Shane Butler: Virginia.
Sravya Madipalli: Hello, people, we’ll get started soon, let’s just wait for… How many people can join? Good morning. This is going to be slightly cozier. Meeting? There was less time that we promoted. We talked about this, and we also changed. Times?
Shane Butler: It should be a pretty good topic.
Sravya Madipalli: Let’s see who all could join… Maybe I could… do you all want to share where you’re joining from? I am based out of… Bay Area, California.
Shane Butler: I’m just getting my camera.
Sravya Madipalli: down, and maybe… Would be great to know where you all are joining from.
Shane Butler: Toronto.
Sravya Madipalli: Toronto, great! Awesome. Okay, let me just admit, people. Coming in… Yeah, hello?
Shane Butler: I just got a few more in.
Sravya Madipalli: Anyone? from… From nearby? So… Awesome. Okay, it’s going to be slightly, smaller audience today, probably, that’s fine. But we can get started soon. Wait, wait, and we’ll just keep getting more folks added to this. Today is, we’re going to talk about, Google Sheets getting… analyzing data in Google Sheets with AI. And how do we do that? I’ll just start sharing that so that you all see it. But… Yeah, I’ll give a quick introduction about myself. Hopefully, you’ve seen me around, but even then, maybe to kick off, a quick intro. I’m Stravia Maripali. I’m one of the co-founders for AI Analyst Lab. I have around, you know, 15, 14, 15 years of experience now.
Most recently in Grammarly, which was renamed to Superhuman. Before that, I was at Microsoft, I was at, later, I was at eBay, and then Nextdoor. And most recently, I’m coming from Grammarly. So, yeah, that’s about me. Along with me, at A Analyst Lab, we have Shane and Hai. Hi isn’t able to make it today, but I’ll hand it over to Shane.
Shane Butler: Hey everyone, I’m Shane. Yeah, I’ve been in data science for a little bit over a decade. Past couple years been, working primarily in AI evals and agility analytics. So, yeah, excited to get into… Analyzing data in Google Sheets with AI. I know we already have a question in the chat, from Attendee. Can you tell us why Google Sheets with AI is important?
Sravya Madipalli: Yeah, yeah, I just saw that. I was, for some reason, not able to listen until now, so I just got my speakers ready. So, not sure if Shane, you spoke or someone spoke, I couldn’t hear you guys.
Shane Butler: Oh yeah, I just did an intro.
Sravya Madipalli: Okay, awesome. Yeah, why is it important? I mean, that’s a good question, but also almost any, tool today, we’d love, we’d want to make sure that we understand how to use that tool with AI. I mean, if with… if using AI is taking you the same amount of time as how you do it regularly, then probably does not, like, make a difference. But once you probably see through demos today, you would understand why it’s important To ensure that we understand how to make our workflows so much more faster and so much more better. And that’s the reason why, we are trying to teach this here, and we’re gonna add this as a part to our cohort as well, that we teach.
So, the simpler reason is that it’s going to be way faster, if you use it with AI. That’s the reason, yeah.
Shane Butler: Yeah, and just to add, like, a lot of the… A lot of the ways we leverage AI to do analytics is through cloud code, or codecs, or, open source models, typically running Python or SQL, or creating charts, By writing scripts, and we’re mostly, like, kind of in cloud code that entire time. But, a lot of our stakeholders, or many people, many, many, many people who do analysis don’t do that. They don’t, you know, do that via the Python or SQL, they do that in Sheets, in Excel, and so it’s important, I think. as these tools get better and better, that we just work them into our existing workflows as seamlessly as possible.
The nice thing with Sheets and Excel, too, is it’s really easy to kind of validate and… understand what’s going on in the sheet itself, whereas, you know, when you’re getting an input into a Python script and output, a lot of that’s kind of obfuscated.
Sravya Madipalli: Yeah, exactly. We have a smaller group here, so please don’t, like, you know, you could just unmute yourself and ask the question, or raise your hand as well, okay? And we could go through. Awesome. Okay, we can probably get started, we might, you know, have a bunch of time at the end as well today for questions, but let’s just get started. Okay. So, I just, today is… we are going to basically look at analyzing data in Google Sheets with AI. We’re going to go through how we could do this with Gemini, with Google Sheets. That’s pretty easy for you to follow.
