207: Watch Me Do a Data Analyst Project in Minutes With Claude Code

A lot of aspiring data analysts ask me, is AI going to take our jobs? Are we cooked? And my honest answer is, let's watch and find out. So I did that exactly in this episode. I gave Claude code a real data analyst project, the kind of analysis that I used to spend hours on, and I just kind of let a do its thing. Here's what happens. So here's the project. I want to get some insights on my YouTube channel. What videos are doing well and why they're doing well, so I can try to make better performing videos in the future and more helpful videos for you guys who are watching, or the you guys who are listening. So what I did is I opened up YouTube creator studio and filtered by the last 365 days. And got different things like the content data, the traffic source data, the geography data, and I just exported all those as CSVs straight to this folder here. And so you can see I have a content geography, new and returning viewers, playlist posts, subscription status, traffic source, and viewer age, zip folders right here on my desktop. And basically I wanna see what's going well with my channel and what's not doing so well with my channel. If I were doing this analysis myself or giving it to another fellow data analyst, I would just kind of hope that they'd go through all the data, explore it, and generate some meaningful insights. If they had prior domain knowledge about YouTube videos, they might know to look through, click-through rates on the thumbnails and average view durations of certain videos. But if I gave it to just a normal data analyst, they probably wouldn't have that domain knowledge of YouTube, and so they wouldn't know where to look necessarily. In this case, I'm not going to use my domain knowledge to give Claude any hints. I just wanna see what it comes up with on its own. So let's go ahead and open up Claude. So if you haven't checked out Claude yet, I highly recommend it. It's produced by a company called Anthropic that's very similar competitor to OpenAI, and Claude is basically a competitor to chat GPT. Now, Claude has gained a lot of popularity in the last, I don't know, six months, because one, it can write really well. Two, it can code really well. Andro has basically decided that it doesn't want to be good at everything. It just wants to be good at writing and good at coding. And here's the truth is when you're good at writing and good at coding, you're very powerful. So they have this chat interface that's very similar to Chat GPT, but they also have this thing called Claude Code right here. Now, Claude Code is basically the coding version. Of Claude, it's more designated for getting tasks done, specifically coding tasks. And you can use the CLI, which is basically the command line interface. Basically, that would look like you opening up your command prompt and then typing in Claude right here. And then boom, Claude Code pops up here. So you can have like a command line terminal version of Claude Code. Or you can just use the desktop app, which has Claude Code right here. Now, personally, I like using the desktop app because terminals and command lines still make me a little bit scared, and it's just harder to know what's going on. So I'll be using the desktop version of Claude Code. Now, I will say that I pay a hundred dollars a month for Claude's Max subscription plan. Basically, that means I have the very powerful version of Claude and I can use their most powerful model. Opus 4.6, which is basically just their smartest, most sophisticated, best model that they have. They've also recently updated it to have a 1 million context window, which is very powerful. Basically, it allows you to have a lot of context, which can be good if you have a lot of data. Now, if you were to change models from Opus to something like sonnet, this is their less powerful model. Now, sauna is still really good, but if you were to change it to Haiku, which is really just their weakest model. It probably wouldn't do as well. So depending on what model you're using, the performance will definitely change. The results will definitely change. So I'm just gonna use the most powerful model just to see what we're working with here. Now, recently when I'm using AI tools, uh, I have really gotten into dictation, and the reason being is I kind of suck at writing. I kind of suck at typing, and if I can just brain dump my thoughts verbally, I find that to be a lot more effective than me filtering my thoughts via my typing with my fingers. So I'll just go ahead and click the record button right here. In this folder, you'll find a set of zip folders that contains the analytics for my YouTube channel directly exported from YouTube Studio. I'd like you to analyze these data sets and provide me meaningful, actionable insights on what I could be doing better on my YouTube channel and what I'm doing well. And that's the prompt. I'm literally going to give Claude, and I'm just gonna press this go button and see what it comes up with. It's gonna ask if it trusts this workspace, and it indeed does. And you'll notice that I have that folder right here. It's literally attached to this folder right here. And that's the thing is cloud code isn't working in the cloud. It's working here locally on your machine. So literally it has this YouTube data analysis folder, the contents of which are inside. Now, I'm kind of curious. It's gonna ask me a