Help us become the #1 Data Podcast by leaving a rating & review! We are 67 reviews away!

Is AI replacing data analysts? Here's my honest answer after testing an actual project.

💌 Join 30k+ aspiring data analysts & get my tips in your inbox weekly 👉 https://datacareerjumpstart.com/newsletter

🆘 Feeling stuck in your data journey? Come to my next free "How to Land Your First Data Job" training 👉 https://datacareerjumpstart.com/training

👩‍💻 Want to land a data job in less than 90 days? 👉 https://datacareerjumpstart.com/daa

👔 Ace The Interview with Confidence 👉 https://datacareerjumpstart.com/interviewsimulator

⌚ TIMESTAMPS

00:24 – What I gave Claude

06:00 – What it came up with

07:24 – First analysis results

19:33 – Building the dashboard

35:50 – Should you be worried?

🔗 CONNECT WITH AVERY

🎥 YouTube Channel

🤝 LinkedIn

📸 Instagram

🎵 TikTok

💻 Website

Mentioned in this episode:

May Cohort of the Data Analytics Accelerator — Now Open

🔗 datacareerjumpstart.com/daa The May cohort of the Data Analytics Accelerator is officially open for enrollment. This is my comprehensive data analytics bootcamp that takes you from wherever you are to landing your first data job. Doesn't matter your background, your degree, or your experience level — we're going to help you get there. What you get: 📊 Full curriculum covering Excel, SQL, Tableau, Python, and R 🛠️ 9 real-world projects across different industries to build your portfolio 💼 LinkedIn, resume, and interview prep so you actually stand out to recruiters 🤝 Weekly office hours, coaching, and a community of 900+ aspiring analysts who are in it with you 🎓 Lifetime access — go at your pace, come back anytime May enrollment deal: 🔥 20% off when you enroll now 🎁 6 free months of my unreleased Data Portfolio Builder tool — this isn't publicly available yet, and every May cohort member gets early access The live kickoff call is with yours truly on Monday, May 11th at 7:00 PM Eastern. Make sure you're enrolled before then so you don't miss it. 👉 datacareerjumpstart.com/daa Or just click the link in the show notes down below. See you on May 11th.

https://datacareerjumpstart.com/daa

A lot of aspiring data analysts ask

me, is AI going to take our jobs?

2

Are we cooked?

3

And my honest answer is,

let's watch and find out.

4

So I did that exactly in this episode.

5

I gave Claude code a real data

analyst project, the kind of analysis

6

that I used to spend hours on, and

I just kind of let a do its thing.

7

Here's what happens.

8

So here's the project.

9

I want to get some insights

on my YouTube channel.

10

What videos are doing well and why

they're doing well, so I can try

11

to make better performing videos in

the future and more helpful videos

12

for you guys who are watching, or

the you guys who are listening.

13

So what I did is I opened up

YouTube creator studio and

14

filtered by the last 365 days.

15

And got different things like the

content data, the traffic source

16

data, the geography data, and I

just exported all those as CSVs

17

straight to this folder here.

18

And so you can see I have a content

geography, new and returning viewers,

19

playlist posts, subscription status,

traffic source, and viewer age, zip

20

folders right here on my desktop.

21

And basically I wanna see what's

going well with my channel and what's

22

not doing so well with my channel.

23

If I were doing this analysis myself or

giving it to another fellow data analyst,

24

I would just kind of hope that they'd

go through all the data, explore it,

25

and generate some meaningful insights.

26

If they had prior domain knowledge

about YouTube videos, they might

27

know to look through, click-through

rates on the thumbnails and average

28

view durations of certain videos.

29

But if I gave it to just a normal data

analyst, they probably wouldn't have that

30

domain knowledge of YouTube, and so they

wouldn't know where to look necessarily.

31

In this case, I'm not going to use my

domain knowledge to give Claude any hints.

32

I just wanna see what it

comes up with on its own.

33

So let's go ahead and open up Claude.

34

So if you haven't checked out

Claude yet, I highly recommend it.

35

It's produced by a company called

Anthropic that's very similar

36

competitor to OpenAI, and Claude is

basically a competitor to chat GPT.

37

Now, Claude has gained a lot of popularity

in the last, I don't know, six months,

38

because one, it can write really well.

39

Two, it can code really well.

40

Andro has basically decided that it

doesn't want to be good at everything.

41

It just wants to be good at

writing and good at coding.

42

And here's the truth is when

you're good at writing and good

43

at coding, you're very powerful.

44

So they have this chat interface

that's very similar to Chat GPT,

45

but they also have this thing

called Claude Code right here.

46

Now, Claude Code is

basically the coding version.

47

Of Claude, it's more designated

for getting tasks done,

48

specifically coding tasks.

49

And you can use the CLI, which is

basically the command line interface.

50

Basically, that would look like you

opening up your command prompt and

51

then typing in Claude right here.

52

And then boom, Claude Code pops up here.

53

So you can have like a command line

terminal version of Claude Code.

