Claude in Excel - Better than Copilot? - video Dailymotion

Alex The Analyst

I just randomly found this, but you can get Claude in Excel. Apparently this isn&#039;t new, but I didn&#039;t see any marketing or push for this.  You can find it here: https://marketplace.microsoft.com/en-us/product/saas/wa200009404?tab=overview  ____________________________________________   RESOURCES:  💻Analyst Builder - https://www.analystbuilder.com/  📖Take my Full MySQL Course Here: https://bit.ly/3tqOipr 📖Take my Full Python Course Here: https://bit.ly/48O581R 📖Practice Technical Interview Questions: https://bit.ly/46pDqqL  Coursera Courses: Google Data Analyst Certification: https://coursera.pxf.io/5bBd62 Data Analysis with Python - https://coursera.pxf.io/BXY3Wy IBM Data Analysis Specialization - https://coursera.pxf.io/AoYOdR Tableau Data Visualization - https://coursera.pxf.io/MXYqaN  *Please note I may earn a small commission for any purchase through these links - Thanks for supporting the channel!* ____________________________________________   BECOME A MEMBER -   Want to support the channel? Consider becoming a member! I do Monthly Livestreams and you get some awesome Emoji&#039;s to use in chat and comments!  https://www.youtube.com/channel/UC7cs8q-gJRlGwj4A8OmCmXg/join ______

Transcript

00:00All right, listen, this is not one of my normal videos.

00:02I just saw this and I was like, I'm gonna record it.

00:05I'm not even gonna, you know, practice anything

00:07or get anything set up.

00:08I'm literally just going to hit record and start going.

00:11Now, as a lot of you guys know, I really like ChatGPT.

00:13I've been using it for a long time,

00:15but I've recently switched over to Claude.

00:18I got a Claude Pro subscription.

00:19I've been using Claude Cowork a lot.

00:21I've been using Claude Code a lot and I really like it.

00:24And I was just browsing.

00:26I just randomly came upon this.

00:28And I saw this right here,

00:29which is Claude by Anthropic in Excel.

00:33It immediately jumped out to me

00:35because I saw the rating of 2.4 and I was like,

00:38this isn't good.

00:39This can't be good.

00:41So I gotta try it.

00:42I literally have not even downloaded it yet.

00:44You are gonna do that with me.

00:45I have a sample data set that I just pulled off a Kaggle

00:48and I'm just gonna see what it can do.

00:50And we're gonna see what it can do together.

00:51I'm really excited because this, I just found it.

00:54And so we're gonna try it out together

00:55and maybe it's amazing.

00:57Maybe it deserves that 2.4 rating,

00:59but I'm gonna ask it to do some basic stuff

01:02and we'll see what happens.

01:03So let's come right over here.

01:05Let's go down here to get it now.

01:08All right.

01:08I just went through like a 20 step process

01:10to log into Microsoft to get here,

01:12but we are finally here.

01:14It may be because I'm now on a Mac.

01:16I switched over.

01:17I'm gonna make a video on the difference between PC and Mac

01:19in the future because it's been really eye-opening.

01:22Let's go over here.

01:23We're gonna open in Excel and of course I need to allow it.

01:26Thanks.

01:27Thanks for that.

01:28Now let's open this up

01:30and I'm gonna full mode this bad boy

01:32because I wanna see.

01:33So right here we have Claude right in your notebooks.

01:35Let me go ahead and log in.

01:37And it looks like we're gonna have this option

01:39of Claude right up here.

01:41I am just figuring this out as we go.

01:43This is fascinating.

01:44I'm gonna go ahead and log in.

01:45I just have to like sign my life away.

01:47It said you can have access to this, this, this, this,

01:49and this.

01:49I'm doing this for you guys but normally I would have like,

01:52you know, not given them access to everything.

01:55But now it's giving me a little bit of information here.

01:57It says make a pivot table.

01:59Select yourselves and ask questions about it.

02:00Let's go ahead and hit next.

02:02Build financial models.

02:04This is, I guess you could call this a financial model.

02:07It just looks like a formula to me.

02:09And we're gonna describe exactly what we want

02:11and it's gonna build it for us.

02:13Let's go ahead and hit next.

02:14It's gonna debug and fix errors.

02:16And we can open Claude anytime with control option C.

02:21It does say that this is in beta,

02:22which I 100% guarantee it is with a 2.4 rating,

02:25but we're gonna try it out.

02:26Now that we have this up and running,

02:28I'm gonna come over here

02:29and I'm gonna get this data set that I had pulled up.

02:33All right, in this data set is just

02:35Apple global sales data set.

02:37I got this off Kaggle.

02:39I'm just, we're just gonna go ahead and try it.

02:44Here we'll just ask some questions.