And also, how do you use Claude Code to get it to work with Google Sheets and, you know, produce trackers and insights and, you know, all of that stuff, all of the cool stuff. So I’m gonna share those demos with you at the end, so please make sure, like, you stay around. I’ll try to, have around 15-20 minutes at the end for questions, so we could probably end this in, like, the next, 30 minutes or so, okay? Awesome. Let’s jump to the next slide. So, before we go ahead, I want to ask you, I know, we have our smaller cozy group here, but I wanted… to know, maybe in chat, if you could talk about who here has a spreadsheet they have been avoiding.
And also, would love to know your role, if you’re a PM, analyst, engineer, you come from finance, founder, whatnot. So, and also would love to understand if… You have, you know, is there a spreadsheet that you’re trying to avoid? And, you know, things like that. Maybe start with the role. Yeah, I see the role, and then we can… so I’ll type mine as well. Okay, there’s a PM here, that’s great. Strategy and analytics, awesome. So, whichever, domain you are, you probably, would be either consuming data via Google Sheets, or either you’re creating Google Sheets yourself.
So, that was the intent of today’s, you know, lightning lesson, is that irrespective of the type of role you have, you are definitely looking at something in Google Sheets, and how could you make, like, you know, use of AI to ensure that you would get more value from the Google Sheets that you are either developing Google Sheets or either, you know, consuming Google Sheets. Awesome.
Okay, so I know this is something that Shane, myself, and, you know, a bunch of us have faced, where there, my question about who here has a spreadsheet they’ve been avoiding, we had, especially in my previous companies, where we had really wonderful dashboards and, you know, complicated, like, insights and all of that always shared. But guess what, was the most popular way of people looking at metrics on a day-to-day? It was Google Sheets. I know maybe for some of you it could be Excel, or, like, anything with, like, a cell and, you know, that structure. It’s basically because it just dumps all the information that you need, and in a format that’s… that a lot of, like, execs and a lot of.
Leaders are used to getting consumed, like, consuming data. So, this is something that I’ve seen time and again, that even though there are, like, nice fancy reports and insights in a dashboard, people… there are, like, a type of reports that always get shared via Google Sheets, and people look forward to those to understand more and more about data and metrics there. So, yeah, this, because your leaders, if they believe in data, like via Google Sheets, it comes to you and your team, especially if you’re an analyst team or a data science team, that you help them with, you know, like, data that goes into those Google Sheets, and it’s very tiring.
It’s not a typical, like, data science job, where you’d rather want to spend time on generating insights rather than working with sheets and tabs, so I’m very grateful that we have AI now that could help you with all of that. And let’s see, we’ll see that later in the, in our demo. Awesome. So here’s the loop. I basically call this, like, a four-step loop. It works for any spreadsheet, right? Like, and, most people probably, like, you know, jump right into, maybe? visualizing data, let’s say if there’s a question, and there, you could create a CSV that was shared with you with a stakeholder that did that.
And you jumping right into, like, the question right away might not be the right thing, because you’re skipping the most important step, where you’re not even, like, asking if the question that you were asked is the right question or not, and if you actually checked, a bunch of data issues. And validated if the data is right at the first place. So that’s the reason you ask. You make sure the question itself is right. And ensure that there is a reason behind the question. We actually get into the whole question loop a lot in one of our courses, AI Analytics for Everyone. The entire week one is based on that, because if the question itself is wrong.
then the entire report, or visualizations or insights that you’re going to bring is going to be utterly meaningless. So that’s the reason it’s very important to get the first step absolutely right. And once you get that, you basically go to the second step of understanding if you even have the data that you need, and is the data formatted right? Is the data, containing right information, or does it, like, you know, is it missing the information that you’re looking for? And then you basically do the visualizing and, you know, summarizing part. Okay, so, so once, like I mentioned before, asking is basically framing the question and not the task itself, right?
