Create a Marketing Funnel with Excel + Copilot

Hello everyone and welcome to today's session. My name is Ree and I'll be your moderator for today. We're going to get started in a couple of minutes once we hit top of the art. just waiting so everyone has a chance to join. If you haven't done so already, please do register for today's session. There is a link to do so in the chat now. You can also scan the QR code that's on screen. And you can also head over to datacamp. com/weinars where you'll find this session as well as all of our future sessions as well. Uh you may already know this uh if you've attended a uh session of ours recently, but we do have our uh flagship virtual conference coming up next week and that is uh data camp radar where we're going to be uh talking about the intersection of AI and humans working together. So if you haven't done so already, please do register for that as well. Uh I'll be posting a link to do so very shortly and we'll also put up a QR code as well on screen at some point so during the session. So yeah, if you want to do that or you can Google uh data camp radar and you'll uh it'll likely be the first hit. So yeah, make sure you register for this session. Make sure you sign up for uh data camp radar and also check out the other uh campaigns that we've got coming up namely uh we're going to be working on AI powered Python for beginners and we've also got a campaign coming up midappril that is all around anthropic but that will be uh released and uh yeah shared with you very shortly but I'm just giving you a sneak peek on that one. uh back to today. So yeah, please make sure you register. Uh if you've already registered, you will have got uh reminder emails notifying you about the setup info and the data set that we're going to be working with today. If you haven't had a chance to look at those, check out the link in the video description. I'll also be showing a link to the resources document, which will link to the setup info document uh in the chat now. So have a look there. Essentially, what you'll need is a uh access to C-Pilot and be able to use Microsoft uh Excel as well. Um we also have the slides that we're going to be covering today in the resources. So, check those out. But again, if you miss anything, make sure you register for the session and we can send you everything along with the recording as well. If you have any questions at any point throughout the session, let us know in the chat. We are going to be uh running through your questions for the last 10 minutes of the session. So, make sure you stick around for that. Cool. I will repeat these messages once more for anyone that's just joined. So, hello everyone. My name is Ree and I'll be your moderator for today's session. We are just about to get started. Also, next week we are having our virtual uh conference, our flagship virtual conference, Static Camp Radar. So, please do uh register for that one as well. I'll be posting a link to that uh very shortly, but we also have a link to it in the resources document too. Uh as well, if you want to join in with us today, please check out the setup info and data set link in the video description. There's also a link to it in the resources document that I've shared in the chat as well. So, have a look at the links in there. And if you want to join in with us today, then uh yeah, have a look at the setup info. There's going to be uh a little bit of time at the start of the session for you to get that sorted, but hopefully you've done it already. And with that, I think I'll hand over to our host for today's session, Richie. Richie, please take it away. Hi there, data scamps and data champs. This is Richie. Uh I'm very excited for this session. So, uh today we're going to be mixing old technology and new technology. So, Excel has been uh perhaps the most popular data analysis tool for almost three decades now and the co-pilot AI assistant provides a modern way to interact with. So, we've been playing around with both technologies and we're going to be putting together a marketing funnel. So, this tracks how people become customers. It's essentially one of the most common