lot of permission things. I'm gonna literally just do, uh, allow always for this project for right now, uh, I'm curious if it's gonna be able to unzip these. It should be able to. Um, but we'll see what it actually does to, to make this go through. So it's just like, Hey, do you wanna make sure that I can run all these commands and I'm gonna let it run all these commands because I, uh, trust it for right now. Now you obviously probably wanna be a little bit more careful than me when allowing these AI agents to do things on your desktop. You probably wanna make sure you know that. What's actually going on. But to be honest, I'm a little bit lazy and I just, I trust her. So far so good. Obviously there's been a lot of issues. I know that there's been one machine learning content creator who like deleted an entire database completely with cloud code, so you, you gotta be careful and make sure you know what it's doing. So right now it basically looks like it created a user memory for me and it's compiling the analysis. It, read all the key files, understand the picture. It's running these bash commands. It looks looking like it's getting the CSV files. And this is, yeah, look, this is what's in the CSV file. You can kinda see one of my titles, how I'd become a data analyst in 2026. Then you're seeing some of the numbers, views on a certain date or something like that, and you can see that it did that with, um, post data, the content data, the playlist data, so on and so forth. So you can kind of see. What it's thinking about, let's see what it's doing right now. Um, it, it just created like a, a file, a memory file for me about, you know, who I am and what this project is about. And it says Avery Smith is a, oh, okay. Let's see what it, it came up with some stuff. I don't know what it's doing. Ah, okay. Here's my full analysis of your YouTube channel. Over the past year, we've had 1.7 million views, a hundred thousand watch hours, 29,000 subscribers gained about $8,000 in ad revenue, 30 million impressions, a click through rate of 3% average view, duration of. Four minutes and 442 videos in the dataset. What we're doing, well, massive reach with how to become a data analyst content. Your top two videos alone account for 22% of the total, um, which is how I'd become a data analyst in 2026 and how I'd become a data data analyst in 2026. I basically just changed. The 2025 video to 2026. Uh, let's editor, let's, let's rule that out. If you're wondering why those video titles are the same, it's because one of 'em is actually for 2025, not for 2026, but I tweaked the title for search purposes, that it would be normal. That way people could actually find the video still and be interested in watching it. 'cause although some things have changed from 20 25, 20 26, a lot remains the same. This also drove 12,000 subscribers, which is 41% of my subscribers gained two strong subscriber conversion on career advice video. Videos like I analyzed 2,893 Data Analyst jobs don't miss these six entry data levels jobs, and nine huge lies about becoming Data analysts have excellent sub to view ratios. Your audience converts to subscribers when you deliver data backed actionable career insights. All right, let's get to know Number three. Excellent watch time on long form interview content, your podcast slash interview video. So I, I like that it. Figured out that I have like podcasty style videos and interview style videos versus just like YouTube videos. I think that's really interesting that it already gleaned that 'cause I didn't tell it that. Right. Um, let's see How to earn 250,000 with multiple remote data jobs. 13,000 views, but 2000 hours of watch time. Data analyst mock interview, 25,000 views, 2000 uh, hours of watch time, and an 8% click-through rate. Wow. See, that's really interesting. Playlist driven watch time is strong in your data career. Uh, podcast playlist alone drives 658 hours of playlist watch time. It's just listing a bunch of things. I don't know. Some of it's actionable. Uh, we'll see if it suggests things. Um, audience demographics, basically 25 to 44. Mostly us, India, uk, and Canada. Um. Browse features and suggested videos are your top engagement sources. By watch time, this actually means YouTube is actively recommending your content. The algorithm likes you. Oh, I'm glad to hear that. Sometimes it feels like it doesn't. Alright, what needs improvement? My CTR is below average. YouTube considers four to 5% healthy CTR. You're at 3% overall. Some of your biggest videos in the 2.4, 2.8%, this means your thumbnails and titles are underperforming relative to your impressions. Your highest. CTR videos give you clues about what works. Okay. Then it says How to export Tableau dashboard to image 15%. So this is interesting because me having those domain knowledge and actually like being a YouTube content creator and knowing YouTube, I know that this is bad advice because let's, let's go. In fact, let's go to how to exports Tableau dashboard to image. And if you actually just go to Google and you search that I'm gonna be the first video that pops up. Right, and so when you're ranking number one on Google. You're going to have a high click through rate because most of this video's traffic comes via search, not via browse or suggested videos. So with search, you're like already know what you really want, right? You're like, this is what I want. And when someone's like, Hey, here you