So, I basically tried to get together some examples of both good and bad here. So, let’s say analyze the spreadsheet is, like, a bad. Example. What does this act… what’s a good one? What’s a good question look like? It is basically, is paid search still profitable this quarter? That’s a good question, because you’re talking about specifically what, is… the product that you’re looking at, here. And you’re also asking the, you know, the actual question for that particular product about profitability. And you also have a timeline here about the profitability in the paid search product. for this quarter.
That’s just so much more, like, specific, rather than someone giving a spreadsheet and asking, hey, what’s wrong? Can you explain? Can you come up with some insights with this? So, yeah, it’s the same with a few other questions as well. Yeah, I try to make the bad look really bad, but you might have seen this, right? People come up with questions like, oh, could you explain how much, you know, is the product healthy? That’s such so vague. You want to be ensuring that you are answering questions that are as precise as they can get about the product, about the timeline, and about why you’re actually, you know, going after the question. Only when you get those specifics right.
that’s when you could go and tackle the, you know, tackle the issue and try to basically eat… this doesn’t have to be something that you do with just Google Sheets or spreadsheets. It could be with anything, right? It could be something that, helps you just avoid so much of, bad, like, you know, like, if you work on this, analysis for a long time, and then it, eventually will be not important, because you didn’t ask the right question in the first place. So, this is something that helps you beyond, a spreadsheet or a Google Sheets-related work, data work. It could just help you in any data work. Awesome. So, yeah, let’s see, we have, like, we, I have here a type A or B in the chat, right?
So give me the revenue numbers, and which channel drove the most growth, and should we increase the budget? What do you think? is, like, you know, a good question in this, A or B? It could be straightforward, but… Trying to get. Okay, Attendee says, B. Yeah, so… Yeah, basically, it’s a pretty obvious one. I was trying to make sure you’re listening. So… but give me the revenue numbers. This is also not pretty uncommon, right? It’s a pretty, you know, a very… you… like, usual type of conversations you have with your leaders, and they’re like, hey, give me the year-over-year score for this metric, or how is the new bookings, or the revenue doing?
And, you know, they don’t even specify if it’s this week, this day, this month. And it’s very important to ensure that you have the question framed right. Okay, so this is another step that, especially when we don’t have time, people tend to skip, about the three tests before you trust any number, right? If you… you need to ensure that if your total adds up or not, right? Do segment numbers sum to the top line? Magnitude, like, is this number in a plausible range or not, right? And do you have one known number? Like, verify one metric that you already know. And guess what?
All of this is very important for you, especially when you’re trying to do evals, when you’re trying to create AI systems that checks itself as soon as there’s an output. So, let’s say you have an AI agent in a skill that produces a tracker, or a sheet, or produces a final number as an insight, and you want to make sure that AI checks itself, and you create evals. For that to happen. Then, these are a bunch of things, these are, like, you know, three things that you could add to your eval, saying, hey, yeah, this is the number that you gave, but what about if you add segments within it, do you… does it end up having the same number or not?
Especially, we are talking about mutually exclusive MEC numbers, right? Mutually exclusive, cumulatively exhaustive numbers, so if all of those adding up, does… Do they add up to the total number or not? And then, if… the range is also very important, this is something that I’ve done as part of my work as well. Where, I created a metric, anomaly engine as part of our, like, you know, end-to-end AI workflow to understand if a metric dropped, why it dropped, and the reasons behind it and stuff we used to run every weekly.
To ensure that basic AI checks were done before a human checked it, who is the owner of the metric, before the owner of the metric checked it, we made sure that the magnitude of the metric makes sense. It is actually within the plausible ranges of it. you could basically do it as part of a sta- like, get a metric if it’s within a standard deviation of that metric variability. Then, you probably are okay that this is something that you would, depending on, again, how what type of variance you see in that metric over seasonality, over, like, different seasons.
Once you understand that part, it totally helps you get to if the metric makes, like, you know, if it… if the metric is within the range or not. Okay, I’ll try to go fast, from here, because we have some really nice demos that I want to make sure that there is time for. Okay, Yeah, so let’s jump into a Google Sheet demo. Let me stop sharing my screen. Okay, oh, it’s so hard to find these things live. Oh, I was actually sharing the right screen before, but that’s fine. Okay. So, let me do sheets.google.com, new sheet… So, I have some CSVs. lying around, and let me just open them, and start working with Google Sheets here. So I have a CSV called NoMart Daily Metrics.