business intelligence tasks and well essential for every business. So, worth knowing about. Our guest is Laura Videl. She is the senior director of life cycle and retention at data camp. Welcome Laura. Great to have you here. — Thanks for having me. — Absolutely. Uh yeah, I'm very much looking forward to this uh session. So uh Laura manages the customer journey for data camp users. That means taking people from uh acquisition through to retention. So she's a datadriven marketing and communications executive. Uh before her time at data camp, Laura was director of retention at the food waste social impact company too good to go. That's actually sounds quite a fun place to work. I mean, do you get to eat leftovers or a lot? — You do. It's much more appealing than you make it sound, but yes, there's a lot of food. — Uh, nice. Wonderful. Uh, okay. I will let you take it away with uh with teaching people about marketing funnels. — Perfect. I'm assuming you are seeing my screen. Perfect. Excellent. Yeah. So using Excel and Copilot to build a marketing funnel uh you hopefully um got a license to Microsoft. Uh if not there's a trial available for 30 days. Um other than that the enablement um to copilot is necessary for this code along. And um yeah there's a practice file that you can download um to follow along while I'm going through this. Um, and if you do have a data camp uh account already, you're probably familiar with the Excel courses that we have uh available. Otherwise, highly encourage you to register and check them out. What we're going to do today, we going to just quickly describe um the setup. So, how do we enable co-pilot? It's going to be a quick one describing the data. We need to clean it, normalize it, and um then merge it. Basically, as you might have seen in the data, the workbook that I shared, it's two tables. Um, and based on that, we're going to build the marketing funnel and we're going to take two deep dives. One into um the workout, the correlation of workouts with um subscription because we're looking at an workout app here um and a channel analysis. And in the end, we're just going to summarize and ship it. All right, let's hop into um Excel right away. So, as I mentioned before, we have two sheets, revenue data and event data. Um, and for us to enable Copilot from home, you see the little icon already here. And if you do not see this icon here, that means um it is not in agent mode. So, you want to um turn it on. There's this little burger menu that you can click. Maybe I'll just show that really quickly. Here you go. And then you can enable it. There you go. All right. So, the first thing that we're going to do is we're going to turn our data into tables. Um, so just so you know, I got my prompts or at least the majority of them already pre-written just so that you don't have to watch me struggle since I'm the person that voice prompting was invented for. Um, and with turning the data into tables, we just create um more organization for co-pilot. Um, it gives clearer results. highly recommended by Microsoft. So, that looks already pretty good. And then let's just check as well the other sheet. And we're all set. Good. And then just for ourselves, I mean, usually you are the person that kind of pulls the data so you do have a good idea of what's in there. But um just to get everyone up to speed and also potentially correct co-pilot if there are some discrepancies on what it is thinking this is. And it actually should be I'm just asking it to describe the data in the workbook. So while that is going on um I'm already jumping ahead a little bit since you can see the data. Um there are some inconsistencies particularly here in the event date and also in uh writing capitalization of the last touch source. Um, we also have some empty cell values which we need to clean in the next step. But let's see what co-pilot came up with. So, two related data sets appear to track a fitness

wellness subscription service. Um, describes number of rows, columns. It describes the values that are in there. the subscription type, annual, monthly, and the revenue amount. And on G2, we get the revenue data. And last but not least, um, both tables are linked by user ID, so it makes that connection already. Um, it shows what we're actually tracking here. Um, and then it has a data quality note that says it has inconsistent date formats. Um, but what it hasn't pointed out so far is actually the different capitalization of the last touch source. So, as we are cleaning this, the first thing I'm going to do is check for a duplication of data by prompting it to find any entries of the exact same data in all column columns in multiple rows. It seems like it's a little bit of work to crawl through both sheets. And here we go. So, in the first revenue data sheet, there seems to be no issue. But then in the event data we see uh some duplicates 124 to be exact. It gives you some examples. So one thing also is to ideally not take it for granted but actually check that copilot has identified this correctly. So by here it gives you already the rows. So this is a little bit of a random checkup to hop in and see that this is actually correct. What I'm going to do from here on is asking it to create a new sheet called event data duplicates removed. Removing the found duplicates but transferring all the other values in there. This way I'm keeping my raw data intact. And actually while it's doing that I'm just giving this a color so I have a little bit of orientation as we keep on creating new sheets. But basically everything I'm going to do is going to happen on that event data side is going to now be working on the event data duplicates removed sheet just so that whenever something happens uh co-pilot chokes there's a mistake I can co go back to original source and can retrace it. Okay, so it already flagged inconsistencies in that um date range or the data on the event dates. Um I do want it to list all the consistencies knowing that it missed one, but I also haven't prompted it to check for that yet. It is bit of larger job takes a little bit of time. Um but once we have identified that we want to create um a new column basically correcting that data. Here we go. Revenue data raw sheet seems to be tip top. There are no duplicates and also the data seems to be very clean. But then on the event data we see there is an issue with the formatting of the date and then we also have inconsistent casing. Um the event types are consistent class types are consistent campaigns no messing data and key fields and then it ranks those according to the issue. So let's attend to the event date first. So what I'm asking is basically starting with the sheet that we newly created. Again the raw data keeps is completely untouched.