02:46We'll do it exactly like it wanted us to do.

02:48And we're gonna see if it gives us actual outputs

02:50that can actually build us things.

02:52I'll start really simply.

02:54We have some really good data here

02:55and then we'll see what happens.

02:57Now I'm just gonna clean this up a little bit.

02:58You'll have to excuse me.

02:59I don't have everything set up.

03:01So I'm using like my main monitor way over here.

03:03Let's go ahead and just get this format a little better.

03:08Now we have a ton of information here.

03:10We have the sale date, we have the year, quarter, month,

03:12country, region, city, the product, category.

03:15I mean, this is a ton of information.

03:17Let's keep scrolling over just a little bit.

03:20And we have things like return status.

03:22It looks like there's an error on this.

03:23We won't look at that too much.

03:24And then we have the revenue in USD.

03:26So I'm gonna ask it some questions here

03:29and let's just see what happens.

03:32All right, so let's bring up Claude.

03:35First, I'm just gonna have it explain this workbook for us.

03:38It says, walk me through the structure.

03:40That looks good.

03:41We're using Opus 4.6.

03:43We can change that to 4.5 or Sonnet 4.6.

03:46I'm gonna use the latest one.

03:47Let's go ahead and run this.

03:49It looks like it is reading it in properly.

03:52This is really good.

03:54It's giving us a lot of information.

03:56I'm going to expand this window quite a bit

04:00just so you can like really see it.

04:04But it gave us a ton of information.

04:05Let's go at the top.

04:06It says, Apple Global Sales Dataset.

04:10And it says we have some key inputs

04:11like transaction ID and date, A through E.

04:15Then we have the geography, product, pricing, volume,

04:18and currency, and that's all towards the left.

04:20Here we have some calculated fields

04:22like discounted price USD.

04:24I guess that's the logic behind it

04:26as well as revenue USD and revenue local currency.

04:29Then we have some analytical dimensions.

04:31So this is just giving us kind of a breakdown of this.

04:33This isn't anything exciting

04:35because this is super simple for any AI system

04:38even two or years ago.

04:40So this is not anything new or anything exciting.

04:42Now let's give it something hard.

04:44So let's come in here and let's kind of highlight everything.

04:49And I'm gonna say, take this data I've highlighted,

04:54which is all of it, and give me a breakdown

04:58of the products that have made us the most money.

05:06I want it in a visualization.

05:11I need to spell that right.

05:13And a pivot table.

05:16So it's gonna ask us to do,

05:19I'm asking it to do a lot here first off

05:22because I basically just highlighted the table.

05:24I didn't even, I don't know if it even reads this in

05:26just by context.

05:27I'm assuming it does.

05:29But let's go ahead.

05:31It says, Claude wants to create a new analysis sheet.

05:33Let's say, always allow,

05:35just so it can continue to do that for us.

05:38Let's go ahead and get rid of this.

05:40Thanks for the double usage, Claude.

05:42Let's see how it runs.

05:43Now this is a little bit more difficult.

05:45I wouldn't think this is like super difficult,

05:47simply because all it's doing is creating another sheet,

05:49taking the data, putting it into pivot table,

05:52using some pre-built in things.

05:54In just a second, I'm going to,

05:56let's see, it looks like it may have encountered

05:58an issue here.

05:59I'm gonna get it to do some kind of higher level thinking,

06:04not just, hey, build this small thing for me.

06:07I'm gonna ask it to do multiple things at the same time

06:09and we're gonna see how it handles it.

06:11So it went ahead and it created this for us

06:14and that's very good.

06:15And it used the category for the rows.

06:17So we have accessories, AirPods, watch, iPad, iPhone and Mac.

06:22And then we have basically the sum of units sold,

06:26sum of the revenue and the average.

06:29And here's our visual and this looks good.

06:31Let's see if it's doing anything else while we're here.

06:34So it looks like it did it pretty well.

06:37I mean, this is a very basic request, right?

06:39Anybody should be able to get in here

06:41and make a pivot table and make a pivot chart.

06:43But if we click into this,

06:45it's going to change to our pivot chart fields.

06:47And if we come right over here,

06:49it's really a lot of the same information that's in here,

06:52which is basically what I wanted it to do.

06:54And it created this bar chart for us.

06:57And so if we look at this,

06:58it's gonna basically reflect a lot of the same information.

07:01Here is our Mac and that's around $8 million in sales

07:05of this data set.

07:06I just pulled this off the internet.

07:08I don't even know what this data set is,

07:09but it looks like it's summing that data properly.

07:11And then it's also visualizing it properly as well.

07:15The next thing I want to do

07:16is a little bit more subjective, right?

07:18Creating a pivot table, I think is, you know,

07:20fairly easy to do for anybody, any human.