So, Nova Mart is a dataset that we work with at AI Analyst Lab. It is basically an e-commerce type of dataset that we synthetically created, where we have all our analysis run through, both in our free and paid courses. So let’s say I have this right now, and let’s work with this in interesting ways with Gemini. I’m going to send a bunch of these things to you as giveaways so that you could, you know, work with it yourself and understand. I think this is a pretty common thing you might be doing with Gemini. If you’re not already doing, please, like, you know, try to leverage Gemini. for your Google Sheets, so look at what I asked it.
What is the revenue trend over the last 3 months in this data? Is it growing? Is it flat, or declining? Give me the month-over-month percentage. And look at what’s happening. Gemini is here evaluating, your data from the CSV, It’s talking about, it’s analyzing the revenue trends, it’s actually analyzing data selection, it is analyzing, like, you know, it’s going in the loop here, and look at what it came up with. It actually gave the monthly revenue trend… let me actually improve this… Wow. It’s a good-looking chart. It actually has everything intended, it talks about total revenue, and how is monthly revenue doing, right? So, I could insert or do a preview. Let’s insert it and see.
So, what happened when I insert it? It basically… let me just… It created this chart, it’s inserted here into one of your sheets, and it also created a nice table for you to look through, right? That’s pretty cool. You could also work with this information, again, like, which day had the highest number of orders, what is the average conversion rate, and things like that. But let’s ask it a different question. Let’s see if it can get this done. I’m actually gonna ask it. Add a new column that shows day of week for each date. Let’s see if it’s going to create this.
So… this part of the demo, I’m going to show you how to work with Gemini within Google Sheets to get this done, and the next part of demo, I’m going to show you how to work with Cloud Code to get the data that’s sitting in your database, how do you work with Cloud Code to create a Google Sheet and, you know, help it help Claude go to help with adding insights to Google Sheets, and, like, you create a Google Sheet that would take you, like, I don’t know, a day or a week. that you could create with Cloud Code in minutes, and share it with your stakeholders. So that’s the second demo that I’m going to get to, but this part I wanted to share with you what, like, inbuilt Gemini could do.
Okay, awesome, isn’t that cool? So, we asked for adding a new column, and it did add a new column. And it also has a formula that you… it used to add a new column, so you could actually validate if that’s true or not. So, and it did it within minutes, within seconds, actually. That’s pretty cool. So one last thing here, and then we’ll do something even cool with Gemini as well. So, I’m asking it which month had the highest total revenue, show me a monthly summary, sorted from highest to lowest. Let’s see if it’s going to do a good sort or not.
One thing that you’ll see with AI, anything, it doesn’t have to be just Gemini or Cloud Code, is the… vague your prompt is, the higher the chances of LLM to interpret your question differently, right? So… I’m generally worried when it comes to simpler questions, but it’s simple enough, but not too, you know, different as well. So that’s the reason, let’s see. So it… I asked which month had the highest total revenue, and look at this! It came up with December having the highest total revenue, October, September, July, and January being the lowest. And does it align with our charts?
Yes, it does align with our chart, because we asked for a monthly revenue trend, and December has the highest, and look at this. Jan, July being the lowest. It also has January, but this didn’t have… this… this chart didn’t have that particular piece. Awesome. Okay, so now we’re going for a complicated one right here, and what is that? So… I am basically going to ask Gemini to help me with creating a tracker. A tracker that… You might have seen at your workplaces, too. So, look at this. What am I asking it? I have a raw data in tab called Novamar Daily Metrics. So, this is a spreadsheet, and I talk about what each column means, and I’m asking it to create a tracker for me.
A tracker that contains, like, you know, what, type of active users, what’s going to be the metric name, and what each, like, you know, column should look like, and what is the formula it should use to come up with a week over week, and month over month. Charts, and let’s see if it’s going to create a tracker for me, right? The reason why I chose Tracker is, in my previous experience, the biggest way that I’ve seen, especially in product data science or in any product world, is people use Google Sheets a lot for understanding the trends of how a metric is doing week over week and month over month. Right? And here, we are trying to understand that.
