

110:52

All right folks. So what we're going to

110:54

do is that we're going to utilize this

110:57

data that I've gotten here the

111:00

extraction into

111:03

another skill. So I'm going to use it as

111:05

an input. But because I have to change

111:07

my model to sonnet, um sonnet is good

111:10

for analysis. So I'm going to quickly

111:11

change it to sonnet. And I'm going to

111:13

dictate my prompt. This time I'm going

111:16

to use the dictate function here.

111:23

Build a twotab spreadsheet. Tab one

111:27

should have the channel performance data

111:30

extracted from this report formatted as

111:33

a clean table with all columns uh

111:36

written. Tab two

111:40

should be a channel by channel analysis

111:42

for each channel one line on whether

111:47

it overperformed or underperformed.

111:50

what the key number is and a verdict of

111:54

scale, maintain or cut.

112:01

All right.

112:04

So before I get that started, uh I

112:07

forgot to add the thing I copied. So

112:12

what I had copied here and now I'm going

112:15

to enter. So it's going to use this.

112:18

It's opened the inbuilt skill as you can

112:21

see here right now. It's going to

112:26

make the spreadsheet. So, see it's

112:30

saying it's creating the spreadsheet.

112:32

See, uh the second tab is actually what

112:35

I want you to watch this for. So not the

112:39

reorganized data because that's very

112:41

easy to do these days but actually

112:45

positioning yourself on each channel and

112:49

understanding with the numbers that

112:52

support it and how to

112:56

see that where you stand on each channel

112:58

where you're spending money. So whether

112:59

you want to scale, whether you maintain

113:00

or whether you want to cut the money. So

113:04

that's the output that probably a

113:05

performance marketer um needs at the

113:08

start of a client conversation. So it's

113:11

not a reformatted PDF but a verdict,

113:13

right? All right. So we're seeing here

113:16

it's recalculating the formulas and it's

113:19

checking for errors. Then we are looking

113:23

at this and there we go. So it is dark

113:26

themed. It is made in that manner. It

113:29

has tab one

113:32

as you can see here and it has tab two.

113:36

So we have two tabs channel performance

113:39

and channel analysis.

113:42

Now if you see here right uh there are

113:47

some things that it has given us in

113:50

terms of tab uh tab one and tab two

113:54

which are part of the analysis. Now

113:57

Instagram if you see here the ro is

114:01

about 5.8 against a campaign target of

114:04

4.2.

114:07

Now when you have this right that's

114:11

available to you. So you see the

114:12

campaign target here you see the ROAS

114:15

here. So the verdict eventually becomes

114:18

after analyzing that you should scale

114:21

Instagram because it has overperformed.

114:24

At the same time, Google display ads

114:26

underperformed, which means you should

114:28

cut back on your money in that

114:30

particular direction. Similarly, if you

114:33

see influencers, it's showing that they

114:36

have overperformed. So, they have hit

114:38

more than the target. So, you should

114:41

scale it. And what you should focus on

114:44

the mid-tier targets. So, the mid tier

114:46

mid-tier creators is somebody that you

114:49

should be focusing on. Similarly the

114:52

channel performance if you see here it's

114:54

given which all has been more uh in

114:58

terms of the returns that we have gotten

115:01

and what has been below the campaign

115:03

average. All right. So this is a

115:06

document that actually I want you to

115:08

open whenever you are going in for the

115:12

client call that is the second tab and

115:15

not the first one because this gives you

115:18

a structured positioning and explaining

115:20

what exactly it is and how uh you have

115:25

spent the money to get

115:29

the performance out of it.

115:37

Tab one, the raw data layer. Right? So

115:40

all the five channels, every column from

115:44

the original report into a clean table.

115:46

This is where you run your own

115:48

calculations ahead. If you wanted to

115:50

calculate more numbers, you can do it

115:53

here. You can even build a chart. So let

115:56

me quickly ask

115:58

uh Claude to make some charts

116:04

that are suitable

116:07

whoops

116:09

suitable

116:11

for

116:13

tab one. Okay. So now it's going to

116:16

quickly add some charts that do in the

116:18

right format. So whatever suitable

116:20

charts are there is going to do it. So

116:23

you see that it's reading the skill for

116:25

chart creation guidance as well. So it's

116:29

doing this right here. And uh after this

116:34

it's going to just go from uploading the

116:39

PDF into making this two tab

116:41

spreadsheet. Now I can give you a quick

116:43

hack as well. Once it's made the charts,

116:46

what I'm going to do is I'm going to

116:48

actually invoke the skill creator and

116:51

ask it to make a skill for the same

116:53

thing that it has done. That is take the

116:56

output of a PDF and convert it into a

117:00

channel performance report in the form

117:02

of an Excel. Right? So, let me quickly

117:05

do this. So,

117:09

build a skill that does this every time.

117:16

Reads a

117:19

what do you say? let's say reads a

117:22

report from an agency

117:26

and then makes

117:31

an Excel

117:34

with

117:36

numbers

117:38

to analyze

117:41

performance.

117:43

Right now you see it's going to create a

117:45

specific skill just for this. So if you

117:48

see here uh this is the channel

117:51

performance. These are some numbers that

117:53

it has picked up and maybe if I quickly

117:58

open this here there will be some charts

118:01

that are going to be

118:04

you know uh

118:07

alongside all right there are I see some

118:10

more things. Oh here we go the charts.

118:13

So there are four charts 1 2 3 and four

118:17

that are here and you can see. So one is

118:21

the campaign average the ROAS by channel

118:24

the budget share the cost acquisition

118:28

and the budget allocation. So you see

118:31

email had the least while the others had

118:35

the most. Now you have the data and the

118:38

charts that are available to you and the

118:39

skill creator skill right now is

118:42

understanding the workflow. So it's

118:45

going to skip the sub agents baselines

118:48

and the review and it'll basically make

118:52

a skill from here. So now you see you

118:56

can actually see here it's reading the

119:00

marketing agency campaign report. It has

119:04

made a skill. So it's reading the skill

119:07

first. Then it's extracting the data.

119:10

Then it's making a workbook out of it.

119:13

And there you go. Now you can save the

119:15

skill. It'll go right in there into your

119:19

skills section. That's you will be able

119:21

to find this in customize here where you

119:24

have all the skills residing. Uh

119:28

similarly if you are not able to find

119:30

this well you can check this out from

119:34

here and you can add it here as well.

119:41

Okay. Another way to save the skill is

119:44

to click on download and it'll save as a

119:47

dot skill file and then you can send it

119:49

to your team as well so that he can also

119:52

utilize this skill. Now an honest note

119:55

here before we move on. The PDF reading

119:57

works best on native PDFs. So documents

120:02

exported digitally from let's say report

120:04

builder design tool. Most agency reports

120:06

from wellestablished agencies are native

120:09

PDFs. So in practice this covers the

120:12

majority of what you receive. If you

120:14

want to try this on a scanned PDF or a

120:16

report that's mostly images of charts.

120:19

So the extraction will be rest reliable.

120:22

So this is something that you should

120:25

know and it is worth knowing before we

120:28

run it on something that matter.