07:24Let's see if it can look at the data

07:26and find issues with it.

07:27I think it's going to be able to identify it,

07:30but I don't know if it's gonna be able to fix it itself.

07:32So let's come in here and let's say,

07:33I want you to analyze each column

07:38and find any issues with the data.

07:42If you find any issues,

07:45I want you to copy the entire data set,

07:51put it in a new tab and clean the data

07:56and then explain what you did.

08:00Let's see if it can do this.

08:01This one is a lot more subjective.

08:03Cleaning data is kind of an art form in some aspects.

08:06And so let's see if it can kind of understand that.

08:09I mean, this especially is obviously messy.

08:11This would be something I would pick up immediately.

08:13If we scroll through here, I haven't looked at everything,

08:16but you know, we may want to clean these NAs up.

08:19Maybe you want to replace that with something.

08:21This is kind of an area that's pretty subjective.

08:23So let's see how it does this.

08:25I'm gonna go ahead and run this.

08:26And again, I'm using 4.6.

08:28This is by far, it's kind of most intensive model right now.

08:32If you want to use a lot less tokens,

08:34because of course you use a lot of tokens with Claude.

08:37If you want to do that,

08:38then go with a previous model like 4.5 or Sonnet 4.6.

08:43Now after this, if it can do this fairly well,

08:46I think I'm gonna be pretty impressed.

08:48I do think that in the future,

08:50a lot of what we do, especially in tools like this,

08:53we'll be using AI, as of right now,

08:56there isn't like a really good option for AI,

09:00don't even get me started on Copilot, just don't.

09:04But we don't have a great option right now

09:07for AI within Excel.

09:10And so who knows, this might be the one that is really good.

09:13I haven't stress tested this, right?

09:16With real data sets, large data sets,

09:18could absolutely break this completely.

09:21Or it may just even be reading in a larger amount of data,

09:24could use a lot of tokens,

09:26and you can't even do one or two tasks

09:28without reaching your limit.

09:29I don't know.

09:30We are on, you know, the frontier with Claude right now,

09:34and we're just trying it out.

09:35So let's see if it's able to do this.

09:37I'm gonna let this run for just a second.

09:39I was rambling, because I'm excited.

09:41But let's let this run for just a second.

09:42Let's see what it does.

09:44All right, it has completed its task.

09:46Let's see how it did.

09:48Now, the first thing that you'll notice is it says,

09:51I've created this cleaned data sheet with 11,500 rows.

09:54I don't even know if that was in our,

09:56how big our original data set was.

09:58Let's go ahead and go all the way to the bottom.

10:00Yeah, so you have 11,500 rows.

10:02So that's a good sign, at least,

10:04that I have the right amount of rows.

10:06Now, if we just scroll back and forth,

10:09I can already tell that there are some things

10:12that are different.

10:14The first thing that I'm noticing,

10:15and this may or may not be intentional,

10:18but we have this sale date right here,

10:20and it actually basically made it a general data type,

10:24instead of something like a short date,

10:26to keep it in that date format.

10:28And date formats are important.

10:30Like the data types are really important

10:31within any database or any Excel type tool.

10:35You are telling different systems,

10:37here's what it can be used for.

10:38And so over here, we had it as a date right here,

10:43but it changed it for some reason to that general.

10:46I don't know why.

10:47I'm gonna keep it like that,

10:48because that's what it thinks is the best.

10:50Now let's go in and see what data quality issues we had.

10:53First, we have this corrupted text encoding,

10:56which we absolutely did.

10:57Let's see what it did to fix this.

10:59So we have this customer age group 45 to 54,

11:03and let's scroll over and let's take a look.

11:06So here's what we had before,

11:08and it looks like it did actually fix this.

11:11So we have 45 to 54, 18 to 24, and let's go back.

11:16And here we have 45 to 54, 18 to 24.

11:19So I think that it correctly got this encoding issue done.

11:23I really think it actually did a pretty good job.

11:25Now, I would add a filter to this.

11:28I'm gonna double check this.

11:30Let's do this while we're here.

11:32Let's see if it got all of them done properly,

11:34and it looks like it did.

11:35So that's a really good sign.

11:37We also have a month mismatch with date.

11:40So this has recalculated the month date from the sale date.

11:43352 were off by one month, 27th showed January,

11:46when the date was in December.

11:48So if we scroll back, we have this sale date,

11:53and then we also have a month.

11:55And what it's claiming is that there was a mismatch

11:57between the day and the month,

12:00when it actually broke it out.

12:01Now, I'm not gonna go back and check,

12:04but that is a real data quality issue where you have mismatching columns, right?

12:09Somebody says they're 75, but their birth date says they're only 73.