54

Or you can just use the desktop app,

which has Claude Code right here.

55

Now, personally, I like using the desktop

app because terminals and command lines

56

still make me a little bit scared, and

it's just harder to know what's going on.

57

So I'll be using the desktop

version of Claude Code.

58

Now, I will say that I pay a

hundred dollars a month for

59

Claude's Max subscription plan.

60

Basically, that means I have the

very powerful version of Claude and

61

I can use their most powerful model.

62

Opus 4.6,

63

which is basically just their

smartest, most sophisticated,

64

best model that they have.

65

They've also recently updated

it to have a 1 million context

66

window, which is very powerful.

67

Basically, it allows you to have

a lot of context, which can be

68

good if you have a lot of data.

69

Now, if you were to change models

from Opus to something like sonnet,

70

this is their less powerful model.

71

Now, sauna is still really good, but if

you were to change it to Haiku, which

72

is really just their weakest model.

73

It probably wouldn't do as well.

74

So depending on what model you're using,

the performance will definitely change.

75

The results will definitely change.

76

So I'm just gonna use the most

powerful model just to see

77

what we're working with here.

78

Now, recently when I'm using AI

tools, uh, I have really gotten

79

into dictation, and the reason

being is I kind of suck at writing.

80

I kind of suck at typing, and if

I can just brain dump my thoughts

81

verbally, I find that to be a lot

more effective than me filtering my

82

thoughts via my typing with my fingers.

83

So I'll just go ahead and click

the record button right here.

84

In this folder, you'll find a set

of zip folders that contains the

85

analytics for my YouTube channel

directly exported from YouTube Studio.

86

I'd like you to analyze these data

sets and provide me meaningful,

87

actionable insights on what I

could be doing better on my YouTube

88

channel and what I'm doing well.

89

And that's the prompt.

90

I'm literally going to give Claude,

and I'm just gonna press this go

91

button and see what it comes up with.

92

It's gonna ask if it trusts this

workspace, and it indeed does.

93

And you'll notice that I

have that folder right here.

94

It's literally attached

to this folder right here.

95

And that's the thing is cloud

code isn't working in the cloud.

96

It's working here locally on your machine.

97

So literally it has this YouTube

data analysis folder, the

98

contents of which are inside.

99

Now, I'm kind of curious.

100

It's gonna ask me a lot

of permission things.

101

I'm gonna literally just do, uh,

allow always for this project

102

for right now, uh, I'm curious if

it's gonna be able to unzip these.

103

It should be able to.

104

Um, but we'll see what it actually

does to, to make this go through.

105

So it's just like, Hey, do you wanna make

sure that I can run all these commands and

106

I'm gonna let it run all these commands

because I, uh, trust it for right now.

107

Now you obviously probably wanna

be a little bit more careful than

108

me when allowing these AI agents

to do things on your desktop.

109

You probably wanna make

sure you know that.

110

What's actually going on.

111

But to be honest, I'm a little

bit lazy and I just, I trust her.

112

So far so good.

113

Obviously there's been a lot of issues.

114

I know that there's been one machine

learning content creator who like deleted

115

an entire database completely with cloud

code, so you, you gotta be careful and

116

make sure you know what it's doing.

117

So right now it basically looks

like it created a user memory for

118

me and it's compiling the analysis.

119

It, read all the key files,

understand the picture.

120

It's running these bash commands.

121

It looks looking like it's

getting the CSV files.

122

And this is, yeah, look, this

is what's in the CSV file.

123

You can kinda see one of my titles,

124

Then you're seeing some of the numbers,

views on a certain date or something like

125

that, and you can see that it did that

with, um, post data, the content data,

126

the playlist data, so on and so forth.

127

So you can kind of see.

128

What it's thinking about, let's

see what it's doing right now.

129

Um, it, it just created like a, a file,

a memory file for me about, you know,

130

who I am and what this project is about.

131

And it says Avery Smith is a, oh, okay.

132

Let's see what it, it

came up with some stuff.

133

I don't know what it's doing.

134

Ah, okay.

135

Here's my full analysis

of your YouTube channel.

136

Over the past year, we've had 1.7

137

million views, a hundred thousand

watch hours, 29,000 subscribers

138

gained about $8,000 in ad revenue, 30

million impressions, a click through

139

rate of 3% average view, duration of.

140

Four minutes and 442

videos in the dataset.

141

What we're doing, well, massive reach

with how to become a data analyst content.

142

Your top two videos alone account for

22% of the total, um, which is how I'd

143

become a data analyst in 2026 and how

144

I basically just changed.

145

The 2025 video to 2026.

146

Uh, let's editor, let's,

let's rule that out.

147

If you're wondering why those video

titles are the same, it's because one

148

of 'em is actually for 2025, not for

149

purposes, that it would be normal.

150

That way people could actually

find the video still and be

151

interested in watching it.

152

'cause although some things have changed

from 20 25, 20 26, a lot remains the same.

153

This also drove 12,000 subscribers,

which is 41% of my subscribers

154

gained two strong subscriber

conversion on career advice video.