Uh we're creating a new column called event date clean next to the original column. The columns are referenced in brackets and with all the values of the original column formatted as Excel serial dates identical to the event date data in revenue data raw. So I basically just want to give it some more direction to say okay I want to match these or merge these two tables later on. So benefit of getting the same data format in is big. So therefore do the exact same thing. Great. The column is already there. One more thing with how the data is um presented. So we already have some serial numbers in here. Um it makes it extremely difficult obviously to do a random check whether this has been translated properly unless you happen to know the serial numbers um for the 26th of March for example which is why I would advise to change the date formatting later on so you're able to actually do that Check get 100% converted zero errors in conversion. Um like I said ideally will be the judge of that and then we can do some random checkups here. So see on the 20th that all looks good. All right, we got one more which is the last touch source. And similarly we want to create a new column next to last touch source with the prefix clean. I could have spelled it out as well. Um, and then standardizing the values to matter organic including organic search affiliates and Google paid search including all variation of Google ads that is quite — I don't understand. Oh, sorry. That was my Google. Um, so in this case, uh, it might be safer and probably also easier to check. So, in this, uh, worksheet, you have about, I think, 1500 rows or something like that. So, not too bad. But uh if you do have a massive spreadsheet obviously um you might want to break it down one by one just to confirm um that everything has been uh changed according to your prompt and then you also don't have such a massive prompt at once. kit. It's been updated. All the variations of meta on our meta organic affiliates. All the trailing spaces have also been cleared from affiliates. So that should leave us ideally. We can again just quickly check that. Um all meta is now meta. Very nice. Good. Um and similarly these two we wanted to have named Google paid search and that is also consistently the case. So that seems to have worked. Good. So now we have everything cleaned. You might want to insert a column here. And this event value is actually the workout time. And now I'm mixing my work with co-pilot's work. Um, and sometimes it's just quicker to just write it yourself than

to prompt co-pilot. So in this case I just say if this is blank then insert zero and otherwise insert the value from here. And now I have this new column. And I could do the exact same thing for this woman. Um, just want to call class type clean. Um, and we do a similar if function. Uh if this one is blank then we want and I otherwise give me that value. Here we go. Good. Before we move on, just confirming with copilot that there are no more inconsistencies here. So, um, ideally I would just do another attack to ask, are there any more inconsistencies? Hopefully we're actually good to move on. This takes uh significantly more time. Here we go. Um a date mismatch was found uh between revenue and event date. Um the event date clean values and event data do not match the event date values in the revenue data for subscription records. Uh even for the same user ID and subscription time. Um here it shows some discrepancy of the revenue data and the event data some sample mismatches and then it describes the root cause. Um it shows that in the event data contain the biggest text dates and different formatting. uh the conversion formula interpreted the format but some dates may have been in a different format causing the incorrect conversion. So in this case um it recommends the date conversion issue stem from the original ambiguous date formats d um cross reference with the revenue data to identify the correct dates or obtain clarification on the original dates format. So there you have two options. I'm going to go with option one. Yep. Obviously you can ask it in this scenario to um create a sheet with the mismatching dates with the user ID and so on. And you can go back and try to identify where the root cause is in the sense that you go back to the actual source of the data and check. Okay. So while it is doing this

and I believe this is actually something that might be at uh the issue of the raw data that I have delivered. Um the next step that we have is to transfer our clean data into another sheet. Once we have managed to do that. All right. So here and that's what I said earlier. I think that has to do with the raw data that I have delivered. It's not an issue of the formatting. It's actually raw data. So here it created a cross reference of the different matches. Exact match, mismatch. And it even provides a recommendation here. So validate against the original source system if needed. So in this case I'm going to ask it well it already created um an overview of these different mismatches and you can therefore basically uh go in and validate the dates in this case in order to move on with this. This was not a planned inconsistency unfortunately but we're just going to transfer the remaining data. So remove the user ID with mismatching event dates from the table and create in a new sheet called event data, cleaned That way we can at least continue. Good. So, this is our next sheet that we're going to continue working with. Um, and in this sheet, we going to use the data user ID, event date. Um, oh, sorry, I jumped on this one. We want to transfer the right columns into a different sheet. So our clean columns, we're going to create another sheet. event and revenue data and cop copy over the column user ID, event date cleaned, event type, class type, worker time, last touch source cleaned and last touch campaign from and we now renamed it from data duped clean. So creating one more sheet and we already called it event and revenue data. Um