go, you're gonna click on it. So that's why you have a high. Click through rate versus if you're just like on, you know, YouTube in general, just on the like, uh, homepage, right? Mm-hmm. Right here, it's like, I don't know what I want. Do I wanna watch the, this is a new chapter, or how I built a lead magnet or, um, what is Databricks? Those types of things. Right? Oh, editor, let's redo that because I don't want 'em to see that I'm watching this video. When you open up YouTube, it's like. What do you wanna watch? Do I wanna watch you know, data with Barr right now? Do I wanna watch Mark Lowe? Do I watch Tom Scott? What am I in the mood for? Maybe I'm gonna play this Burger Life game. There's just a lot more distractions and earning the click here is a lot harder to do than if you're like, you know. How to export to Tableau dashboard to image. Oh, okay. I'm number one. Click on Google for, or YouTube for this click. Like, that's the reason why that click through rate is high is because it's mostly search base versus browse or the homepage. Um, same with these videos. These are all going to be, um, higher search videos. This one's not, which is interesting. Um, I'll admit, becoming a data analysis is sustainable right now. Um, so that ones may be interesting to look into. Titles with specificity, controversy, or clear utility. Get clicked more. You're brought how to become a data analyst. Titles get massive impressions, but lower CTR consider AB testing thumbnails on your top impression videos, even with a 0.5% improvement. That's 30,000 more views. View your retention is short. Yes, I agree. Let's see. Um, action. Front load your value, your short form content isn't driving meaningful engagement. That's why we stopped posting short form, uh, engagement. Really, to be honest, 89% of views come from non-subscribers. Let's see your attorney viewers have a 3.7 click through rate versus a 2.81 for new viewers. And watch a minute, 20 longer, add stronger CTAs for subscribing mid video, not just the end. Consider a subscriber CTA in the first two minutes. Okay? If you guys are watching right now, hit subscribe so that way you can make Claude and me happy. Right? And that's, I'm doing my CTA right here. Hit subscribe right now. Um, okay. YouTube search is high volume, but low engagement brings in a lot of YouTube search. Your number one source by view count, but it doesn't have long view duration. I don't even know if that's true. I'm gonna go back to the actual raw data here, and I'm just gonna go to traffic source. Oh, wow. See, I don't think I knew that that search, which is this. Now see, oh, I don't know. Search is this, um, blue bar right here, right? And it's saying it's my number one traffic source over the last 365 days. And I don't know if it is browse, is this green? And it looks like it's above blue most of the time, except for, um, a little bit of a period last year. I don't know. That's really interesting. I don't know if blue is always higher. It's more steady, which is really nice. At suggested videos and browse videos are very similar. So if you take green plus yellow, that's always gonna be bigger than blue. But I guess it's not combining those, it doesn't know, once again, it's missing that domain knowledge. Browse and suggested are very similar style type videos because once again, you're, you're having to earn the click in those. Um, maybe, maybe I'm over assuming how similar they are, but from my domain experience, these two are very similar. Okay, interesting. Action search viewers are looking for quick answers. Consider creating dedicated answer videos. Under five minutes optimize for search while keeping your longer career advice for browse. Interesting. Your Tableau export video. Of a 15.64% CTR proves this model. Well, okay, it proves that we have a high click-through rate, but do I get anything from that video? I'd argue not really. Let's open it up and look at the data a little bit more closely. Okay. I have that video popped up. As you can see, it has 37, almost 38,000 views. 473, uh, watch hours. It's only gotten me $135 and 73 subscriber. So although this video is getting a lot of views, like how am I really benefiting from it? I've made, let's see, I guess, how many dollars a day do I make? 135. I make like 10 cents a day from this video, right? 1 36 divided by 14, 16 days. Yeah, I make 9 cents a day with this video, um, which isn't nothing, I'm grateful for it, but like, that's not really the point. Like even though this has an unreal, uh, click through rate, like this, this click through rate is really, really high, right? Like at 13%, I guess, since published. When you compare it to a video like this, how to become a data analyst in 2026, like, look at this. So many more subscribers, so much more money, you know? And even though our click through rate. Is really low. It's probably like 2% for this video. Yeah, look at the click-through rate's only 2.8%. Even though the click-through rate kind of stinks, like the overall number of subscribers and revenue is way better. So I don't know why it's kind of fixed on, like the idea that. Clickthrough rate is everything. Now clickthrough rate's really important, but just 'cause your clickthrough rate is high doesn't mean that it's good for you as a YouTube creator. But I do kinda like the idea of creating dedicated answer videos under five minutes for optimized search. That is an interesting concept. India viewers