120:32

Hi folks, we are now going to begin and

120:35

we're going to create the document for

120:37

the B2B management consulting firms. So

120:41

Kova Consulting is a B2B management

120:43

consulting firm with about 40

120:45

consultants. It's extremely fast

120:47

growing. The firm has a voice, right? It

120:50

just lives in people's heads right now

120:53

and comes out differently depending on

120:55

who's writing a proposal. The document

120:58

I'm building is called a tone of voice

121:00

guide. Something every consultant at

121:04

Kova can use when they're writing client

121:08

communications, proposals, LinkedIn

121:11

posts, anything that goes out under the

121:15

firm's name. So you see the claude

121:18

interface. you're very very uh

121:20

acquainted with this and I hope you are.

121:23

Now let me give it a quick prompt. So I

121:26

am using a tool called whisper flow

121:27

here. It lets you think of ideas, right?

121:32

It has two options of whether to polish

121:37

what you're saying or prompt engineer.

121:39

I'm choosing prompt engineer right now

121:41

and I'm going to quickly uh utilize the

121:44

skill so that

121:48

I can dictate what exactly I want.

121:52

Tone of voice document for Kova

121:54

Consulting. KA Consulting is a B2B

121:57

management consulting firm with about 40

122:00

consultants. The document audience is

122:04

all consultants plus the marketing and

122:07

business development team.

122:09

uh four tone of voice principles need to

122:12

be followed. Number one, direct and

122:16

precise. No jargon, no filler. Say

122:20

exactly what you mean in as few words as

122:24

it takes. Include a do and do not

122:28

example for that. Number two, uh make it

122:32

confidence without arrogance.

122:35

authoritative but never dismissive of

122:38

the client's existing thinking. Include

122:42

a do not and do example in that as well.

122:46

Number three, human despite the subject

122:50

matter. Consulting

122:53

can feel cold and abstract. KOA's

122:57

writing should feel like advice from a

123:01

smart colleague, not a report.

123:05

um from a firm include a do not and do

123:09

example in this as well. Number four,

123:12

practical over theoretical.

123:15

So every recommendation comes with a

123:17

concrete next step. No principle without

123:20

application. Include a do do not example

123:24

in this as well. Document structure um

123:29

should be overview section explaining

123:31

why tone of voice matters for the firm.

123:34

Then the four principles each with a do

123:37

do not example formatted as a two column

123:40

table. Then a writing checklist of 10

123:44

practical rules. Then a onepage quick

123:47

reference summary at the end. The format

123:50

should be properly structured word

123:52

document with heading hierarchy. Page

123:55

breaks between major sections. do do not

123:58

examples as two column tables with do

124:02

don't as column headers.

124:07

So it is going to process and if you see

124:10

it has everything here right it's almost

124:14

like a prompt.

124:16

So I have this here. So

124:22

now I'm going to add something more

124:24

using this document.

124:27

create a custom skill.

124:32

So let's do this step later once it's

124:34

given me the document. Right now my

124:36

focus is building the document. See a

124:40

tone of voice document has one job,

124:42

right? It needs to be clear enough that

124:45

someone who has never thought about

124:48

brand voice can read it and immediately

124:50

write better. It needs to be more than

124:53

be authentic or be like the brand

124:56

because that's not guidance or do do not

124:59

example showing the difference between a

125:01

w opening and a direct one that would be

125:04

better guidance right so watch for that

125:06

output uh we will look at that for the

125:10

output once it's done so it is creating

125:13

the document at the moment so just let's

125:16

quickly take a break and come back once

125:19

it's generated

125:30

All right. If you see here, I'm a bit

125:32

confused cuz the first page seems to be

125:36

upside down somehow. Okay. This can

125:39

happen with claw. So, I'm going to

125:41

quickly write this. The first page is

125:45

upside down.

125:48

Fix it. This much is enough for claw to

125:50

fix it. But you see um we have an

125:54

overview. I don't know if you can read

125:57

it but then we have the principle the

126:00

first principle and then what's going to

126:03

happen the dos and the don'ts right? So

126:10

similarly here principle number two,

126:13

principle number three and principle

126:14

number four right and then the writing

126:17

checklist a quick reference and what to

126:19

remove what not to remove.

126:23

Okay so

126:27

there we go it's here. So this gives

126:30

completely what it is and this is the

126:32

brief document that you can save and

126:35

then you can share it with your team. At

126:39

the same time you can also open this in

126:42

drive. So it automatically saves to

126:45

drive and you can utilize it there as

126:49

well. Well this was all for the first

126:55

thing. Now I will show you a quick trick

126:58

that you can use and you can share this

127:00

as not just a document but I'm going to

127:03

actually quickly

127:05

save this

127:07

and

127:11

all right hi friends. So what I'm going

127:14

to do now is that I am going to utilize

127:18

the dox skill and create the other three

127:22

documents. the brand voice system, the

127:24

brand kit foundation and the brand audit

127:27

proposal the all three of them so that

127:32

we will understand how we can utilize a

127:37

docx skill to create it and at the end

127:39

using these documents that we have

127:41

created we will create a custom skill

127:44

that is going to be helpful for my

127:48

entire team which has access to claude

127:50

to automatically format at and adhere to

127:54

these guidelines. So, not only are we

127:57

going to give them the guidelines, we're

127:59

also going to teach them how to utilize

128:02

it. Now, um the second brief is a little

128:05

bit more demanding. Uh so, the brief

128:08

that I'm going to give Claude, it's

128:10

going to ask Claude to make judgment

128:12

calls about what KA should and shouldn't

128:14

sound like based on the company's

128:16

positioning and the uh and the

128:19

competitive context. So what I have done

128:22

is I have written the entire prompt for

128:26

you already. Uh I am just going to

128:29

quickly copy paste it here. And if you

128:32

see here the brand voice system for KA

128:34

consulting, it's given the company

128:37

positioning. So what exactly does it do?

128:40

What it is not? What's the comparative

128:43

context? What's the document structure

128:45

going to be? So there's going to be a

128:47

vocabulary guide, a sentence structure

128:48

guide, a register by channel because

128:52

what you write on LinkedIn cannot work

128:53

on Instagram and what you write on

128:54

Instagram will not work on LinkedIn. And

128:57

then section number four. So what good

129:00

co writing would sound like. I'm going

129:02

to quickly give this and wait for it to

129:05

run. While that is running, what I can

129:08

do is I can actually place all of these

129:12

conversations into a project. Now,

129:15

what's a project? A project is a space

129:19

where I can keep all my conversations

129:23

that are related to one particular topic

129:27

in a folder. So now you see it says KA

129:31

consulting/brand voice system. I'm going

129:34

to do the same thing and I'm going to

129:36

add this as well into the KA consulting

129:41

project. So now when I go to project and

129:44

I go to Kova Consulting, you see that

129:48

you will be able to see this. Now this

129:53

says that it wasn't sent. It's because I

129:56

moved it.

129:59

Oh, it is working already.

130:03

Is it? It is working. Just give me a

130:06

second. I'm going to quickly send it.

130:08

And now it's making the entire brand

130:10

voice document. Okay, while that is

130:14

happening, I'm also going to open up

130:16

another new chat and I'm going to give

130:19

it the prompt for the brand kit

130:21

foundation

130:22

and the prompt for the brand audit

130:26

proposal. Now mind you, the brand audit

130:29

proposal is actually a proposal that is

130:33

for investment.