because we want to add basically the data from our revenue data raw sheet from here. So when we're looking at the data here, let's get a little bit small. Apologies for that. When we look at the data here um we actually have subscription data and renewal data um and a subscription type. Whereas in the event data we have an event type that is either annual subscription or monthly subscription. If I find one there you go. Yeah. Um but no renewal events of any sort. So therefore we need to merge this. So now we have our event data dduplicated cleaned all in here. Now we want to merge it using this would be a V lookup actually an X lookup because we want to match based on the user ID the event date we want to match the data from the revenue type the subscription type sorry for that and revenue amount from that raw data revenue raw data sheet to the event type being either annual subscription or monthly subscription. So just to clarify what I've just done here, um all the columns are referenced with brackets and then the values being annual subscription and monthly subscription in uh our sheet. What that means is basically that we're going to get the data from here which is it a renewal? Is it monthly or annually? We get added here. It already added those columns. So, we got the revenue type here, the subscription type, and the revenue amount. Okay. So, in order to again um confirm that this is correct, we would basically look up the user ID, go into our revenue data raw sheet and confirm that this is actually the right uh revenue type, subscription type and revenue amount. Um and then on the next step one more thing in the sheet is to add a column called event type all copying over the data from the column event type but if annual subscription or monthly subscription fill in the data from the column revenue type instead. So basically what I want is I want to replace my annual subscription, my monthly subscription with just the value new subscription since I now have a different column that actually tells me how long that subscription is lasting. And after we do have that, we can actually move on to building the pivot table, which will then uh allow us to build our marketing funnel.

Still not quite there. All right. So, I actually have a question about workflows here. So, — one thing I've noticed like with Copilot is it takes a few minutes thinking often and I don't know how you deal with um these weights. Like how do you change your workflow to cope with this cuz it seems like in order to not just be sat around like for annoyingly short pauses each time, you got to be able to switch to something else. Have you found a way to deal with this? Um, to be honest, I think the best way to do this is probably to think of the next prompt already. Um, because you obviously need to move on after this. Um, other than that, I also think uh just summarizing well usually you don't just build an Excel. Obviously now we're kind of waiting on a point where we don't have any insights that we can review necessarily, but since we had some issues with dates that were unaligned, I probably work on that trying to figure out what where's the issue in the raw data. If I already have some insights, I probably would start building, you know, some documentation, some deck, whatsoever that starts translating this analysis into something presentable. — Okay. So, yeah, I like the idea of like kind of figuring out what the next step is going to be so you can kind of cue stuff up already. Um, I guess the other thing is because there are now like at least three different ways of working with Excel. So you can either write formulas or you can use the point andclick interface. Um and now there's copilot as well. Do you have a sense of like what sort of tasks you want to do with each of the three interfaces? Like when do you choose one or the other? — Um I guess uh to be as I mean personal preference probably to some degree, right? Um I do think most people do have yeah a personal preference what they like to work with. Um also the raw data set probably also determines that to some degree as in how big is it? What are you dealing with? Um yeah, I guess those would be my criteria, decision criteria. — Okay. I suppose it depends yeah. How complicated the task is and whe like do you actually know what the formula is you need to type because if you got to go and look up the formula yourself — that's going to slow you down. Whereas if you know it then it's probably quickest to just type it yourself. — Um okay. And do you have a sense of how complicated um a problem um copilot can solve in one go in this? Like is it best to do like simple prompts where it's like just like uh create one new column or like do one small calculation or you can you go for like longer more abstract things? — You can go more abstract. The good thing is that you don't have to necessarily figure it out yourself as well as in you don't have to necessarily uh Google for formulas or anything or look at a cheat sheet. You can actually ask it what the best way of solving this is and it basically breaks it down step by step. Um and then yeah going along those steps usually helps also with this loading time right now I do since I have prompted it previously with that um it takes significantly longer than what I've previously experienced so it might be as well a worker connective issue of some sort right now um but uh I haven't