155

Videos like I analyzed 2,893 Data

Analyst jobs don't miss these six

156

entry data levels jobs, and nine huge

lies about becoming Data analysts

157

have excellent sub to view ratios.

158

Your audience converts to

subscribers when you deliver data

159

backed actionable career insights.

160

All right, let's get to know Number three.

161

Excellent watch time on long

form interview content, your

162

podcast slash interview video.

163

So I, I like that it.

164

Figured out that I have like podcasty

style videos and interview style

165

videos versus just like YouTube videos.

166

I think that's really interesting

that it already gleaned that

167

'cause I didn't tell it that.

168

Right.

169

Um, let's see How to earn 250,000

with multiple remote data jobs.

170

13,000 views, but 2000

hours of watch time.

171

Data analyst mock interview, 25,000

172

time, and an 8% click-through rate.

173

Wow.

174

See, that's really interesting.

175

Playlist driven watch time is

strong in your data career.

176

Uh, podcast playlist alone drives

658 hours of playlist watch time.

177

It's just listing a bunch of things.

178

I don't know.

179

Some of it's actionable.

180

Uh, we'll see if it suggests things.

181

Um, audience demographics,

basically 25 to 44.

182

Mostly us, India, uk, and Canada.

183

Um.

184

Browse features and suggested videos

are your top engagement sources.

185

By watch time, this actually means YouTube

is actively recommending your content.

186

The algorithm likes you.

187

Oh, I'm glad to hear that.

188

Sometimes it feels like it doesn't.

189

Alright, what needs improvement?

190

My CTR is below average.

191

YouTube considers four to 5% healthy CTR.

192

You're at 3% overall.

193

Some of your biggest videos in the 2.4,

194

2.8%,

195

this means your thumbnails

and titles are underperforming

196

relative to your impressions.

197

Your highest.

198

CTR videos give you

clues about what works.

199

Okay.

200

Then it says How to export

Tableau dashboard to image 15%.

201

So this is interesting because me having

those domain knowledge and actually

202

like being a YouTube content creator

and knowing YouTube, I know that this

203

is bad advice because let's, let's go.

204

In fact, let's go to how to

exports Tableau dashboard to image.

205

And if you actually just go to

Google and you search that I'm gonna

206

be the first video that pops up.

207

Right, and so when you're

ranking number one on Google.

208

You're going to have a high click

through rate because most of this

209

video's traffic comes via search,

not via browse or suggested videos.

210

So with search, you're like already

know what you really want, right?

211

You're like, this is what I want.

212

And when someone's like, Hey, here

you go, you're gonna click on it.

213

So that's why you have a high.

214

Click through rate versus if you're just

like on, you know, YouTube in general,

215

just on the like, uh, homepage, right?

216

Mm-hmm.

217

Right here, it's like, I

don't know what I want.

218

Do I wanna watch the, this is a

new chapter, or how I built a lead

219

magnet or, um, what is Databricks?

220

Those types of things.

221

Right?

222

Oh, editor, let's redo that

because I don't want 'em to see

223

that I'm watching this video.

224

When you open up YouTube, it's like.

225

What do you wanna watch?

226

Do I wanna watch you know,

data with Barr right now?

227

Do I wanna watch Mark Lowe?

228

Do I watch Tom Scott?

229

What am I in the mood for?

230

Maybe I'm gonna play

this Burger Life game.

231

There's just a lot more distractions and

earning the click here is a lot harder

232

to do than if you're like, you know.

233

How to export to Tableau

dashboard to image.

234

Oh, okay.

235

I'm number one.

236

Click on Google for, or

YouTube for this click.

237

Like, that's the reason why that

click through rate is high is

238

because it's mostly search base

versus browse or the homepage.

239

Um, same with these videos.

240

These are all going to be,

um, higher search videos.

241

This one's not, which is interesting.

242

Um, I'll admit, becoming a data

analysis is sustainable right now.

243

Um, so that ones may be

interesting to look into.

244

Titles with specificity,

controversy, or clear utility.

245

Get clicked more.

246

You're brought how to

become a data analyst.

247

Titles get massive impressions, but lower

CTR consider AB testing thumbnails on your

248

top impression videos, even with a 0.5%

249

improvement.

250

That's 30,000 more views.

251

View your retention is short.

252

Yes, I agree.

253

Let's see.

254

Um, action.

255

Front load your value, your

short form content isn't

256

driving meaningful engagement.

257

That's why we stopped posting

short form, uh, engagement.

258

Really, to be honest, 89% of

views come from non-subscribers.

259

Let's see your attorney viewers have a 3.7

260

click through rate versus a 2.81

261

for new viewers.

262

And watch a minute, 20 longer,

add stronger CTAs for subscribing

263

mid video, not just the end.

264

Consider a subscriber CTA

in the first two minutes.

265

Okay?

266

If you guys are watching right

now, hit subscribe so that way

267

you can make Claude and me happy.

268

Right?

269

And that's, I'm doing my CTA right here.