130:34

So uh the situation usually in a brand

130:39

or a proposal will be related to a PDF

130:42

and not a DOCX. So I'm going to wait for

130:44

that one because these three are the

130:46

only ones that we're going to utilize to

130:50

actually get a skill made for the brand

130:54

voice.

130:56

Okay, here's how you're going to go

130:58

about creating a skill using all the

131:01

docs that you just created. So, I'm

131:03

going to attach the four

131:07

the three documents. I keep saying four

131:09

cuz I'm teaching you four skills, but

131:11

the three documents. And I'm going to

131:13

invoke the skill creator and say create

131:17

a specific skill

131:20

that captures

131:23

all the

131:26

aspects

131:28

in these documents. Uh so instead of

131:31

attaching these documents here you can

131:33

also attach them here inside the project

131:37

so that every time you invoke it invoke

131:40

something let's say a conversation so

131:42

instead of attaching it here right you

131:44

can directly add this

131:47

in while you are reading this out. So it

131:52

will now reference those three documents

131:55

to

131:57

understand how to create the skill. It

132:01

may ask you some more questions. So let

132:03

it ask the questions and then it'll

132:06

basically talk about how it is right. So

132:10

it's now building a tight well

132:13

ststructured skill that captures all of

132:15

them. So now it's going to use the main

132:18

skill MD file. Okay, this is the skill

132:22

MD file and it's going to write

132:26

the main skill MD file of what it is,

132:29

what it has to do, what it doesn't have

132:31

to do, etc. And then it's going to give

132:35

you all the information. So, it's read

132:37

the skill creator instructions. This is

132:39

an inbuilt skill in Claude, right? And

132:45

yeah, so now you basically wait for the

132:47

magic to happen. Once the magic happens,

132:50

you have this ready for you to go. Now

132:54

while this is happening, right, uh it

132:57

can run parallel things. So now what I'm

133:01

going to do is I am going to give it a

133:04

document where I want actually a PDF to

133:09

be created. Right? Now this we were

133:10

working with dog access. Now let's talk

133:13

about a PDF and how you can do that. So

133:17

let's say the situation right so you're

133:18

Kovas brand manager and they need to

133:20

make the case to let's say six senior

133:23

partners that the firm should invest in

133:25

formalizing its brand the partners are

133:28

time poor and skeptical of the brand

133:30

spend right so Google doc is not going

133:32

to do the job for you you will have to

133:34

make a proper PDF report note what this

133:37

document actually is is actually a brand

133:40

investment proposal and not a full audit

133:42

a proper audit is about 20 pages of

133:44

research findings Right? This is a tight

133:47

executive facing document that makes the

133:49

case for commissioning one. Call it what

133:52

it is when you're briefing and the skill

133:54

will produce the right document. So I

133:56

had already generated

133:59

uh using my prompt engineering bit

134:03

a complete

134:07

briefer of what my investment proposal

134:10

should look like. And I'm just going to

134:12

quickly send it and it's going to do an

134:14

entire audit and going to use the skill

134:17

MD for PDF creation and create a PDF.

134:20

See the brief is written the way you

134:24

would brief a consultant. So there are

134:25

specific findings, the specific data

134:27

points, explicit audience that is

134:29

mentioned, explicit purpose that's

134:31

mentioned, the 71%

134:34

uh client survey that is mentioned there

134:38

as well. And um

134:42

you know it's going to be the stat

134:45

that's mentioned is going to be very

134:47

interesting for the senior stakeholders.

134:50

So there is specific numbers that make

134:53

the case in a way that the vague

134:54

observations don't. The stat

134:59

that stat came from a fictional internal

135:03

survey for KA but the logic still holds

135:06

right. So find a specific number and put

135:08

it in the brief so that you can utilize

135:10

it for your brand. And now I'm going to

135:13

quickly Okay, by the time my

135:17

formalization business case is happening

135:20

here, I am going to quickly look at the

135:22

brand creator and the brand writer skill

135:25

that they have made. They have utilized

135:28

all the different things and now I have

135:30

a skill that does it for me. In fact,

135:34

your brand references, it is full of

135:37

guidelines that we had first generated

135:41

with the breeze. Right? I could have

135:44

done this without the documents also,

135:45

but the documents can act as great

135:47

references. So now what I'll show you is

135:50

how you can manage the skill and add

135:54

references here. So for example, there

135:56

is a markdown file, right? So let's say

135:59

in the skill I want to add more files.

136:03

So I'm going to say uh something like

136:07

edit with claude

136:11

and let's see what happens. I have not

136:14

tested this out myself so I am figuring

136:17

this out with you.

136:20

So, I'm going to quickly put these three

136:23

and say

136:26

I want

136:29

to add these as

136:32

reference

136:34

documents

136:36

in

136:38

the

136:40

in the skill,

136:43

right? You can we have also talked about

136:44

how you can tweak the actual skill

136:48

and let's see whether it's going to be

136:50

able to add all of these

136:54

texts from these documents into the

136:58

brand references. So while it's doing

137:01

that um

137:04

brilliant

137:06

now you can cross question it.

137:14

See?

137:20

Yes.

137:22

So now it's going to quickly just edit

137:25

it. And once this is done, now I've

137:27

shown you how to edit your skill from

137:30

the back end as well. So now it's adding

137:34

the parts that were missing and you can

137:36

see completely what's going on.

137:40

Okay, one thing that is worth flagging

137:43

here is that I have created this right

137:46

but all of these numbers are

137:48

illustrative and the ranges are also

137:50

illustrative. So it has used this PDF

137:53

using report lab and it has made the

137:55

entire PDF. So now you see that with the

137:58

briefing like this where you tell it the

138:00

key findings what it to do and they kind

138:03

of create like an executive summary

138:06

report that can be done right here right

138:10

and it's going to be extremely easy. So

138:12

it's given a proper thing for now the

138:15

formatting may be off here and there

138:17

which it is. Uh I have no idea why what

138:20

is happening with claude in terms of

138:24

it's

138:25

uh see it's now all right sometimes you

138:28

just have to refresh it. It just goes a

138:30

little haywire right but now you have

138:32

this entire document and this is

138:34

something that you can send to your

138:36

brand manager to actually uh to send to

138:38

your hire officials not your as a brand

138:41

manager to get some money that you can

138:44

utilize to make your life better. Now um

138:48

so what we've seen is from a brief to a

138:50

propos to a professional PDF proposal

138:54

and that to in about 5 minutes. So this

138:59

is what it is and this is what it will

139:02

look like across a brand manager's ear

139:06

uh you know across a brand manager's ear

139:08

of the kind of things that they're

139:09

doing. You see brand managers produce

139:11

the same document categories

139:15

um on repeat whether it's guidelines,

139:18

voice systems, briefs, reports,

139:20

proposals. It's just that the format

139:22

varies but the pattern does not. We're

139:26

going to be covering five skills today

139:28

from one source. So, I'm going to have

139:30

the agency report from the Midian Media

139:33

Orena campaign and we're going to read

139:36

the PDF, analyze it using the XLS skill,

139:41

make a leadership deck using PPTX, have

139:44

a LinkedIn post crafted, and then use

139:47

the humaniz humanizer skill by Blader

139:50

Insight Claw to make it human and

139:52

natural and ready to post. Right? So the

139:55

same campaign data with zero tool uh

139:58

switching end to end I'm going to be

140:00

covering in this video.