have low engagement shorts aren't working. Yep. Uh, either invest seriously in short to the clear strategy or stop spending time there. And that's unfortunately. Uh, what I've done is we just haven't done any shorts. Now, we might start doing shorts in a little bit, but um, we'll have to see. End screens only drove 2,800 views and video cards, 1592 views. That's a very small amount, so yeah, you're right. We don't use end screens nearly as well as we probably should, which is unfortunate. Okay, top strategic recommendations, double down on data backed career insights, AB test thumbnails. We're already doing that. I like this idea. Create a quick answer series. I like that. Front load hooks in the first 30 seconds. I mean, I try, but it's harder than it, I guess, than it looks and leverage your podcast. More interview content drives disproportional. Watch time in the Data Career podcast playlist. Use your strongest pay, consider promoting it more actively in your recommended videos. Okay. Interesting. Um. So this is like all the analysis it did. I don't even know. Okay. It did extract all of those ZIP files and to get all those separate CSVs, and for each one of these CSVs, it usually has the table data, the totals and the chart data. Okay. It kind of like crunched a bunch of these numbers, but I didn't give much direction. And one thing I actually really wanted to actually see is like a dashboard. So I'm gonna actually ask it to create a dashboard. Can you create a dashboard that monitors these metrics and shows me. The key things that I should be looking at as a YouTube content creator. Once again, I'm leaving it pretty. Open-ended in general because I wanna see what it does. I'm gonna go ahead and click go. And you can see it's currently puzzling. It's currently thinking it's booping, it's gonna try to create some sort of a dashboard. Now, Claude Code really likes JavaScripts, so my guess is it's going to create some sort of a JavaScript dashboard. Um, these JavaScript dashboards are usually probably going to be done in React. React is a JavaScript library that's really good for like. Creating websites basically, and in turn data visualizations. Um, let's see what it's thinking. So it's entered planning mode. Whoa. Okay, here's the plan. I guess it already planned. So it said, uh, Avery's exported these CSVs, and this is the data. We're gonna build a single self-contained HT mal file with embedded CSS and JavaScript. Use chart js for the charts, parse all CSV data and embed it directly as JavaScript objects so the dashboard opens instantly to the browser with no server needed. It's gonna have KPI cards, views over time. Top 15 videos. Traffic sources, audience, age, geography, new versus returning viewers, subscriber versus non-subscriber. Top playlists and content performance scatter. Interesting style will be YouTube studio inspired dark theme. And, uh, okay, let's go ahead and approve that and see how it does. Now, just for reference, so most of you, I'm assuming, don't have domain knowledge of what YouTube studio looks like. Um, this is kind of what YouTube Studio looks like right now. Um, it has my latest video, which was she became a data analyst in 67 days. Um, how many views that has, what's the click through rate? What's the average view duration? My current subscribers, which is 65,000 and a summary of the last 28 days. Um, it's not very graphics based. It's not very chart based. Like, I don't think there's any charts on this page. Right. Um, there's like a lot of cards with like their own like almost ads for different YouTube things. Um, if you go to analytics, this one's like a little bit more of what we're trying to look for. Um, this is basically what would be beating on the overview page. You basically have your views over the last 28 days, your watch time, your subscribers, and your revenue of the last 28 days. You have a little realtime counter for your subscribers and your views in the last 48 hours. And then you have your top content in this period, which is interesting. My, my shorts, there's a few of my shorts who actually, that actually do really well, which is interesting. So this is kind of what we have to beat. Um, it's very tab based, which I kind of don't love. It'd be nice if you could just like customize your own version of this, I guess. It has your audience channels, your audience watches, those types of things. So we'll see if cloud code can kind of beat this and what it looks like. Uh, right now you can kind of see that it's just extracting and preparing all the CSV datas for embedding. And then the next step will be to build the H TM L dashboard and then verify dashboard renders correctly in the browser. It's been working for about. Three minutes and it's still thinking we'll go ahead and fast forward until Claude tells me something more interesting. Right. Alright. I just finished the dashboard here and, uh, it has the KPI cards and then the 10 interactive sections. It has this little preview for me over the right hand side. I'm not. Optimistic. I don't think it's gonna look that good, but let's go ahead and see how it looks. I'm gonna go back to my analysis page and then I'm gonna open up the dashboard here separately. Okay. It doesn't look as bad as I thought. Lots of scrolling on the dashboard. Um, but at least it has a lot of information. So as total views, watch hours, subscribers gained revenue impressions, view, duration. Um, the daily views versus the seven day moving average. Peak periods, which is interesting. Top 15 videos by views and the revenue, the traffic source. So search is 31%. Oh wow. So 31.4 versus 31. So I guess search by itself actually is my number one source. Um, but browse and suggested I feel like are so similar and that would be. 52. Um, and I've always linked these together, so that's interesting. Okay. Then you do see that this is the interesting thing where my search average duration is quite low. Um, oh, and I guess this is red. 