270

Hit subscribe right now.

271

Um, okay.

272

YouTube search is high volume,

but low engagement brings

273

in a lot of YouTube search.

274

Your number one source by view count,

but it doesn't have long view duration.

275

I don't even know if that's true.

276

I'm gonna go back to the actual

raw data here, and I'm just

277

gonna go to traffic source.

278

Oh, wow.

279

See, I don't think I knew that

that search, which is this.

280

Now see, oh, I don't know.

281

Search is this, um, blue

bar right here, right?

282

And it's saying it's my number one

traffic source over the last 365 days.

283

And I don't know if it

is browse, is this green?

284

And it looks like it's above blue

most of the time, except for, um,

285

a little bit of a period last year.

286

I don't know.

287

That's really interesting.

288

I don't know if blue is always higher.

289

It's more steady, which is really nice.

290

At suggested videos and browse

videos are very similar.

291

So if you take green plus yellow,

that's always gonna be bigger than blue.

292

But I guess it's not combining

those, it doesn't know, once again,

293

it's missing that domain knowledge.

294

Browse and suggested are very

similar style type videos because

295

once again, you're, you're having

to earn the click in those.

296

Um, maybe, maybe I'm over assuming how

similar they are, but from my domain

297

experience, these two are very similar.

298

Okay, interesting.

299

Action search viewers are

looking for quick answers.

300

Consider creating dedicated answer videos.

301

Under five minutes optimize

for search while keeping your

302

longer career advice for browse.

303

Interesting.

304

Your Tableau export video.

305

Of a 15.64%

306

CTR proves this model.

307

Well, okay, it proves that we have

a high click-through rate, but do

308

I get anything from that video?

309

I'd argue not really.

310

Let's open it up and look at the

data a little bit more closely.

311

Okay.

312

I have that video popped up.

313

As you can see, it has

37, almost 38,000 views.

314

473, uh, watch hours.

315

It's only gotten me

$135 and 73 subscriber.

316

So although this video is getting

a lot of views, like how am

317

I really benefiting from it?

318

I've made, let's see, I guess,

how many dollars a day do I make?

319

135.

320

I make like 10 cents a day

from this video, right?

321

1 36 divided by 14, 16 days.

322

Yeah, I make 9 cents a day with

this video, um, which isn't

323

nothing, I'm grateful for it, but

like, that's not really the point.

324

Like even though this has an

unreal, uh, click through rate,

325

like this, this click through rate

is really, really high, right?

326

Like at 13%, I guess, since published.

327

When you compare it to a video

like this, how to become a data

328

analyst in 2026, like, look at this.

329

So many more subscribers, so

much more money, you know?

330

And even though our click through rate.

331

Is really low.

332

It's probably like 2% for this video.

333

Yeah, look at the

click-through rate's only 2.8%.

334

Even though the click-through rate kind

of stinks, like the overall number of

335

subscribers and revenue is way better.

336

So I don't know why it's kind

of fixed on, like the idea that.

337

Clickthrough rate is everything.

338

Now clickthrough rate's really important,

but just 'cause your clickthrough

339

rate is high doesn't mean that it's

good for you as a YouTube creator.

340

But I do kinda like the idea of

creating dedicated answer videos under

341

five minutes for optimized search.

342

That is an interesting concept.

343

India viewers have low

engagement shorts aren't working.

344

Yep.

345

Uh, either invest seriously

in short to the clear strategy

346

or stop spending time there.

347

And that's unfortunately.

348

Uh, what I've done is we

just haven't done any shorts.

349

Now, we might start doing shorts in a

little bit, but um, we'll have to see.

350

End screens only drove 2,800

351

That's a very small amount,

so yeah, you're right.

352

We don't use end screens nearly as well as

we probably should, which is unfortunate.

353

Okay, top strategic recommendations,

double down on data backed career

354

insights, AB test thumbnails.

355

We're already doing that.

356

I like this idea.

357

Create a quick answer series.

358

I like that.

359

Front load hooks in the first 30 seconds.

360

I mean, I try, but it's harder

than it, I guess, than it looks

361

and leverage your podcast.

362

More interview content

drives disproportional.

363

Watch time in the Data

Career podcast playlist.

364

Use your strongest pay, consider

promoting it more actively

365

in your recommended videos.

366

Okay.

367

Interesting.

368

Um.

369

So this is like all the analysis it did.

370

I don't even know.

371

Okay.

372

It did extract all of those ZIP files and

to get all those separate CSVs, and for

373

each one of these CSVs, it usually has the

table data, the totals and the chart data.

374

Okay.

375

It kind of like crunched a bunch of these

numbers, but I didn't give much direction.

376

And one thing I actually really wanted

to actually see is like a dashboard.

377

So I'm gonna actually ask

it to create a dashboard.

378

Can you create a dashboard that

monitors these metrics and shows me.

379

The key things that I should be looking

at as a YouTube content creator.

380

Once again, I'm leaving it pretty.

381

Open-ended in general because

I wanna see what it does.