140:04

Okay before anything let me be specific

140:07

by what I mean by chaining because it's

140:11

more important than just using multiple

140:13

skills in a row. Right? So in a standard

140:16

workflow right you probably use a tool

140:19

you get an output. So if I were to say

140:23

here you have an agency report then

140:25

you'd probably copy paste it reformat it

140:28

and then send it to the Excel where you

140:31

will again rewrite and adapt and build

140:33

the slides in PowerPoint. Then maybe you

140:37

would use this you know to write uh your

140:40

LinkedIn post and then probably strip

140:43

the AI tone and decide to humanize it.

140:46

Right? So each tool has a gap right uh

140:51

between the next one. Then this is a

140:54

human step that prepares the output for

140:58

the next input. A scale chain

141:00

essentially eliminates these gaps.

141:04

Right? The output of each skill is

141:06

already formatted as the input for the

141:10

next skill. So you're not preparing

141:12

anything between the steps. So you're

141:13

just handing it off directly, right? So

141:17

the manual handoffs get eliminated and

141:20

the hours disappear. So now you have one

141:22

chain with no gaps and the speed

141:24

automatically compounds. See the analogy

141:27

that I keep coming back to is the

141:30

difference between a relay race and a

141:33

series of sprints. Right? In a series of

141:36

sprints, each runner finishes, hands the

141:41

bat to let's say the next to the the to

141:44

the maybe a coach and then the coach

141:46

decides what to do with it and

141:49

eventually the next runner starts.

141:51

Right? In a relay on the other hand

141:54

where it's not a series of sprints, the

141:58

first player or the first runner

142:00

automatically gives the batn in the best

142:04

way that can be carried by the second

142:07

runner. Right? The batn goes directly

142:10

from hand to hand. Nothing is lost

142:12

between the steps. So the speed

142:14

automatically compounds here. And that's

142:17

what going to what we're going to see

142:19

today. five skills. Each one's output

142:23

becomes the next one's input. So, it's

142:26

one campaign worth of data processed end

142:28

to end without you managing the handoffs

142:32

manually.

142:35

Before I even run the chain, I want to

142:37

name what this workflow looks like

142:39

manually. See, I don't want to be

142:41

dramatic about it just because the

142:44

contrast is going to be the point. Well,

142:47

you receive the agency report, right?

142:49

and it probably receive it as a PDF.

142:52

Once you open it, you'll probably read

142:54

through it, find the entire narrative,

142:56

right, with the actual numbers, the

142:59

copy, the performance table into the

143:01

Excel sheet, right? Reformat the columns

143:04

maybe to match your own naming

143:06

convention and this would take

143:08

approximately 40 minutes, right? Then

143:10

let's say you want to build a debrief

143:13

deck in PowerPoint. So, you maybe

143:15

already have a template if you, you

143:17

know, uh are somebody who has their own

143:19

brand strategy. In case you haven't seen

143:22

how to make your own brand strategies,

143:24

um well, there's a video in this same

143:26

series. I will link it up here so you

143:28

can go watch that. And uh let's say you

143:32

have your own template, right? You

143:34

create the slides from the results. You

143:36

have the channel breakdowns, what

143:38

exactly worked, what didn't, and the

143:41

recommendations,

143:42

everything, right? you write it down

143:44

that's another 90 minutes off the bat

143:47

right minimum if you're working quickly

143:49

I'm talking about this 90 minutes a

143:51

minimum sometimes uh you know my friends

143:54

in corporates they would spend about a

143:56

day just making a deck right let's say

143:59

you do this as well right and now you

144:01

think that oh I want to write a LinkedIn

144:03

post out of it um seeing what are the

144:06

key insights from let's say this

144:08

particular campaign so you draft it then

144:11

you edit it then you strip the AI tone

144:13

out of it, right? If you use something

144:15

like Claude 2 drafted and that's another

144:18

30 40 minutes spent there. So total

144:21

somewhere between 2 and 1/2 to 3 hours,

144:25

right? Spread across 1 2 3 four tools at

144:29

the moment without

144:32

uh you know using any kind of uh AI

144:37

tool. you'd probably be using multiple

144:39

tools having manual hands off and the

144:42

time gets lost in these gaps right so a

144:48

scale chain workflow a connected one

144:50

here as I have shown would probably work

144:55

better than the manual workflow

144:59

I'm going to be using the same agency

145:01

report that we used back in episode

145:03

number six you can check that out up in

145:06

the icard the episode number six if you

145:07

haven't seen it but we were working with

145:09

Meridian media's Diwali campaign rap for

145:13

Orenda skincare so if you haven't seen

145:15

that episode this is basically a

145:17

fivepage PDF with a narrative right and

145:21

a channel performance table inside it so

145:25

the five skills that I'm going to be

145:26

working with I am going to be using

145:29

claude for that so I'm going to quickly

145:31

attach this here um it says that I'm out

145:34

of usage credits but I am Not uh I

145:37

checked my settings. I think there is

145:39

some issue from Claude's end that this

145:41

error message is popping up. Um all

145:45

right so I have attached this in the

145:49

chat section right it's asking me how

145:51

can I help you right so let's uh look at

145:55

PDF reading first so I am going to make

146:00

claude read the PDF right so I've

146:02

uploaded the report and uh you've seen

146:05

this step before multiple times right so

146:07

the skill reads through the PDF pulls

146:09

out the structured data it ignores the

146:11

pen padding all of it Right. And I'm

146:14

going to ask to uh for it to make a two

146:18

tab spreadsheet. So I'm going to say

146:22

turn this PDF into a two tab

146:26

spreadsheet,

146:29

right? Uh

146:31

what all do I need to have? So tab one

146:34

is going to be let's say about channel

146:36

performance

146:40

data

146:42

from this report and tab two for that

146:46

matter let it be channel by channel

146:50

analysis right

146:53

uh I want the verdict of scale let's say

146:58

maintain

147:00

or even cut for that matter

147:03

for each channel,

147:08

let's say with the key numbers

147:11

uh that support it.

147:16

Right? I'm going to be quickly running

147:18

this. If you see that the error message

147:20

is there, but it's still working. There

147:22

is some backend problem, but it is now

147:24

working. It's reading the Excel skill

147:27

and it'll also use the PDF. Uh, so it's

147:32

already extracted it from there and now

147:34

it's going to be turning this PDF into a

147:36

two tab spreadsheet.

147:38

So also from episode 6, right? Raw data

147:41

in one tab, channel vertics in the

147:44

other. If you watch that episode, you

147:45

know exactly what's about to come. So

147:47

watch what happens after this one.

147:55

All right, good. So now I have it in the

147:59

form of let's say a spreadsheet. It has

148:01

all the channels that are mentioned, the

148:02

performance of each channel, what it is,

148:05

whether to scale, whether to cut or

148:07

review or even maintain. Right? Um, we

148:10

have the analysis and the performance

148:12

both in different tabs as requested. If

148:14

you see here, Google display consumed

148:17

about 9.1 times the return on ad spend.