How do we do something like that within Gemini? Oh, it’s… And how do we do something like that, with Cloud Code? So currently, we are working with Gemini here. Okay, so, this, I think, would take some time. And look at this. It looked at my my prompt, what I was asking it, and it basically created this. It understood that it has to create a daily tracker dashboard. I will create a new daily tracker tab at the beginning of your spreadsheet. This tab will feature a 30-day rolling view of your key metrics, including daily values, week-over-week percentage. 7-day summary. I will also apply the requested Navy headers, so in my prompt, I actually gave it some formatting as well.
So, and that’s the reason it got it here, okay? So look at the steps. It actually… understood my prompt, and it gave the entire steps that it needs to do. Initialize… let me approve this so that it keeps working. It talks about initializing daily tracker tabs, setting up headers, populating metric labels, implementing value formulas. Implementing week-over-week formulas, implementing last seven summary formulas, apply number and conditional formatting, add key insights and verification. One thing you’ll see here is. I actually, at the top, asked for it to come up with a month-over-month, as well.
But… I… looks like the prompt that we gave, like, even though we talk about give month over month. Looks like it didn’t quite take the month-over-month part into the… its, you know, execution layer. So it only has implementing weak formulas, and because of that, we’ll probably not see month-over-month. in our tracker dashboard. But let’s see, what this’ll create. So, this is the set of steps that it’s… you know, tackling. And I just approved it so that it could do that edit to my Google Sheet. And this is going to take its time. While this works, I want to share with you, Cloud Code, how things would work in Cloud Code. Hmm… Okay. Let’s do things in Cloud Code now. Okay.
So, this… is my… I’ll try to make it bigger for you guys, yeah. So, this is my Cloud code, and here is what we’ll try to see how it works with Google Sheets. So, I did Claude, and we are in Opus 4.8, 1 million context, and let me ask… similar… thing in Cloud Code. The difference between Cloud Code and Gemini is, I actually have skills and agents that is going to help create this. For us in Google Sheets. So, my prompt to Claude Code is. And let me move to auto mode, so that I don’t have to keep giving it permissions. So this is an auto mode on Cloud Code, where I don’t have to give it permissions, and it’s going to execute for me.
So my prompt here is, create a daily metrics tracker in Google Sheets for October 2024 using the no more data in data practice. Including revenue, orders, average order value, which is AOV, active users, sessions, new signups, conversion rate, and stuff. I’m also asking it to break it down by the acquisition channel, device, product category. If you’re working, like, you would probably know that this is the type of tracker you would have in your thing as well. and add week-over-week comparisons and conditional formatting, add revenue trend charts. I’m also not just asking it for a tracker, I’m also asking it for a chart, and look at this.
I’m also asking of it for then add an insights tab with an executive summary, KPI comparison versus, you know, September. Let’s see if this is doing such detail thing or not. I’m asking it to have it by channel, device performance tables, and 6 data-driven takeaways that it could give me now that I have this data, okay? Look at this! It already created the file for me. Okay So, let me… I’ll ask it to open it. Open the Google Sheet. Okay, let me stop… let’s see if it opened or not. So… I don’t even… so, because… Okay, it opened the Google Sheet, let me just share with you what it opened for you guys. Okay, look at this! This got created by Claude Code, an entire tracker.
Top-line metrics, by segments, breakdown by channel, paid search, social, referral, breakdown by device, breakdowns by category. Wow, this is something that just… Takes hours if you’re creating it for the first time. this… it took such a small time for it to create. So, the first time I ran this, guys, it didn’t take me, like, a minute or two. It did take me 10 minutes, because I made it go through, and, you know, I was working with it to come up with the format, to come up with, you know, make it red if it’s negative, make it green, like, you know, green if it’s positive. I worked with it, but once I came up with the format that I want my tracker to look like.
I… helped… I worked with Cloud Code to let it create skills and agents, and also Python files, to make this a lot more easier for these workflows later on. And in a minute, it basically took all the data that kinda looks like this. into a Google Sheet in this format. How cool is that? So, this… I don’t know. It was mind-blowing for me, because I didn’t work with Google… it’s been a while I started working with Google Sheets, so I wanted to try this myself to share with you all. And I was extremely, like, surprised and, with how fast we can get Claude Code to help you get this information fast.