382

I'm gonna go ahead and click go.

383

And you can see it's currently puzzling.

384

It's currently thinking it's

booping, it's gonna try to

385

create some sort of a dashboard.

386

Now, Claude Code really likes JavaScripts,

so my guess is it's going to create

387

some sort of a JavaScript dashboard.

388

Um, these JavaScript

dashboards are usually probably

389

going to be done in React.

390

React is a JavaScript library

that's really good for like.

391

Creating websites basically,

and in turn data visualizations.

392

Um, let's see what it's thinking.

393

So it's entered planning mode.

394

Whoa.

395

Okay, here's the plan.

396

I guess it already planned.

397

So it said, uh, Avery's exported

these CSVs, and this is the data.

398

We're gonna build a single

self-contained HT mal file with

399

embedded CSS and JavaScript.

400

Use chart js for the charts, parse all CSV

data and embed it directly as JavaScript

401

objects so the dashboard opens instantly

to the browser with no server needed.

402

It's gonna have KPI

cards, views over time.

403

Top 15 videos.

404

Traffic sources, audience, age,

geography, new versus returning viewers,

405

subscriber versus non-subscriber.

406

Top playlists and content

performance scatter.

407

Interesting style will be YouTube

studio inspired dark theme.

408

And, uh, okay, let's go ahead and

approve that and see how it does.

409

Now, just for reference, so most of you,

I'm assuming, don't have domain knowledge

410

of what YouTube studio looks like.

411

Um, this is kind of what YouTube

Studio looks like right now.

412

Um, it has my latest video, which was

she became a data analyst in 67 days.

413

Um, how many views that has,

what's the click through rate?

414

What's the average view duration?

415

My current subscribers, which is 65,000

and a summary of the last 28 days.

416

Um, it's not very graphics based.

417

It's not very chart based.

418

Like, I don't think there's

any charts on this page.

419

Right.

420

Um, there's like a lot of cards

with like their own like almost

421

ads for different YouTube things.

422

Um, if you go to analytics, this

one's like a little bit more of

423

what we're trying to look for.

424

Um, this is basically what would

be beating on the overview page.

425

You basically have your views

over the last 28 days, your watch

426

time, your subscribers, and your

revenue of the last 28 days.

427

You have a little realtime

counter for your subscribers and

428

your views in the last 48 hours.

429

And then you have your top content

in this period, which is interesting.

430

My, my shorts, there's a few of my

shorts who actually, that actually

431

do really well, which is interesting.

432

So this is kind of what we have to beat.

433

Um, it's very tab based,

which I kind of don't love.

434

It'd be nice if you could

just like customize your own

435

version of this, I guess.

436

It has your audience channels, your

audience watches, those types of things.

437

So we'll see if cloud code can kind

of beat this and what it looks like.

438

Uh, right now you can kind of see that

it's just extracting and preparing

439

all the CSV datas for embedding.

440

And then the next step will be

to build the H TM L dashboard

441

and then verify dashboard

renders correctly in the browser.

442

It's been working for about.

443

Three minutes and it's still

thinking we'll go ahead and

444

fast forward until Claude tells

me something more interesting.

445

Right.

446

Alright.

447

I just finished the dashboard here

and, uh, it has the KPI cards and

448

then the 10 interactive sections.

449

It has this little preview for

me over the right hand side.

450

I'm not.

451

Optimistic.

452

I don't think it's gonna look that good,

but let's go ahead and see how it looks.

453

I'm gonna go back to my analysis

page and then I'm gonna open up

454

the dashboard here separately.

455

Okay.

456

It doesn't look as bad as I thought.

457

Lots of scrolling on the dashboard.

458

Um, but at least it has

a lot of information.

459

So as total views, watch hours,

subscribers gained revenue

460

impressions, view, duration.

461

Um, the daily views versus

the seven day moving average.

462

Peak periods, which is interesting.

463

Top 15 videos by views and the

revenue, the traffic source.

464

So search is 31%.

465

Oh wow.

466

So 31.4

467

versus 31.

468

So I guess search by itself

actually is my number one source.

469

Um, but browse and suggested I feel

like are so similar and that would be.

470

52.

471

Um, and I've always linked these

together, so that's interesting.

472

Okay.

473

Then you do see that this is the

interesting thing where my search

474

average duration is quite low.

475

Um, oh, and I guess this is red.

476

'cause it's low and green if it's good.

477

So that's interesting.

478

So external, okay.

479

Top country views.

480

The age distribution for views and

watch time subscribed versus non

481

subscribed, new versus returning users.

482

And then CTR versus view.

483

Each bubble equals a video.

484

Size is the watch time,

color is the subscriber.

485

Rate impressions, and

then click through rate.

486

So this has a ton of impressions, a ton of

impressions, but low click through rate.

487

But the green.

488

Is lots of subscribers.

489

So what is this one?

490

This one is, Tableau is

easier than you think.

491

Oh, it's a short, so there's

no subscribers coming from it.