148:21

So, definitely cut or review. So uh the

148:24

verdict is cut or review because it's

148:26

about you know uh consuming 18% of the

148:31

9% of the total revenue share right so

148:34

the verdict it's cut and that's the

148:36

inside and we'll hold on to this because

148:37

this is what we will need in step number

148:40

four right now let's move on to the PPTX

148:43

skill uh I have a prompt that I already

148:45

written about let's say a campaign

148:48

debrief tech for leadership team the

148:50

source data is the channel performance

148:53

uh analysis

148:55

above. So I'm just going to quickly

148:57

above. So I'll lead with the headline

148:59

result, the return on ad spend, uh

149:02

whatever there is and all of it, right?

149:04

I need 10 slides in total. So I've

149:06

mentioned that and I've given this to

149:09

them. Now here's what's new, right? In

149:12

episode 1, I wrote a brief by hand to

149:15

feed it into PBDX. In episode 5, the

149:18

marketing manager probably wrote a de

149:20

debrief brief. This time the data from

149:23

Excel analysis is the brief. So the

149:26

chain is actually feeding itself. The

149:28

output of step two became the input of

149:31

step three without me writing anything

149:33

new. Right? Uh so if you see if you if I

149:36

take a moment here it's already

149:37

generating and once it's generated I'll

149:40

be back.

149:43

Okay folks the output is here and here

149:46

is what you can see we meet the target.

149:49

So that's good. They've given the entire

149:52

campaign right in different uh methods.

149:56

They've put this here as well. So there

149:59

are three recommendations right one is

150:01

scale Instagram. Number two is uh

150:05

cutting or restructuring the Google

150:06

display ads or capping the emails to two

150:10

cents. Right? Um and also refining the

150:13

influencer creator mix here. Right? So

150:16

that's the debrief deck. So every time

150:19

that you know somebody wants to actually

150:21

make this they can directly use it. So

150:24

they have done this. Now let's

150:27

understand uh how we can use this

150:30

debrief deck as let's say uh something

150:35

that's going to allow me to attach this

150:39

debrief deck in let's say uh LinkedIn

150:43

and uh write

150:47

let's say uh LinkedIn post related to

150:51

this and how we are learning the lesson

150:54

from let's say our last campaign uh

150:57

about Google display ads and what

150:59

happened and how you know we have

151:02

figured it out. So I had made this. So

151:04

now let me just quickly put this in. So

151:08

it should be professional direct and

151:10

insight first. So it's reading my

151:12

humanizer perfect skill. So this

151:15

humanizer personal skill is basically a

151:17

skill that I created uh so that it get

151:21

the general humanizer skill also gets

151:24

refined according to my own personal

151:25

tone. So now it's going to be using

151:28

that. So using the same caption

151:30

generator skill right for LinkedIn that

151:32

you saw from episode number four. The

151:34

source material has changed but in

151:35

episode 4 we were generating let's say

151:37

from social content uh you know for

151:41

let's say a particular campaign topic.

151:43

Here what we're doing is we're

151:44

generating it from specific data point

151:46

that came out of an Excel sheet. The

151:48

insight isn't invented at all. It just

151:50

came from the chain. Right? So let's see

151:53

here. What does it say? We ran Google

151:56

display with 18% of our budget last

151:58

campaign. It delivered 9% of revenue.

152:01

ROAS was this much. Campaign average was

152:05

4.6.

152:06

Right? We're cutting it for the next

152:09

one. Uh and honestly the bigger lesson

152:12

isn't about Google display. It's about

152:13

the habit of keeping a channel in the

152:15

mix just because it's there because it's

152:17

familiar because it ran last time. So

152:19

this is writing in the same tonality

152:21

that I write my papers in. So good

152:24

structure, right? It's also insight

152:27

first and there's a specific number.

152:28

There's a clear decision. Now the last

152:31

step of humanization is already applied

152:34

in here because it automatically invoked

152:37

my personal humanizer skill. uh because

152:40

I'd already written something about

152:41

tone, right? Let's say if uh I were not

152:45

to do this, right?

152:48

If I were to, you know, not invoke let's

152:51

say the humanizer personal skill here, I

152:53

probably just say /humanizer

152:55

and write something like humanize the

152:58

above answer, right? And that's how it

153:00

would be. So now we've seen the version

153:04

that goes out that is the version that

153:06

is completely humanized

153:08

from

153:10

uh you know the Excel which was then

153:14

converted into a debrief PBD. So we have

153:17

seen five steps coming out of one agency

153:19

PDF in so the spreadsheet the leadership

153:21

deck and a LinkedIn post out right total

153:24

time for all of these five steps were

153:26

probably the time uh that was this

153:29

entire video. So here you know

153:34

we are not doing the same work that can

153:37

be done manually uh across different

153:40

tools with reformatting in the gaps

153:43

right that's the better part of the

153:46

morning and this is how you can optimize

153:49

your time and save up time in your

153:52

workday.

153:54

See, the practical implication of a

153:56

chain like this is not just speed. It's

153:59

the category of work that used to

154:02

require blocking out a morning that now

154:04

fits inside just 30 minutes. That

154:07

changes what you actually put in your

154:08

calendar. So, a campaign brief that

154:11

previously needed only 2 hours of

154:12

focused document production time can

154:16

happen the same morning the report even

154:18

arrives. And the LinkedIn post as well

154:21

can be crafted that could have been

154:24

delayed you know can be crafted

154:27

in the same time that someone had to

154:30

write it um comes out of the session as

154:33

a debrief right so one honest note on

154:37

these change uh chains though they

154:40

require clean handoffs so if your source

154:43

PDF is poorly structured and it's heavy

154:47

on images or is scanned and rather than

154:50

exported the extraction in probably step

154:53

one will be rest reliable and that will

154:56

then affect the entire chain. So it's

154:59

extremely important to also check the

155:02

output at each step and the input

155:06

automatically then gets checked right

155:09

the chain is only as good as its first

155:11

step for well structured agency reports

155:14

right which is probably most of them

155:15

this will run cleanly for many source

155:18

documents you might need to tidy the

155:20

step one output before feeding it

155:22

forward. See the broader point is that

155:26

start looking at workflows not as a

155:28

sequence of tools but as a sequence of

155:30

handoffs. So where are the gaps? Where

155:33

are you doing manual reformatting

155:34

between steps? Those are gaps that can

155:37

be fulfilled by building chains. Before

155:42

anything else, right, I want you to

155:44

first understand what actually a skill

155:47

is and what are the components that are

155:49

part of let's say this skill. Right? So

155:52

understanding the anatomy of it. So

155:55

number one is a trigger condition,

155:57

right? So when should the actually the

156:00

skill activate, right? Whether it's when

156:03

a campaign brief is input, whether a PDF

156:05

report is there or a brand review is

156:08

requested. So understanding what

156:10

activates this skill, what kind of

156:12

output uh is required from it comes at a

156:15

later stage. But understanding what kind

156:17

of input or context tells Claude in its

156:20

skill mode rather than you know uh the

156:22

general chat mode for a production skill

156:25

this is usually a file right and usually

156:27

a file type or a task type so a campaign

156:29

brief or you know a PDF report for a

156:32

review skill the kind where you have to

156:34

actually review work it is probably

156:36

content plus the instruction to evaluate

156:39

it now the second is what is the task

156:43

right because having a trigger just

156:44

means that you activating it. So imagine

156:47

it as a domino. So the first part of the

156:50

domino is like you know when you get

156:52

dominoes to fall the first step that you

156:54

do is actually take your finger point it

156:57

at the domino and then push the first

157:00

block. So the pushing the first block

157:02

the act is the trigger. The second is

157:04

how do you define a task right? So the

157:07

task of all the dominoes that are

157:08

falling one by one. So what does the

157:11

skill actually do? Not weekly but

157:13

precisely. So instead of saying, you

157:15

know, review this copy, well, that's not

157:16

a task definition, you'll probably say

157:18

something like check this copy against

157:20

the following brand voice criteria and

157:23

flag anything that viates them, right?