And, if you remember, now that we have week over week for every Topland metric, and also for… by different channel. for every date here. I also asked it to create a chart. Let’s see if it created a chart. Okay, it did create a chart. I would probably like the chart to have some more text and annotation, but this is okay. Maybe I could, like, work with it to make it slightly smaller. But yeah, this is the chart it created. And I also wanted to create an Insights tab that I could share. Okay, so look at this. This is the Insights tab that it actually shared, and it actually gave the key takeaways. This… I asked for it to be a certain format, I created this and the scale, the type of format.
This is literally something, once I check the data, and if I agree with the data, that I could share with anyone, like VP of Growth, VP of… data, like, whoever, right? It’s a pretty good-looking spreadsheet that contains insights in the right way. Awesome. Okay, so let’s see what did Gemini do? Where are we? Is Gemini here? Okay? Okay, look at this! Gemini also created the chart based on what we asked for. Not as fancy, it didn’t get all as many, like, obviously, when compared to how fancy we got the Cloud Code to work with. Because Claude Code, I mean, I kind of cheated here, right?
Because Cloud Code has a really detailed skill, a really detailed Python file, a really detailed agent on, like, all of these workflows. But, this worked off based on this prompt. But it’s still not bad. I think it did a good job, at creating this. Like I was suspecting, the month-over-month didn’t get added, though we asked for it, because Gemini didn’t create formula for it. When we saw in the steps, it didn’t have month-over-month, so that didn’t get created. But it did create, the week over week, and it got some of the formatting right. It got the key insights, though I’d like the key insights to be probably… oops. In a different place, maybe? I’ll just copy-paste here. Okay.
So, but it got… it froze the pains… not bad. So what this means is I could basically make my prompt even more clearer. And that could probably help me create this, you know, tracker within Gem… with Gemini as well. But, how much flexibility we have with Cloud Code to create a prompt like this is pretty cool. And I want to share with you how Claude Code created this as well. Let me do that, where am I? Okay, we are here. Let me ask this. Could you… Can you share? a simple, ASCII. diagram of… How you created this? Google Sheet. Okay, we can basically ask for it in as detail if we want even more detail, but we can also ask for it with any Details as well. Look at this. So what’s happening?
What happened? It took the data sources, orders, sessions, users, order items, and products CSV. When I created first time, they were not even CSVs, they were actually in database, directly. So we have a database. DuckDB, so it directly created from DuckDB, but to make, to use them in Gemini as well, I made them into CSVs for this demo, but you don’t even need the data to sit in CSV. Your data could sit directly in your database. Yeah. For this demo, it did take data from CSV, it loaded the data, look at this. I have an orchestrator, I created a Python file in my script. You can see it in the scripts.
This is the Python file I created to ensure that, this tracker file can get, you know, it is repeatable, and I don’t have to, like, you know, keep, working with Claude Code every time to do it, and I wanted specifically how I like the formatting to be, how I like specific things to be, and I worked with Cloud Code to create this tracker file for me. So it loads the data, it computes the metrics, it computes the week over week, and it builds all rows. Looks like that’s what… And then it uses Google Sheets.python file and Google Auth. Auth is where I have all my command line CLI-related, you know, information in there. And then it, created Google Sheet and, you know, Insights tabs.
Let me ask it… what… Skills and agents. and other tools. you use… to create this. Google Sheet. Okay, so it basically used a Create Tracker skill. So I have a create tracker skill, created for this, so the project skill that owns this workflow, invoking it, loaded the playbook, telling me the reference script, helper module, auth setup, metric type conventions that already existed. This is what saved me from building anything from scratch. So, I created… so, for any of these other things to get popped up and to get used, you want all of that to be saved as a skill, right? And that’s the skill that we created.
I also had an agent that I used last time, but looks like it didn’t call it this time, because I think the skill in Python vials were sufficed. In this workflow, but I also have a Google Sheet Creator agent that used a review… it has a review feature inside it as well, but since I think we got the Python file to be so, like, you know, concrete, it didn’t have to call the agent this time. And the tools, scale, read, and bash, so… and these did real work, already exist in the repo, so we have googlesheets.python, the sheets help wrapper around the Google Sheets API, and, you know, things like that.