'cause it's low and green if it's good. So that's interesting. So external, okay. Top country views. The age distribution for views and watch time subscribed versus non subscribed, new versus returning users. And then CTR versus view. Each bubble equals a video. Size is the watch time, color is the subscriber. Rate impressions, and then click through rate. So this has a ton of impressions, a ton of impressions, but low click through rate. But the green. Is lots of subscribers. So what is this one? This one is, Tableau is easier than you think. Oh, it's a short, so there's no subscribers coming from it. There's a lot of views and a decent amount of a click through rate. Okay. Interesting. I'd almost. Remove shorts from this because they don't really compare top playlist. This is interesting. So SQLs data analyst, this is something I've been really interested in. This playlist is doing quite well. Uh, average due duration. So shout out to my Kenyans. It looks like you guys watch my videos the longest. Other than us and India, why aren't you guys watching the videos longer? Same with Indonesia and Pakistan. Come on guys. Come on. Maybe it's 'cause I talk crazy. I don't know. This dashboard is okay. I don't think it's anything amazing. I mean, obviously I didn't have to make it and it was quick, so I appreciate that. There's not like any crazy insights in terms of like, is it better than, than the YouTube, you know, studio mode. Probably not. There's some things I like maybe more and some things. Uh, I don't like, I mean, I don't love scrolling on dashboards really that often. But then the other equivalent is you have to tab it like this, right? So are you gonna tab or are you gonna scroll? It just, I guess, depends on how you like your information viewed. I'd rather tab to be honest. Um, because then I can actually like choose where I'm going to versus if I want to get to, you know, the traffic sources. I have to go by the top 15 videos by views. So on and so forth. This isn't bad. It's not great. It's not bad. I'm actually just gonna ask Claude. I'm gonna say review the dashboard, find the pros and cons of your data displayed and the, um, UI and create a second better version. So I'm basically telling. Claude, you go, Hey, go look at your dashboard, see what you did. Well see what you didn't do well, and then recreate it based off of, you know, your findings. Um, this is something I found that's really interesting is when you have AI grading ai, it often works better if you give it to like a different model. Like I gave this to, you know, chat GPT or GPT 5.2 or something like that. Or Gemini or something like that, it might do a little bit better. Um, but I'm actually interested to see how it does, it looks like it's struggling. It's trying to open the dashboard, uh, in Chrome and it's not working. Let's see what it's trying to do now. Okay, let's see. It's been trying to open up the dashboard for quite a bit here and it looks like it's really struggling to, when this sort of thing happens, it's a big red flag to me because it keeps trying to do this and eventually it's gonna give up and it's going to just probably guess what it looks like instead of actually knowing what it looks like. Although it should be able to like read the HTML that it created. But I get nervous when it like starts to do these like failures and repeat itself over and over again because eventually it's gonna give up and it's just gonna make stuff up and I. Done this enough. I've spent hundreds of hours analyzing data with Claude that I know eventually, uh, it will make something up. Basically, it's like, this is what I think it looks like, and you have to be really careful because you, unless you're paying really close attention, you can actually see what's going on in these log files. You might not notice it, it might not tell you. So that's something I have a red flag. I'm gonna be looking really closely to see what it says and actually make sure that it's. Talking normal versus this is gonna make something up. Alright, so after about five minutes, it did a full review of V one. Some of the V one, uh, cons were three outta the seven charts are completely broke. Uh, no month over month. Trends missing, computed efficiency metrics, no upload frequency analysis, post data completely unused, no content category grouping, all these different things, right? Well, I mean, like it is working. We just saw the graph, right? Like all this. Stuff is working. All of this is here, so I don't know why it thinks it's broken. The truth is it doesn't actually see it, it doesn't actually know what's going on, is my guess. Or it's just not being rendered correctly when it's trying to to view it, it can actually view it. And so it tried to, you know, uh, it says it's building V two now, but it's not, it's literally stopped. So maybe I'll say, you know, keep going and we'll see if. It actually does build version two, but