492

There's a lot of views and a decent

amount of a click through rate.

493

Okay.

494

Interesting.

495

I'd almost.

496

Remove shorts from this because they

don't really compare top playlist.

497

This is interesting.

498

So SQLs data analyst, this is something

I've been really interested in.

499

This playlist is doing quite well.

500

Uh, average due duration.

501

So shout out to my Kenyans.

502

It looks like you guys

watch my videos the longest.

503

Other than us and India, why aren't

you guys watching the videos longer?

504

Same with Indonesia and Pakistan.

505

Come on guys.

506

Come on.

507

Maybe it's 'cause I talk crazy.

508

I don't know.

509

This dashboard is okay.

510

I don't think it's anything amazing.

511

I mean, obviously I didn't have to make

it and it was quick, so I appreciate that.

512

There's not like any crazy insights in

terms of like, is it better than, than

513

the YouTube, you know, studio mode.

514

Probably not.

515

There's some things I like

maybe more and some things.

516

Uh, I don't like, I mean, I don't love

scrolling on dashboards really that often.

517

But then the other equivalent is

you have to tab it like this, right?

518

So are you gonna tab or

are you gonna scroll?

519

It just, I guess, depends on how

you like your information viewed.

520

I'd rather tab to be honest.

521

Um, because then I can actually like

choose where I'm going to versus if I want

522

to get to, you know, the traffic sources.

523

I have to go by the

top 15 videos by views.

524

So on and so forth.

525

This isn't bad.

526

It's not great.

527

It's not bad.

528

I'm actually just gonna ask Claude.

529

I'm gonna say review the dashboard,

find the pros and cons of your

530

data displayed and the, um, UI and

create a second better version.

531

So I'm basically telling.

532

Claude, you go, Hey, go look at

your dashboard, see what you did.

533

Well see what you didn't do well,

and then recreate it based off

534

of, you know, your findings.

535

Um, this is something I found that's

really interesting is when you have AI

536

grading ai, it often works better if

you give it to like a different model.

537

Like I gave this to, you

know, chat GPT or GPT 5.2

538

or something like that.

539

Or Gemini or something like that,

it might do a little bit better.

540

Um, but I'm actually interested to see how

it does, it looks like it's struggling.

541

It's trying to open the dashboard,

uh, in Chrome and it's not working.

542

Let's see what it's trying to do now.

543

Okay, let's see.

544

It's been trying to open up the dashboard

for quite a bit here and it looks like

545

it's really struggling to, when this

sort of thing happens, it's a big red

546

flag to me because it keeps trying to

do this and eventually it's gonna give

547

up and it's going to just probably

guess what it looks like instead of

548

actually knowing what it looks like.

549

Although it should be able to like

read the HTML that it created.

550

But I get nervous when it like starts

to do these like failures and repeat

551

itself over and over again because

eventually it's gonna give up and

552

it's just gonna make stuff up and I.

553

Done this enough.

554

I've spent hundreds of hours analyzing

data with Claude that I know eventually,

555

uh, it will make something up.

556

Basically, it's like, this is what I think

it looks like, and you have to be really

557

careful because you, unless you're paying

really close attention, you can actually

558

see what's going on in these log files.

559

You might not notice it,

it might not tell you.

560

So that's something I have a red flag.

561

I'm gonna be looking really

closely to see what it says and

562

actually make sure that it's.

563

Talking normal versus this

is gonna make something up.

564

Alright, so after about five minutes,

it did a full review of V one.

565

Some of the V one, uh, cons

were three outta the seven

566

charts are completely broke.

567

Uh, no month over month.

568

Trends missing, computed efficiency

metrics, no upload frequency analysis,

569

post data completely unused, no

content category grouping, all

570

these different things, right?

571

Well, I mean, like it is working.

572

We just saw the graph, right?

573

Like all this.

574

Stuff is working.

575

All of this is here, so I don't

know why it thinks it's broken.

576

The truth is it doesn't actually

see it, it doesn't actually know

577

what's going on, is my guess.

578

Or it's just not being rendered

correctly when it's trying to to

579

view it, it can actually view it.

580

And so it tried to, you know, uh,

it says it's building V two now, but

581

it's not, it's literally stopped.

582

So maybe I'll say, you know,

keep going and we'll see if.

583

It actually does build version two, but

at this point I'm not super optimistic.

584

We'll, we'll see how it does.

585

Okay.

586

After some coaxing, I think it got

this new version of dashboard two here.

587

Let's see how it looks.

588

Okay.

589

Oh, and look, it did go to a tabbing.

590

Oh, it's both tabbing.

591

Wow.

592

It's like it was listening

to our conversation.

593

I was like, I like tabs

more, and I added tabs.

594

This is interesting.

595

Uh, this is like the

number of views on a day.

596

I don't get why this is here.

597

That's interesting.

598

Um, let's see.

599

Subscriber efficiency per 1000 views.

600

Oh see this is actually interesting.

601

So this, this is normalizing it by views.

602

So what video brought in the most

603

And that's really

interesting because like.