157:26

And that do with a specific reason and a

157:28

suggested alternative. That's a task,

157:30

right? And that's a task definition. The

157:33

precision of this component determines

157:35

whether the skill produces useful output

157:38

or generic feedback. Now the third one

157:41

in this uh you know anatomy is the

157:44

output. So what should the actual even

157:46

result look like? So the result of all

157:49

the dominoes falling is the dominoes

157:52

putting themselves in a pile. Now here

157:54

we will understand what does the result

157:56

look like? What format what sections

157:58

what level of detail you know a

158:00

production skill output uh probably is

158:03

going to be a file. A review skill

158:05

output could be a structured verdict and

158:07

that structure of that verdict also

158:08

needs to be defined otherwise you'll

158:10

just get like one big chunk with you

158:12

know a paragraph of general feedback

158:14

that's even useless. It's not even

158:16

something that you can you know act on.

158:18

And then the fourth component in this

158:20

entire anatomy is the rubric that is

158:24

what standard should it follow. Right?

158:26

So just getting it to create an output

158:30

does not necessitate that the output is

158:33

going to be correct. It could be that

158:35

the output is wrong. Right? So you need

158:37

a fourth component that basically

158:39

reviews these skills based on brand

158:41

voice, compliance rules, formatting

158:43

standards or even checklists. Right? And

158:46

this is going to act as a rubric, right?

158:49

The standard against which the content

158:51

is judged. Now this is what makes a good

158:54

review skill. Okay? and it's going to be

158:56

genuinely useful rather than asking some

158:58

sort something like okay is this good

159:00

because a rubric actually contains your

159:01

criteria like your band phrases your

159:03

tone principles your specific dos and

159:05

don'ts right without a rubric the skill

159:07

is judging against its own general sense

159:09

what is good writing so you can't really

159:12

judge whether the copy that you've

159:14

written is actually good or not but with

159:16

a rubric it's judging against yours and

159:18

that's the difference between a

159:20

pre-built skill and a custom one

159:22

pre-built skills have these generic

159:24

standards that are kind kind of baked

159:25

into the system and a custom skill has

159:28

your standards baked in. So it's very

159:31

very specific to you. Now let's quickly

159:34

understand of how we can do this using

159:36

claude. So let's get on with it.

159:40

Okay. So before I even get on with

159:42

actually understanding and developing

159:44

this custom skill with claude, if you

159:47

look here, here is you know a quick

159:50

snapshot of a difference between a

159:51

custom skill and generic AI and why

159:53

prompting generic AI can cause problems

159:56

and why a custom skill is needed. I've

159:59

already explained this in the previous

160:00

section but if you want to just quickly

160:02

go through this you can take a

160:03

screenshot even even save it for future

160:05

so that you can understand what exactly

160:08

happens here. and what happens in these

160:11

systems when you actually create a

160:13

custom skin. So this is what a dashboard

160:16

looks like, right? We've already seen

160:18

this. We have been very familiarized

160:20

with this. We also have different kinds

160:22

of models. We saw how we can utilize the

160:25

different kind of models for specific

160:27

use cases. Now uh this one I have

160:30

specifically created a this this time

160:33

I've specifically created a free uh

160:37

claude account so that the viewers who

160:40

are viewing this who may okay uh we are

160:44

in the panel and I have opened the site

160:47

pane in the side pane you can look at

160:49

customize. So when you go to customize

160:52

you see the skills that are available to

160:54

you. Now before I even click skills

160:56

right you can already see here that

160:58

Claude actually asks you how you can do

161:00

this. So you can connect your own apps

161:02

you can create new skills and you can

161:04

even browse plugins. Now what are

161:06

plugins? These are basically pre-built

161:07

knowledge for your field. So in case you

161:09

want to understand how the entire claude

161:11

ecosystem works, you can drop a comment

161:14

below and I can explain how you can

161:16

utilize Claude for your benefit and

161:18

especially for your roles. So the kind

161:20

of roles that you're interested in,

161:21

please remember to drop a comment below

161:23

so I can actually uh look at what kind

161:26

of videos you know our audience wants to

161:29

write your own set of skill

161:31

instructions. So a skill like we

161:33

understood in the previous videos is a

161:34

markdown file. So you explicitly tell

161:37

Claude in markdown what all is required

161:39

based on the um file structure that a

161:44

skill understands and then here you can

161:46

directly upload a skill. So we had

161:48

already uploaded the humanizer skill

161:49

which was written according to somebody

161:51

else's standards. So now I just quickly

161:53

click on create a new skill with claude

161:55

right and it's given this let's create a

161:57

skill together using a skill creator

161:59

skill. Well that's a mouthful first ask

162:01

me what the skill should do. I am just

162:03

going to quickly enter this and let's

162:05

see what Claude responds. So now Claude

162:07

is going to give me I'll use the skill

162:08

creator skill to guide us through this

162:10

properly and let me check it first.

162:11

Right. So now it's checking what the

162:13

actual setup is how it's going to be and

162:16

it tells us you know what we're going to

162:18

do testing etc etc if you can see you

162:21

can see here the trigger context the

162:23

output and the test cases now the input

162:26

okay the anatomy that I mentioned here

162:28

the trigger the task the output and the

162:31

rubric if you see here all of them are

162:33

here the trigger context so when should

162:35

it actually kick in what is the output

162:37

going to look like so what exactly is it

162:39

going to do and what are the test cases

162:41

so test cases are basically checking and

162:44

reviewing for whether the skill that we

162:47

have created is working in the right

162:49

direction or not. Now, uh the skill that

162:53

I'm building today is going to be a copy

162:54

review skill for Orenda. So, this is the

162:56

brand that we've been using across this

162:58

series. And the idea is that any piece

163:00

of copy going out under the Orenda brand

163:03

name, right, gets run through the skill

163:05

before it's even published. it gets

163:06

checked against ora specific brand voice

163:09

criteria and structured um basically a

163:13

structured output that is going to be

163:15

actionable. So this is the architecture

163:17

that is somewhat in my mind right that

163:19

is to build a copy review skill for

163:22

Renda right so we're going to do this

163:24

entirely and this is the architecture so

163:26

my trigger should look something like

163:29

review this for reinda is this on brand

163:31

something like that then I'm going to

163:33

look at the brand voice rubric then I'm

163:35

going to review it right and then I'm

163:38

going to give it a structured outward

163:40

now this is the architecture for the

163:41

skill that I'm going to use all right

163:44

folks so instead of actually typing the

163:46

entire prompt. I have already, you know,

163:49

written it somewhere for myself. So now

163:52

look at this. What am I saying? Build a

163:53

copy review skill for Orena. Right? The

163:55

name of that skill is going to be this.