at this point I'm not super optimistic. We'll, we'll see how it does. Okay. After some coaxing, I think it got this new version of dashboard two here. Let's see how it looks. Okay. Oh, and look, it did go to a tabbing. Oh, it's both tabbing. Wow. It's like it was listening to our conversation. I was like, I like tabs more, and I added tabs. This is interesting. Uh, this is like the number of views on a day. I don't get why this is here. That's interesting. Um, let's see. Subscriber efficiency per 1000 views. Oh see this is actually interesting. So this, this is normalizing it by views. So what video brought in the most subscribers per 1000 views it got. And that's really interesting because like. Yeah, you know, something like this one, I analyze this many jobs. Like this is getting 37. Let's see, it says, oh yeah, 37 subscribers per 1000 views. See, that's interesting. I think this is like the most interesting graph it's created so far. Um, let's see here. This one's interesting. Watch time, duration by source. Um, and then this is the average duration and this is the watch time. I don't think that's very useful comparison. Um, yeah. Here's my funnel. I guess this is the number of impressions. This is the number of views. This is watch greater than one minute. This is the subscribers. That's kinda interesting. I like that idea. This is views versus average duration. I don't know. Okay. Overall, like this is, this is just fine. I think this is nothing amazing, nothing terrible. What I think where it gets really powerful. Is where instead of me just saying, analyze this, where me as like a data analyst, me as a domain expert come in and be like, I have things that I actually want you to look at. I have my brain, I know it's important. Help me to do the actual dirty work of the analysis. So for example, um, one thing that I think is, is really powerful or would be really interesting to see also what monthly heat map grid 12 monthly cells. I don't even see where that's at. Did you guys see a heat map? Am I blind? Uh, is it like in the old version? If I refresh the old version, I don't see it there either. So I'm actually gonna call Claude out real quick for that. I'm gonna say, um, where, where is the monthly heat map? Oh, it's calling. Oh, okay. It's calling this thing a heat map, which it's hardly a heat map, but that's fine. Uh, it's calling. This a heat map. I mean, that's a terrible heat map if I'm being honest. So one thing I think would be really powerful or be interesting actually for me to see as someone who's, you know, invested in this data set is something like it tried to make with this, uh, bubble chart right here. I really like bubble charts 'cause it can show you a lot of variables at once. Uh, but I don't think this is quite what I want in terms of the bubble chart. So I'm actually just gonna say, I'm actually just gonna tell Claude what I'd like to see. Please make just a standalone bubble chart of the click-through rate on the x axis. The views on the Y axis where each circle is equal to a video, the size is equal to the number of new subscribers, and the color is the percentage of those viewers that came by the search traffic. This way, I'm able to see the click through rate, the views, the number of new subscribers, and the percentage. Coming from search all in one place, and that'll let me see outliers a little bit better visually. Um, because for me data is really hard to understand unless I can visually see it and visually understanding it, uh, makes it really helpful. Okay. Traffic source data is only available at the channel level by date, not per video. Hmm. That's interesting. Didn't earlier, didn't you tell me, Claude, that there was a video? I guess it's by aggregate. So it turns out that even though on YouTube studio, you can see how viewers found this video and see that search was, you know, 60%, that that is not available in the dataset that we have. There is, that is not included in the export in YouTube studio. So in order to get that data, we'd have to like take a screenshot, uh, or, or jot down these numbers right here, or use the YouTube API and that's for a another video. So we're not gonna do that today. So instead we're gonna just go ahead and kind of create a similar bubble chart. Um, but instead of the color being the percent of search. Will make it the view time, the average view duration. So it's really not that much different than this chart, to be perfectly honest. I would've liked to have been a little bit different. Um, but I guess we didn't give it the data. However, I did notice on the first version right here, it gave like this little optimization down here where you have high impressions, low CTR, which basically it says thumbnails and tidal problem. Top left quadrant. While high impressions, low CTR are down here. It's not the top left quadrant right here and it, but that does mean that there I could improve the title and thumbnail. So I just got the quadrant area wrong. High CTR, low views. Well, we don't have views on this chart anywhere actually. So high CTR, low views. It doesn't even make sense. Right. It says it's the bottom right, but that doesn't make sense. Uh, videos in the top right are your proven winners. I don't have any with high impressions and high clickthroughs, so I have no winners, I guess. Uh, anyways, let's go ahead and say, okay, instead of doing the percent by search traffic, make it the average video