604

Yeah, you know, something like

this one, I analyze this many jobs.

605

Like this is getting 37.

606

Let's see, it says, oh yeah,

607

See, that's interesting.

608

I think this is like the most

interesting graph it's created so far.

609

Um, let's see here.

610

This one's interesting.

611

Watch time, duration by source.

612

Um, and then this is the average

duration and this is the watch time.

613

I don't think that's

very useful comparison.

614

Um, yeah.

615

Here's my funnel.

616

I guess this is the number of impressions.

617

This is the number of views.

618

This is watch greater than one minute.

619

This is the subscribers.

620

That's kinda interesting.

621

I like that idea.

622

This is views versus average duration.

623

I don't know.

624

Okay.

625

Overall, like this is, this is just fine.

626

I think this is nothing

amazing, nothing terrible.

627

What I think where it

gets really powerful.

628

Is where instead of me just saying,

analyze this, where me as like a

629

data analyst, me as a domain expert

come in and be like, I have things

630

that I actually want you to look at.

631

I have my brain, I know it's important.

632

Help me to do the actual

dirty work of the analysis.

633

So for example, um, one thing that I

think is, is really powerful or would

634

be really interesting to see also what

monthly heat map grid 12 monthly cells.

635

I don't even see where that's at.

636

Did you guys see a heat map?

637

Am I blind?

638

Uh, is it like in the old version?

639

If I refresh the old version,

I don't see it there either.

640

So I'm actually gonna call

Claude out real quick for that.

641

I'm gonna say, um, where,

where is the monthly heat map?

642

Oh, it's calling.

643

Oh, okay.

644

It's calling this thing a heat map, which

it's hardly a heat map, but that's fine.

645

Uh, it's calling.

646

This a heat map.

647

I mean, that's a terrible

heat map if I'm being honest.

648

So one thing I think would be really

powerful or be interesting actually

649

for me to see as someone who's, you

know, invested in this data set is

650

something like it tried to make with

this, uh, bubble chart right here.

651

I really like bubble charts 'cause it

can show you a lot of variables at once.

652

Uh, but I don't think this is quite what

I want in terms of the bubble chart.

653

So I'm actually just gonna say,

I'm actually just gonna tell

654

Claude what I'd like to see.

655

Please make just a standalone bubble chart

of the click-through rate on the x axis.

656

The views on the Y axis where each circle

is equal to a video, the size is equal

657

to the number of new subscribers, and

the color is the percentage of those

658

viewers that came by the search traffic.

659

This way, I'm able to see the click

through rate, the views, the number of

660

new subscribers, and the percentage.

661

Coming from search all in one place,

and that'll let me see outliers

662

a little bit better visually.

663

Um, because for me data is really hard

to understand unless I can visually

664

see it and visually understanding

it, uh, makes it really helpful.

665

Okay.

666

Traffic source data is only available at

the channel level by date, not per video.

667

Hmm.

668

That's interesting.

669

Didn't earlier, didn't you tell

me, Claude, that there was a video?

670

I guess it's by aggregate.

671

So it turns out that even though on

YouTube studio, you can see how viewers

672

found this video and see that search

was, you know, 60%, that that is not

673

available in the dataset that we have.

674

There is, that is not included

in the export in YouTube studio.

675

So in order to get that data, we'd

have to like take a screenshot,

676

uh, or, or jot down these numbers

right here, or use the YouTube API

677

and that's for a another video.

678

So we're not gonna do that today.

679

So instead we're gonna just go ahead and

kind of create a similar bubble chart.

680

Um, but instead of the color

being the percent of search.

681

Will make it the view time,

the average view duration.

682

So it's really not that much different

than this chart, to be perfectly honest.

683

I would've liked to have

been a little bit different.

684

Um, but I guess we

didn't give it the data.

685

However, I did notice on the first version

right here, it gave like this little

686

optimization down here where you have

high impressions, low CTR, which basically

687

it says thumbnails and tidal problem.

688

Top left quadrant.

689

While high impressions,

low CTR are down here.

690

It's not the top left quadrant right here

and it, but that does mean that there I

691

could improve the title and thumbnail.

692

So I just got the quadrant area wrong.

693

High CTR, low views.

694

Well, we don't have views on

this chart anywhere actually.

695

So high CTR, low views.

696

It doesn't even make sense.

697

Right.

698

It says it's the bottom right,

but that doesn't make sense.

699

Uh, videos in the top right

are your proven winners.

700

I don't have any with high

impressions and high clickthroughs,

701

so I have no winners, I guess.

702

Uh, anyways, let's go ahead and

say, okay, instead of doing the

703

percent by search traffic, make

it the average video duration.

704

Also, please exclude.

705

All shorts, or I guess rather

make two separate charts, one for

706

shorts and one for longer videos.

707

Because previously it

had put those together.

708

Right.

709

And that's where you, I mean,

this is a short right here.

710

This is a short right here.

711

And this is a short right here.

712

So basically all of the red.