12:12That's a data quality issue, and so you'd wanna fix that.

12:15It may have done that, and maybe it needed to convert this to a general data type in order to

12:21do that.

12:21I don't know.

12:22But it's saying it fixed that, so, you know, we're gonna go ahead and give it the benefit of the

12:25doubt.

12:26And we also have inconsistent storage format.

12:29So, removed SSD suffix, so all storage values are unified.

12:34One terabyte of SSD, one terabyte, 256 gigabytes SSD, 256 gigabytes.

12:39Now, let's go back.

12:41This is the last one we're gonna check, and then we'll be done.

12:43But let's go over to storage real quick, right in here, and let's find our storage.

12:49All right, so storage is right here.

12:51Let's open this up, and so we have one terabyte, 128 gigabytes, two terabyte.

12:56So, it looks like it's standardized this.

12:57Let's come back, and let's take a look.

13:00I'm going to come right here.

13:02I'm gonna add a filter, just so we can look at it.

13:06And let's look at our storage.

13:08Here we have one terabyte and one terabyte of SSD.

13:11All right, standardization and kind of standardizing your data is 100% a great data cleaning thing.

13:17I don't think this is correct because it's very possible that the one terabyte is not SSD.

13:24And so, it's making an assumption here that, because this could be, what is it, HDD?

13:28I can't remember before, solid-state drives.

13:31But this is like the old style with the disk.

13:34And it very possibly could be that these are two completely different things.

13:37And so, I could just go in here and I could prompt and say,

13:40hey, these are not the same things, we don't want them to be matched, you know, switch it back,

13:45or this isn't something we want to change, and we can fix that.

13:48But this is where, you know, just having understanding of your data is really important.

13:52It made some assumptions here.

13:54I'm gonna give it, again, the benefit of the doubt, although I do think this is kind of an error

13:58in my opinion.

13:59If I'm scrolling down, we also have missing customer ratings, and they just left it as is.

14:05And at least it's telling us to do this.

14:07Sometimes you want to fill in those values with some default value, or you don't want to populate that in

14:12some way.

14:13But we're keeping it as is, and then it's giving us some other information.

14:17Overall, I don't think I'd give it a 2.4 rating.

14:21As of right now, I would give this a solid 4.

14:25The reason I'm giving it a 4 and not any higher or any lower is because I think that it

14:29did do a really good job.

14:30I mean, Claude, in and of itself, is a great tool for data analysis.

14:34I'll be coming out with lots of videos on how to use Claude Cowork, how to use Claude Code, things

14:39like that.

14:39But integrated in the tool like this, I think it did fairly well.

14:43Here's the only real issue, and this is something I might cover in another video, is using larger data sets.

14:49When you have multiple data sets, when you start joining these data sets together,

14:52you need to start creating custom formulas and things like this.

14:55That's where things can get really tricky, and AI can sometimes have trouble with that.

14:59We saw that with Copilot when it first came out. I mean, Copilot was a mess. Copilot was not good.

15:05But, this is pretty decent. I think this is worth testing and trying out.

15:10And I literally just saw this, you know, randomly, and I was like, I'm filming a video,

15:14and so here I am, just randomly filming a video. I didn't get changed into my stuff or anything.

15:19I am looking at myself, and I feel a little silly that I'm wearing a hat.

15:22But, I was like, I gotta get this. I gotta try it. I'm just gonna do this live and see

15:26what happens.

15:27And, overall, I think it did pretty well. I think that, when it comes to structured data like this,

15:32there's gonna be a lot of use for AI, but it's gonna really come down to, do you know the

15:38data well?

15:38Do you, as a user, understand the context of this data?

15:42And then, later on down the line, it's gonna be really important to create these semantic models of our data,

15:47in order to give it a lot of context to let it know, here's what we need, here's what we

15:52want.

15:52And, that is gonna take, just as an enterprise, a long time to create.

15:57These are not overnight things. We're looking at, you know, three, five, ten years before a lot of adoption

16:02of these types of things in most companies.

16:06But, as just a plug and play for Claude, I think it did well. I think it did good.

16:10I will make more videos like this on tons of different AI tools in the near future,

16:15because I'm diving into so many, and I'm trying to keep up, and it's a lot.

16:19But, it's a lot of fun. And so, if you are interested, try this out.

16:23Create an account, try it out, see if you like it, see if it works for your use case or

16:27not.

16:28Thanks for joining me on this random video that I just decided to record.

16:31I hope it was useful or interesting. Maybe you didn't even know this existed.

16:34I did not. But, it makes perfect sense. And, we'll see more tools like this coming out.

16:39And, we'll see how they go. Thanks, everybody. I'll see you in the next one.

Claude in Excel - Better than Copilot?

Category

Transcript

Comments

Recommended