duration. Also, please exclude. All shorts, or I guess rather make two separate charts, one for shorts and one for longer videos. Because previously it had put those together. Right. And that's where you, I mean, this is a short right here. This is a short right here. And this is a short right here. So basically all of the red. Uh, dots on this page were shorts, so it doesn't really make sense to have, you know, shorts and long videos on the same page because they're quite different products and quite different audiences and quite different purposes, to be honest. Okay, I'm gonna hit allow and let's see what it creates here. So finally, after about seven minutes, uh, I think it finished. Um, I did notice there was a few funny things going on. Like for instance, there was some issues with the. Number of data we were trying to look at. It looks like, like in terms of context windows, that makes me a little bit nervous and it was having a hard time actually screenshotting them. Um, because there is a lot of, uh, different, uh, bubbles going on, a lot of different data being displayed. Um, let's see how it went. Okay, so here is our click through versus bubble chart, and we have it for shorts and we have it for long form views. So. Um, this is interesting to me. Um, you obviously have like, almost, uh, I forgot the, what this is called, but like this like shape where it's like l almost, right, which just goes down and then, right. And then it looks like we just have some huge outliers and shorts. Um, this interest, this is interesting to me 'cause now. I can like, interpret this data. In fact, let me ask Claude how it interprets this data. Um, okay. How do you interpret this data? What action should I take? Uh, we'll see what it says while I, I give you my thoughts. So what, what is the size of the bubble that's number of subs? So even though, like I was saying earlier, even though these, these have huge click through, right? They have low subs, um, this one's the closest one, so data analyst, mock interview, we're still getting a decent amount of subscribers, but not even really then we're, we don't get meaningful number of subscribers till about this video right here. Um, and then these are obviously where the subscribers are, are quite substantial. Uh, the color is the video duration. Um, I don't think I asked for the that, did I? If I did, I didn't mean to, what did I say up here? I said that the, oh yeah. I make it the average video duration. I meant a VD uh, wait, that is. Average view, duration, not video duration. Ah, that's my fault for, for saying that, but eh, okay. So this video right here, this is 45 minutes. Okay. Yeah, that makes sense. Okay. So, um, let's see. So the color's basically meaningless. I mean, I guess we can say that the, the interviews, which are usually the longer videos, um. Don't get a ton of views. I guess that's one thing we can look at it and none of them get a particular large amount of sub subscribers. Okay. Um, let's look at shorts. So we have Tableau, remote Data, job, Tableau, Harvard Saturated data sets, Google Analytics certificate. And what's the difference between these? Okay. So it's almost like we have like a low line, a little bit higher line, and then like these three outliers, uh, over here. So I think I could just basically take anything that's above. This like bottom line and try to make more shorts like that. That would be kind of my takeaway. And pay attention to the titles on these. Um, we're missing such a key metric. The being able to know like what the traffic's from is so important. So we'd wanna try to get that data for the future. But since we don't have it right now, that's my takeaways. I would just try to make more videos like these titles in the shorts and on this one over here. I mean, really the important thing is anything I would say that's like. In this circle, uh, right here, we should probably include, try to include the click through rate on most of these. See if we could improve it. Um, and on these. So that's my takeaway. Let's see what Claude says. Claude is saying, um, your biggest videos have the worst CTR. Yeah, we've known about that. Higher CTR is good, obviously. Um, not a whole lot there. Shorts two massive outliers. We talked about that. Shorts don't convert to subscribers. We talked about that to. Yeah, I don't think this is necessarily very meaningful, but I didn't really come up with that much better analysis on my end. I think really, in order for me to get really meaningful data outta this, we need to have the traffic source involved, so hopefully that helps. You see how I personally use Claude Code as a. Helpful tool as a data analyst, it makes my work so much faster. Doing all this previously would've taken me so much time to get through everything. It does a great job of creating graphs for me. It does a great job of coming up with some sort of suggestions or some sort of actual analysis and, uh, insights. That being said. I still need to prompt it. I still need to ask it what to do. I need to, you know, obviously be a domain expert to try to know what all this stuff means and to ask it the right question. So I don't really foresee it replacing any data analyst. I kind of just see it being as the new tool for data analysts to actually use to do their analysis. But let me know what you think in the comments down below. I appreciate you guys watching or listening, and I'll see you in the next episode.