Yeah, so if you’re creating these for the first time, you’ll have to get the Google CLI connected, like, your cloud code needs to be connected with Google. Along with that, you need these files to get it as fast as I got this working. Yeah. So, just wanted to share this magic with you all. I felt it was magic the first time I ran it. So, anything… so this is the type of tracker we created here. You could create any type of Google Sheets. All you need to make sure is you know your exact request. The more clear you are with the type of request, the better you’ll be able to work with Cloud Code to get it to help create this.
not just the first time, but repeatably, that you could create skills and agents and, you know, helper files and Python files for it, and ship it to the repo and, you know, share it with your team, where they could build these trackers In minutes as well. Yeah, so… this, let me stop sharing. I could talk through about… Give me a few minutes, and I’ll complete my… -Oh, what is this? Okay. Let me share my screen back, and I’ll go back to the slides, and… Beacon. Give me a minute. And we can go to Q&A in a quick moment. Okay, so we did the demo in Google Sheets, we also did the demo in Cloud Code.
And you saw that what Claudecore just built, it basically took the 3 CSVs that we have, and it created a live Google Sheet with 3 tabs, the first tab being a top line plus segments, and second being a chart, third being insights. And everything under 5 minutes, because we had a bunch of tools that had all this information that could help create this for us. Okay, what else Cloud Code could do? Like I was telling you, it could do a lot of things for you in a lot more faster fashion. It can answer a follow-up, build a financial model, clean a messy data, audit a dashboard, it could help you with scenario analysis, like, what if this happened? How do you model the impact? And things like that.
And cohort retention analysis. All of these are some things that I’m sure all of us have worked with, or either asked. For these requests, if you’re a non-data person, and if you’re a data person, you answer these requests from your stakeholders. Okay, so I wanted to quickly share what you’ll walk away with today. You’ll have a Google Sheets AI Prompt Pack, all the stuff that you could talk to Google Sheets with, the 3-Check Validation Checklist, and the Google Sheets plus, like, how do you set up things with Google Sheets, and, you know, things like that. And we share all of these and a lot more in detail in our upcoming course.
We have a course coming on July 13, literally a couple of weeks from now, and we also have another cohort coming in August, where we have one-on-one bootcamp. In this bootcamp, you will learn to create skills, agents, how do you even build an entire system that does this, not just with Google Sheets, but with, like, takes a very vague data question into something very specific that helps you run end-to-end analysis. With your database, and helps you create, like, these insights and, analysis, decks and all of that within, like. Few hours, when compared to, like, weeks of time that it would take you regularly. So we teach you how we build an AI analyst system. So we have a free repo.
I hope you all have looked at the repo, our free repo. So we chat with not just the free repo, we also have a paid, course repo that we’ll probably make free soon, but it is, like, it has 60 skills, 30, 40 agents. And we talked through how we designed and actually built that analyst system, and how you could build something like that yourself for your work.
So, yeah, please check out the 101 Bootcamp that’s coming up, and for people who are interested in more than 101, where who’d want to understand, like, you know, how do you get this, and also ensure that you create systems that are, gather context of your entire enterprise that you’re working at, and also validate how does validation work to ensure that Because… all the LLMs are non-deterministic, and if you want a same answer every single time, we need to build eval systems around it, and that’s what we teach in 201 Advanced. In fact, we just taught one.
Last weekend, and we are continuing, to teach a bunch of course, a bunch of material on open source models, context, how do you provide context to your… and how do you engineer the context, how do you validate things? And how do you do this in a bigger, wider scale in a big company? So we teach that in 201 Advanced Course. And along with these things, we also have another track about mastering thinking. That’s AI Analytics for Everyone. This is a 5-week course. It is currently running right now, and we have another one coming in August 3rd.