seen it uh struggling with like a standard Excel formula combination. So far it handles those very smoothly. — Okay, that's good. Um, so yeah, I say I tried the equivalent with um Gemini in Google Sheets and that gets stuck on like anything that's kind of non-trivial. So I'm impressed this is actually doing something is it's a little bit slow but um uh it is at least doing something. Uh actually there's a question from the audience here. So maybe we can take that while we're waiting for co-pilot. — Uh so uh Sue says why using square brackets in column names? What happens if you don't? — Yeah, it is just a recommendation by Microsoft to use square brackets for um columns. Um what happens if you don't? Uh if you actually have unique naming, nothing will most likely happen. Um, but if you do have uh a lot of names that are somewhat repetitive, as in it could be a value um within a data set, but it could also be column um header, then you will probably run into inaccuracy. But I've had I've tried both to be honest with brackets and without brackets. Uh yeah, it just the same with um so Microsoft SQL Server the database you need to use square brackets around column names there. Uh and I think especially if you've got spaces within the column name then you got to use square brackets to let it know it's all one entity. — Exactly. So I've just tried to relaunch this in a new chat um so that it's been taking forever. Um, otherwise I do have a sheet where I've done this prompt successfully. Uh, but uh, yeah, here we go. Okay, — I'll let you move on. — It seems like it choked. Okay, that looks much better. Good. Okay, just basically picking up from where we left off in case you forgot. I wouldn't be a I couldn't blame you honestly. Uh it's been a bit but basically what we wanted to do is we wanted to no longer have these values of annual subscription or monthly subscription in the event type but rather use from the revenue data raw the event type which is either a new subscription with a subscription type being annual or a renewal. Um let's see if we have some renewal data that I can display. No. Okay. So in this case probably this already gives us some insights that when we did the cross reference the renewals were most likely being kicked out. All right. So checking the um the actual results here that translated everything nicely into this column. So we're good here. Okay. So finally we can actually move on to create our pivot table. Um this is just transferring. So now we still have event counts, right? We do need unique users. Um, so doing this via creating a pivot table and a new sheet based on the data in event revenue data with rows being the user ids, columns event type all what we just created and values being the count of event type all. So when we send that Wonderful. Okay. So now we have our pivot table here. Um we see user with that ID has done one registration but nothing else. And now we have this person for example who started a trial and who worked out three times but didn't subscribe. So with that we can now build our funnel asking it to create a new sheet called marketing funnel counting the number of unique user IDs in the pivot table with

registration being noted in zero and doing the same for trial start workout and new subscription. Good. So, here we have already it's not quite done yet our funnel. So, we can already see that it seems quite healthy on the top. And now we see quite a bit of drop off, but there's obviously a nicer way to visualize that. And again here I think so I'm going to prompt co-pilot but obviously it would be very simple for you to calculate the funnel conversion rate by simply um dividing all of these down the funnel events by the top funnel registration event. Um, I got my prompt here. I actually think to be honest, uh, just calculating this yourself in Excel would be quicker to be honest. Um, there we go. Um, and then another thing is, so we already saw this in total numbers that the funnel gets extremely slim here. Um we can also prompt it to calculate the drop off rate from stage to stage. So it's basically just two sides of the same coin but um yeah getting that in different expressions. So, we actually do see when it says base, by the way, um you can obviously um I would probably change that into 100 um% because it's just weird if you do have great to visualize in a bit which we're going to do. Um base is not a value that it can work with. Um so here we basically just want to create a bar chart displaying um the conversion rates. Um and I've realized that it has a tendency to flip around um these funnel stages. Um, which is why I'm instructing it exactly how I want it to be. Um, so I want the registrations at the top and then basically a proper funnel. So the widest part at the top, the most narrow part at the bottom. And uh most I mean um most people won't necessarily need the help of copilot in analyzing the funnel. Oh, I see. So it picked up actually on something. All our renewal values were previously um filtered out because they had mismatching um event values. So this can actually be removed. Um which will be interesting to see if it actually takes it right now from the revenue raw data which originally had renewal events. Um so in this case you will need to manually um remove those obviously um or prompted but that was based on a mistake in my prompt up Yeah. And we have our funnel.