713

Uh, dots on this page were shorts, so

it doesn't really make sense to have,

714

you know, shorts and long videos on the

same page because they're quite different

715

products and quite different audiences and

quite different purposes, to be honest.

716

Okay, I'm gonna hit allow and

let's see what it creates here.

717

So finally, after about seven

minutes, uh, I think it finished.

718

Um, I did notice there was

a few funny things going on.

719

Like for instance, there

was some issues with the.

720

Number of data we were trying to look at.

721

It looks like, like in terms of context

windows, that makes me a little bit

722

nervous and it was having a hard

time actually screenshotting them.

723

Um, because there is a lot of, uh,

different, uh, bubbles going on, a

724

lot of different data being displayed.

725

Um, let's see how it went.

726

Okay, so here is our click through versus

bubble chart, and we have it for shorts

727

and we have it for long form views.

728

So.

729

Um, this is interesting to me.

730

Um, you obviously have like, almost,

uh, I forgot the, what this is

731

called, but like this like shape

where it's like l almost, right,

732

which just goes down and then, right.

733

And then it looks like we just

have some huge outliers and shorts.

734

Um, this interest, this is

interesting to me 'cause now.

735

I can like, interpret this data.

736

In fact, let me ask Claude

how it interprets this data.

737

Um, okay.

738

How do you interpret this data?

739

What action should I take?

740

Uh, we'll see what it says

while I, I give you my thoughts.

741

So what, what is the size of the

bubble that's number of subs?

742

So even though, like I was saying

earlier, even though these, these

743

have huge click through, right?

744

They have low subs, um, this one's

the closest one, so data analyst,

745

mock interview, we're still getting

a decent amount of subscribers, but

746

not even really then we're, we don't

get meaningful number of subscribers

747

till about this video right here.

748

Um, and then these are obviously where the

subscribers are, are quite substantial.

749

Uh, the color is the video duration.

750

Um, I don't think I asked

for the that, did I?

751

If I did, I didn't mean

to, what did I say up here?

752

I said that the, oh yeah.

753

I make it the average video duration.

754

I meant a VD uh, wait, that is.

755

Average view, duration,

not video duration.

756

Ah, that's my fault for, for

saying that, but eh, okay.

757

So this video right

here, this is 45 minutes.

758

Okay.

759

Yeah, that makes sense.

760

Okay.

761

So, um, let's see.

762

So the color's basically meaningless.

763

I mean, I guess we can say that

the, the interviews, which are

764

usually the longer videos, um.

765

Don't get a ton of views.

766

I guess that's one thing we can look

at it and none of them get a particular

767

large amount of sub subscribers.

768

Okay.

769

Um, let's look at shorts.

770

So we have Tableau, remote Data,

job, Tableau, Harvard Saturated data

771

sets, Google Analytics certificate.

772

And what's the difference between these?

773

Okay.

774

So it's almost like we have like a low

line, a little bit higher line, and then

775

like these three outliers, uh, over here.

776

So I think I could just basically

take anything that's above.

777

This like bottom line and try

to make more shorts like that.

778

That would be kind of my takeaway.

779

And pay attention to the titles on these.

780

Um, we're missing such a key metric.

781

The being able to know like what

the traffic's from is so important.

782

So we'd wanna try to get

that data for the future.

783

But since we don't have it

right now, that's my takeaways.

784

I would just try to make more

videos like these titles in the

785

shorts and on this one over here.

786

I mean, really the important thing

is anything I would say that's like.

787

In this circle, uh, right here, we should

probably include, try to include the

788

click through rate on most of these.

789

See if we could improve it.

790

Um, and on these.

791

So that's my takeaway.

792

Let's see what Claude says.

793

Claude is saying, um, your

biggest videos have the worst CTR.

794

Yeah, we've known about that.

795

Higher CTR is good, obviously.

796

Um, not a whole lot there.

797

Shorts two massive outliers.

798

We talked about that.

799

Shorts don't convert to subscribers.

800

We talked about that to.

801

Yeah, I don't think this is

necessarily very meaningful, but

802

I didn't really come up with that

much better analysis on my end.

803

I think really, in order for me to

get really meaningful data outta this,

804

we need to have the traffic source

involved, so hopefully that helps.

805

You see how I personally

use Claude Code as a.

806

Helpful tool as a data analyst,

it makes my work so much faster.

807

Doing all this previously would've taken

me so much time to get through everything.

808

It does a great job of

creating graphs for me.

809

It does a great job of coming up with

some sort of suggestions or some sort

810

of actual analysis and, uh, insights.

811

That being said.

812

I still need to prompt it.

813

I still need to ask it what to do.

814

I need to, you know, obviously be a domain

expert to try to know what all this stuff

815

means and to ask it the right question.

816

So I don't really foresee it

replacing any data analyst.

817

I kind of just see it being as

the new tool for data analysts to

818

actually use to do their analysis.

819

But let me know what you think

in the comments down below.

820

I appreciate you guys watching

or listening, and I'll see

821

you in the next episode.