163:57

What is the trigger? Anytime the user

163:59

paste or uploads or copy and then asks

164:03

for a review brand check or a

164:05

pre-publish check, right? Also to be

164:07

triggered on phrases like is this on

164:09

brand, review this fora or check before

164:11

I send. What it does exactly? It

164:14

basically reviews any piece of forenda

164:17

copy whether it's ad, email, social

164:18

post, product description, anything,

164:20

right? It checks against brand voice

164:22

criteria, right? So, I want it to be

164:24

more direct and warm. I wanted to be

164:27

specific about the ingredients and

164:29

benefits. I want no fear-based language

164:32

in it. So, something like problem skin,

164:33

anti-aging, damage, nothing. These are

164:36

all I would say phrases that are banned.

164:39

Then uh from I'm going to also talk

164:41

about you know whether it celebrates the

164:44

customer's existing routine rather than

164:46

positioning against the alternatives

164:47

that are there and then also I want the

164:49

sentence rhythm rhythm to be short lead

164:52

sentences with supporting details right

164:54

then the vocabulary what kind of you

164:56

know words I don't want it want it to

164:58

use words like clinically proven right

165:01

because as a skincare brand it could be

165:02

possible that it's not yet clinically

165:04

proven it's under it's under trials and

165:06

that's what happens with most skincare

165:08

brands So I need to be very careful of

165:10

what actually I put out and then I have

165:12

basically created this and then the next

165:14

thing I have you know after the review

165:16

section of what actually it's going to

165:18

review the output format I have said

165:20

that okay you know pass pass pass with

165:22

edits or review needed there are three

165:24

things right that my senior copywriter

165:27

was doing which is either they were

165:28

passing it or they were saying that it

165:30

will pass with certain edits and an

165:33

entire rewrite that could have been

165:35

needed for that particular copyright. So

165:37

you flag the line and then you basically

165:39

give the tone summary of whether even

165:41

the overall piece sounds like orinda or

165:43

what orinda is supposed to sound like.

165:46

Now I'm going to quickly hit enter on

165:47

this and we are going to now come back

165:50

and see when and you know why it while

165:52

this builds when this completely builds

165:54

out. It is also going to ask me certain

165:56

questions right uh whether it should

165:58

check to all paragraphs. I'm going to

166:00

say all paragraphs for this one. Uh I

166:03

want all variants of this. Okay. Do you

166:06

want to test cases before we finalize?

166:09

Yes, I want to also set up the test

166:11

cases. So, it's also going to ask me

166:12

prompting questions that my instructions

166:16

may not actually specify. And now it's

166:19

done everything and now it's setting it

166:21

up. Right now, you see that the skill

166:24

creator is taking this specification and

166:26

then turning it into a configured skill.

166:28

One that probably, you know, Claude can

166:30

trigger automatically with all of these

166:31

criteria already loaded. So I don't need

166:34

to paste these brand guidelines into

166:36

every review request now. They basically

166:37

live inside the skill. So I'm going to

166:40

quickly wait till this you know creates

166:42

the entire skill MD and I'm going to be

166:44

back then.

166:48

All right. Now you see that there is a

166:50

skill.md that is there right? It's

166:52

already drafted. It has done the three

166:55

test cases. It has checked it entirely.

166:58

Now what is it asking me here? Are this

167:01

thing? How do these look? any suggested

167:03

and I'll say no it's all right

167:08

build me the skill now so it's actually

167:11

not yet built build the skill it's only

167:14

build the skill.md and now it's going to

167:16

run the script and build it into the

167:18

proper directory so now it's going to

167:20

package it so once it is actually

167:23

convinced or it has convinced me in fact

167:26

that the skill is actually you know

167:28

going to be used ahead then it now Once

167:32

the validation has been passed, it now

167:34

creates a writable output location

167:36

something right. Okay, while that is

167:39

happening, let's quickly also write a

167:41

bad copy. So, uh I want you know I have

167:46

this bad copy that I had written

167:49

here. So, I'm going to quickly paste

167:50

that as well. Now, if you see it's done,

167:53

see the name is exact same orenda copy

167:55

review. All the criteria that I

167:58

mentioned uh is mentioned here. So you

168:01

have it needs to be direct and warm

168:02

specific right no broad claims etc

168:05

vocabulary to f you know to flag and

168:08

stuff like that everything is here and

168:10

this is the output format now I'm going

168:12

to save this table quickly okay and I

168:14

had already written a bad set of copy so

168:17

this has deliberate errors in this so I

168:21

want you to I want to show you how this

168:24

works so I can just invoke it by the

168:26

slash and I've copied it so you see the

168:28

word revolutionary is here which is bad

168:30

transformative. So now it's reading it,

168:33

right? And then it's reviewing and then

168:35

it's given its verdict. What is the

168:37

verdict that an entire rewrite is

168:39

needed, right? So this is going to be a

168:41

piece of the copy. And here are the

168:43

lines that are flagged. It says what?

168:45

Revolutionary banter. You need a

168:48

suggested alternative transformative

168:50

formula. Right? So it's going to say

168:52

things like signs of aging are you know

168:55

um see again signs of aging fear-based

168:58

language alternative softens fine lines.

169:00

So it's giving you how you can fix it

169:03

right. Um it's saying that the suggested

169:05

alternative also you know needs to

169:07

probably be framed around skin gains

169:09

that is not understanding what you're

169:11

fighting against but what you can gain

169:13

and that's usually a brand's you know

169:15

voice principle uh that is going to be

169:17

showing up in this rewrite. Now this is

169:20

a good catch right that I want you to

169:21

see here that it is not just looking at

169:25

single phrases it's also looking at

169:27

context. So a simple single phrase can

169:31

also like show up in two different

169:34

places. So one is let's say naturally

169:37

nourishing powerful blend of active

169:39

ingredients right here the word of

169:42

natural you have to mention it like what

169:44

exactly is there so it's like an

169:45

unspecific claim right and because there

169:48

is no active ingredient that is there

169:50

that criteria is also something that is

169:53

you know broken. So now it has dissected

169:56

it completely and this is the kind of

169:59

non-generic grammar checker thing that

170:02

we actually want this to do. It's to

170:04

review at a higher level. Now, um you

170:07

understand the tone summary, right? Uh

170:09

the two sentences on whether even it

170:11

sounds like or it says that it's

170:12

completely bad when it comes to sounding

170:15

like the brand. So, every violation gets

170:17

caught, right? Everyone with a specific

170:19

reason, a specific alternative is also

170:21

proposed for it and that's the skill

170:23

doing the job and that currently falls

170:26

probably to a senior copywriter or even

170:27

a brand manager, right? Who's reading

170:29

every piece of coffee before it even

170:31

goes up. Now, here's probably what the

170:34

corrected version looks like after

170:36

applying the flags. Now, what I'm going

170:39

to do is I have the corrected version

170:41

with me and let's quickly paste this in

170:43

and actually ask it to evaluate this.