I did that. Take my front. And the next thing would be I will ask it to analyze the funnel health. Um although you probably um I mean most people will see where the issue lies. Yeah. And uh here we go. This is exactly what I'm saying, which is why specificity really helps is to now I would need to reorder basically the series or prompt it to start with registration and change that. For the sake of time, I'm going to leave that for now. Ask it to analyze the final health. It's going to generate some insights that I like to then just copy into this sheet as a kind of like a milestone that you can then later on reference. Um we do have one more thing to look at which is basically checking um the if workout time correlates with subscription uh likelihood. There we go. Um Copilot doesn't have a very good sense of aesthetics as in it will most certainly um put any written insights behind a graph. Um, so yeah, here you have a little bit of an overview over the key metrics and it identifies the biggest revenue opportunities already. So obviously it says an 8% uh or yeah an industry benchmark uh would be 8% plus which is also interesting. Um obviously when you're working with this these would need to be confirmed um generally I wouldn't take them um for yeah for granted. Um so now we have that uh and obviously the next question would be okay how can we influence that would be okay workout time to subscription do these people just simply not work out enough as in use our product enough which is why we would want to look at is there relationship between the workout time and the subscription conversion rate. Um, so we're going to create one more pivot table. We're going to reference the exact same event and revenue data table. We're building a pivot table in a sheet called workout data with rows being the user ID, columns the class type, event type all as a filter and we're setting that to workout and then the value is the sum of the workout time. Um so here we see the different classes and now we see the total amount being worked out by the different users. Um get now we want to um look at if actually yeah we basically just need the uh the individual uses workout time. Um, so I want to use that in a new sheet that I'm then cross referencing with the subscription. So I'm creating new sheet called workout time subscription to display the number of user ids per sum of workout time. So I'm actually looking at okay so for 1 minute how many users do we have? Um so this is basically the grand total. Um just to see that distribution.

Here we go. We see that the lowest sum, it's not quite done yet, but it seems like the lowest amount of minutes someone can have worked out within our um sample uses is 50 minutes. And then that goes all the way down. This is the grand total which we can delete once it is completed. to 309 and again confirming that basically. So, we already said um we might want to just uh look at maybe the count if 15 minutes. count. If in this range um the criteria is uh 15 up here and it comes out with eight. So that's a good way to just confirm actually this is not rubbish that fits. And then in the next step we want to basically map the distribution in a graph to look at how does it look in a distribution. So based on the data in this sheet at a bar chart with x-axis sum of worker time and y-axis number of user ids and the series is number of user ids. Okay, it added one more series here where I need to prompt it remove the serious. sum of workout time on the graph. Hopefully that will fix it. So looking at this, we already see that um there's definitely some sort of cliff around the between the 55 and the 61 minutes. Uh here we go. This looks much better. Um so there's something around that one hour mark um that seems to um yeah where people um usually cluster around. Um and then the one thing that I want to ask is how do I basically cluster those like based on this distribution what are good ranges to cluster into high, medium and low uh workout pens groups. Then once we have that um it will give us a recommendation recommended clustering. So this is low, medium, high. Um we have a very equal tilebased distribution or there is behavioral based natural breaking points. um with within that range. Um it even recommends

um it recommends the behavioral based uh yeah this aligns with the typical fitness app behavior. So in the end I think this is where you pull um your experience and knowledge that you have from within this the business to make a call of on this. So in this case I'll need to define that um I want to create another sheet um based uh on the user ids from the workout table uh the pivot table workout data and a column called workout time total with the value in grand total from the same sheet. So basically I'm just taking this column here and I'm taking column A and actually G and I'm going to put that into another sheet um where I am going to classify every user as high, medium or low. So we're keeping that here. Okay. Don't have this one. Now based on the option one of workout pens some classification. Um, enter a column called workout group. Um and enter high, medium, uh or low based on their work. Total sum of workout minutes. All right, we're just coming up to time. So, uh we're carrying for a little bit longer to finish the last task here. Um from the audience, Ash is off. Um so uh if you need to jump then please do come back for the last few minutes to watch you're on the recording. We'll send that out the next day or two. Tomorrow we've got a session on data visualization for data storytelling. So of course data storytelling is how you provide a narrative with your results. We're going to focus on the data visualization side of things. Uh and then uh in a week's time on Wednesday of course as Reese mentioned we've got the radar virtual conference. It's one of our biggest events of the year. That's going to be a lot of fun. And then moving into April, we've got a whole week of getting started with AI assisted Python. So, uh please do uh check out the upcoming webinars. Uh come back to those and definitely uh sign up for radar. Uh all right. Uh Laura, I'll let you go through do the sort of final task and then uh we maybe take time for one or two questions. — Definitely will do. Uh it's actually one more um problem. Probably we need to merge the subscription data. So from the um pivot table event types reference the event new subscription and enter it in a new column in workout time. I believe that's the full name of the sheet two. All right.