170:47

Okay,

170:50

let's see what's the verdict on this,

170:51

right? U maybe it's going to be better.

170:53

Maybe it's going to still say see

170:55

overall verdict passed because this

170:57

actually sounds like the brand. So, what

170:59

does it have? It has the orinda glow

171:02

serum what there is. It tells you what

171:04

all ingredients are there and then it

171:06

talks about how it can fit into your

171:08

routine. So this version would

171:10

definitely pass the skill. See the copy

171:12

review skill is actually one example.

171:14

Before I give you the broader picture,

171:16

one distinction that's worth knowing is

171:18

what category it actually fits into

171:20

because it changes which kind of skill

171:22

you would build for what kind of a

171:25

problem. Now there are three categories

171:26

broadly. If you see here you have

171:28

production based, review based and

171:30

transformation based skills. Now

171:32

production skills generate any kind of

171:34

output. So whether it's a document, a

171:36

deck, a report, right? Everything that

171:38

we've covered from episodes 1 through 8

171:41

in this particular series. Then the

171:44

review skills actually evaluate. So this

171:46

is something that we have seen right now

171:48

with the copy reviewer, right? So they

171:50

flag problems, they suggest

171:52

alternatives, they give a verdict

171:54

because that's how a review skill is

171:56

actually meant to be built. And then the

171:57

last kind is the transformation one

171:59

where we see that we take an existing

172:02

output, right? And then we rework or

172:03

repurpose the content for a new

172:05

particular need. So whether it's

172:07

humanizing, whether it's changing or

172:09

creating a LinkedIn post from your

172:12

brief, right? We've saw we've seen this

172:13

in the last you know couple of episodes

172:16

and that's something what a

172:17

transformation skill does, right? So

172:19

most marketing teams, well, they at

172:21

least need one of each. So here's what

172:24

you know what's worth building in my

172:26

head. Um because you just don't want a

172:30

chatbot that's giving you answers. You

172:32

want a workflow engine, right? Somewhere

172:34

you're giving an input, then you have

172:35

the rules set in, then you review the

172:37

logic and the output format, and then

172:39

you can actually get results into your

172:41

workflow that are extremely important.

172:44

So if your brand is specific right to

172:47

let's say um is specific enough for a

172:50

particular use case which is going to be

172:51

in most cases then instead of using the

172:53

humanizer skill right you can maybe

172:55

create a brand voice skill then find a

172:57

skill that not only just removes the AI

172:59

patterns but a skill that checks for it

173:01

right and then in terms of production

173:03

you can think about campaign briefs if

173:05

your team is producing campaign briefs

173:07

regularly and they always need to follow

173:08

the same format right then a campaign

173:11

brief is something that you should think

173:12

about creating so with your section

173:14

ctions, your names, your field names,

173:16

your agency's requirement, everything

173:17

that's baked into it, right? If you work

173:19

with clients and produce reports in

173:21

let's say a specific format, right? Then

173:23

you can have a skill that basically

173:26

changes it into a specific format. So

173:28

you reformat it for the client facing

173:31

thing that you require. If you operate

173:33

in a regulated category, for example,

173:35

BFSI or healthcare or even FMCG for that

173:39

matter, right? anything that has any

173:40

kind of compliance checks. Then you can

173:43

create a compliance review, an evaluator

173:47

skill that basically evaluates all the

173:49

pre-compliance claims that you are

173:51

making and all the claims that you could

173:53

make. Now, these are probably, you know,

173:56

skills that will also help your legal

173:57

team to review these. So, you're not

174:00

essentially replacing the legal review.

174:02

You mind that. It's about catching the

174:04

obvious problems before it even reaches

174:06

legal, right? So that legal doesn't just

174:09

point out the obvious things and be like

174:10

dude this is so obvious right it's just

174:12

going to save time for them and for you.

174:14

So the principle across all of these is

174:15

just the same. So you find your task

174:17

that your team does on repeat where a

174:20

single person is currently the quality

174:21

control layer. You basically standardize

174:23

it for everybody and say okay this is

174:26

how we're going to go ahead. This is how

174:29

the skill is going to be built and this

174:30

is how you're going to use it. So

174:32

instead of creating a first pass editor,

174:34

you basically create a reviewer, right?

174:36

And this is going to be the last demo in

174:38

this series. Um where we have unpacked

174:41

all the different kinds of workflows and

174:43

we've also unpacked what exactly it is

174:47

and how exactly we can create our own

174:49

skills for our particular workflows as

174:52

well. Let me say something to you

174:55

clearly and what I think you should

174:56

actually take from this entire series.

174:58

See, I don't want you to get lost in the

175:00

volume of content. There is so much

175:01

content on the internet itself. But this

175:04

wasn't a series about prompting either.

175:06

It was not about even getting better

175:08

outputs from Claude, you know, by asking

175:10

better questions. Um, the whole argument

175:13

from episode 1 all the way to through

175:17

this one has been about a mental model

175:20

shift from using Claude as a chatbot

175:23

that drafts for you, right? and it

175:25

drafts things for you that you can work

175:27

with to using it actually as a workflow

175:30

engine that produces finished

175:33

deliverables. That shift basically

175:36

changes how much value you get from the

175:38

tool. You know, not by just a tiny

175:41

margin but by a multiple I would say

175:45

compounding kind of an effect and value

175:48

that it can generate. The one thing that

175:50

I'd ask you to do this week, right, is

175:52

to pick one workflow um that your team

175:54

does on repeat where the output is

175:56

currently inconsistent or probably takes

175:58

longer than it should, ask the three

176:01

questions. So, does it repeat? Does it

176:03

have a specific format? Does someone

176:06

review it for consistency? If the answer

176:08

is yes to all three of them, build the

176:11

skill. It'll probably just take an hour

176:13

for us, you know, if you get well versed

176:15

with it. It took about just 15 to 20

176:19

minutes to do this and the return on

176:21

that 20 minutes probably compounds every

176:23

time the workflow runs. But before you

176:26

do a quick question for the comments,

176:28

what's the first custom skill you're

176:30

going to build? Drop it below. You know,

176:32

I read every comment in this series and

176:34

I actually want to understand what all

176:36

new content you want to see on this

176:39

channel.

176:40

And now that this series is closing,

176:43

I'll just say the question isn't whether

176:46

AI is going to change how marketing

176:48

teams work because we already know it

176:51

has. The question is whether you're

176:53

using it as a chatbot or as a workflow

176:58

engine. And I will leave it with you

177:00

guys and I will see you in our next

177:02

series. Don't forget to subscribe to

177:04

this channel to get updated with the new

177:06

series that we come up with and we also

177:09

conduct lives every week. So do check

177:12

them out as well. Thank you so much and

177:15

I hope you really enjoyed the series and

177:17

I will see you the next time. Bye-bye.

Interactive Summary

The video argues that most users underutilize Claude by treating it as a chatbot rather than a powerful workflow engine. By transitioning from simple prompting to using 'skills'—preconfigured, repeatable workflows—marketing professionals can automate end-to-end tasks like creating stakeholder decks, generating social media content, and processing reports. This shift moves the focus from manual production to efficient execution, enabling users to create finished deliverables rather than just raw text.

Suggested questions

3 ready-made prompts