Good. Now we know the people who have actually subscribed as well. So um the next step is to create a table clustering the number of users in each workout time category. Um, and the amount of subscriptions of that group. So again, this one you could basically then validate if it's saying, okay, people in the workout group um high. There we go. Um have 85 subscriptions. The way to validate that is to just filter here and then check for that. Good. And then uh again you can ask it basically. So we already see the subscription rate is significantly higher here. So meaning that we went with that um percentile to say we need to move people obviously to high um the insights from this clustering insight generation. And uh basically you get your in this case I'll just need to take a peek as well where we decided on the clustering. Oh, here we go. that even mentions it. Cool. Um, there's a clear correlation. More workouts, higher conversion. So, once we're actually moving from over 50 minutes, from 51 to 58, it doubles. And then there's another step once we reach 86, it more than doubles one more time. Um so that would basically be a way uh strategically to say good we have our marketing funnel here. We have a clear bottleneck in a monetization. The way to work with this is to move people to 86 minutes of working out basically. Um and then beyond that obviously you can look at the class types and so on and so forth. um with a similar uh structure creating a pivot table and then creating that correlation between revenue data and the event data and that's about it. All right, wonderful stuff. Uh that's very cool and I have to say um it just seemed like a nice alternative way of working here. the fact that like you don't need to like worry so much about the nitty-gritty of like what's going into a particular formula or what the micro task is you're trying to solve. You can just go high level but this is the thing I'm trying to achieve and let uh let co-pilot figure out all the details. Um all right so I guess one more question is so we've been using co-pilot from inside Excel most LLMs now will also deal with spreadsheet files so you have got an opportunity to just like upload your spreadsheet to claude or judgi or any of these others and or I guess even like the standalone co-pilot and say can you do something there like do you have a sense of like which way round like it is better to work the with uploading letting cla all the very small tools handler is the validation of it. So therefore I have I mean even just in a marketing funnel creation and then looking at the workout groups um I have created I can't even count how many sheets right and when you break it down you can at least validate is that true um whereas if you use include it usually gives you a nice

little wrapped version that you can ship but there's actually no way of confirming um which I would say if you do have a hunch if you have some good. You know, you kind of you already know what the data says, you just have interpretation issues or something, then it's great. Otherwise, you might run the risk that, you know, there's a bit of hallucination going on. — Ah, yeah. So, we got the trade-off in that it's they're more powerful models than what's running with the copilot side. So you can do more complicated task in theory, but then having that — feedback loop to visually inspect the spreadsheet, you've probably got to then — ask it to create a new spreadsheet file and then go and open it up in Excel and then go back to your model. So I can see how that's much more of a pain. So it is again it's weird workflow loops where sort of that in between situation where like copilot side Excel is a bit slower but you have the tighter feedback loop and so you're going to probably have to like play around with like what workflows work best for different situations. — Yeah, certainly. Um it's great for brainstorming as well to look at how do I best solve this. Um so I probably would use it for that. Same with uh insight generation. I also think it's great that if you're really reliant on something being 100% correct, you probably want to be able to actually validate that. — Okay. Yeah. Actually, that leads me to one last question before we wrap up then is when things go wrong, I mean, we saw it even here that it doesn't always get the answer right first time like what do you how do you deal with occasional failures and like making sure that you are given that quality control there. So uh if there's an error, you do spot it and you can fix it. — Yeah, I hoped I can share a little bit on at every step. We al always do a little bit of a validation. Um I guess it's more about like breaking down definitely keeping raw data untouched so you can always go back. So if everything falls apart, you probably have to start from scratch, but you haven't compromised your raw data. So that's one thing. And then the other thing is um creating a different sheet a different excerpt of your data on every step so that worst case scenario something breaks you go one step back instead of starting completely new. And uh yeah, you do have obviously your um your formulas so you can check like even just with little counts and all these kind of things otherwise I had scenarios where I did ask it but that number is off from that number what's going on and then it corrects itself. So you need to be switched on right you can't just lean back and let it do the work. — Can't outsource your entire brain to the AI. You actually still have to think. Okay. Uh, no, I do really like that point about like don't let it mangle like your original data source. If you got raw data in there, then you want to be working on a copy. So, if it messes it all up, you've not lost your data. Uh, yeah, that does seem like a good uh the safe option. Okay. All right. Uh, with that, we are well past time. So, uh, thank you so much, Laura. That was uh very cool stuff. Very much enjoyed that. Thank you to the audience who showed up today. Please do come back tomorrow for our storytelling session. next week for radar. Hope to see you.