This… in this course, we basically teach people the frameworks Let’s say if you are an analyst, or if you are a pro- like, you know, non-data person, like product or design, and you want to understand the framework of how to think like a Silicon Valley product data scientist, how do you think about experimentation? How do you think about metrics? How do you think about causal inference and understanding. We teach all of that stuff. You could think of that as, like, almost like a mini master’s in data science from Silicon Valley Tech Workflows. So we teach that in AI analytics for everyone. We also teach that, basically, how to use all of that with AI. Yeah, so we have a bunch of bundles.
We do 101 plus 201, so that you could save $600 there, and we also have Builder’s course, where if you get your 101 bootcamp, then you could, like, if you… if you get, if you buy Builder’s course, you also get. 101 bootcamp, free with it. So it’s almost like two-for-one deal. And yeah, we also have 20% off everything here. These are the, you know, codes, but I’ll share all of these with you over email as well. Okay! It’s time for any Q&A. Any questions you have? I could just take any questions, yeah.
Shane Butler: I think we had a question in chat around, can you, like. Update the Google Sheet with, like, live assumptions and do scenario analysis. from, Attendee. Attendee, if you wanna… if you wanna add any other context to that.
Sravya Madipalli: Attendee, you could ask, like, live as well, if you have any other context, yeah.
Attendee: Oh, yeah, can you hear me?
Sravya Madipalli: I can hear you.
Attendee: Yeah, so basically I was just wondering if we… if it is possible to… for the model to pick, like, assumptions? from the file, and, like, or allow the flexibility to change assumptions. For example, you know, if we have… if you’re forecasting, and if you change, like, one metric, like the number of salespeople, how does the overall forecast change? So… That, and also, like, if it could, like, create different scenarios. or… I’d like a conservative estimate, or an optimistic estimate. So… so, like, make it more dynamic instead of just, like. Creating a forecast, and that’s it.
Sravya Madipalli: Absolutely, Attendee. So, what I tried to show you here was… like, how do you create something out of just raw data that’s sitting in your database? And, like, I wanted to show you the power of that, like, within minutes, it could create something like that. And what you’re asking for is even more detail, not just a tracker, but a tracker that’s dynamic in nature, that basically, either… there are two ways you could work with it, right? Either it’s in that situation you have a question for the tracker, and can it help answer you? Absolutely. You could literally work with Cloud Code on this. This is something that’s beyond Google Sheets, I would say.
This is something, let’s say you have an insight that you worked with Cloud Code, that it generated an insight, and you have a question on that insight, right? And at that moment, you could work with Cloud Code to understand a lot more detail. about that particular data, and you let it output that in one of the, you know, sheets or tabs, or maybe even create a Google Doc out of it, right? That’s one way. The second thing, Attendee, is if you want the sheet itself to be dynamic, right? If you want the sheet itself to take a bunch of parameters, and based on how the output looks like, you want it to dynamically change the type of Google, like, you know, sheet reporting that it creates?
Is that what you meant?
Attendee: Yeah, exactly, like, it picks up numbers, and if I adjust the assumptions, it modifies everything.
Sravya Madipalli: Absolutely. So that is something that I would definitely think we could work with Cloud Code to help it, because I didn’t do that particular thing with Google Sheets yet, but I’ve definitely done that with a bunch of HTML dashboards I created, where you basically are… it’s like parameters, right? You basically create, share a bunch of input parameters, and you want your output to change based on the input parameter. And I could either… I mean, and if you’re specific about how you want it to look like, I would create a Python file for that. Based on the input parameters, you have a different output based on the inputs to the Python files.
and you let your skill pick that Python file based on if it sees that type of data, and if it sees, then that’s when you trigger that Python file. So.
Attendee: funny.
Sravya Madipalli: I can think of ways on how you could do it. I’ve done it before, with other workflows, yeah.
Attendee: Great, thank you.
Sravya Madipalli: Awesome. Yeah, any other questions? I know we have a good group of people here today. Any other questions about the Google Sheets workflow, or about the courses, or about any… it could be, like, generic questions as well. You don’t have to worry about it, yeah.
Shane Butler: I think the thing I like about the Google Sheets stuff is, you know, it’s not just… it’s about all these types of analyses, we can still… you can still do them in Cloud Code and have them… Output in a deck, or output in a document. But now, when someone asks you, hey, I want to basically, like, see the work behind this analysis, you’re able to kind of communicate that and illustrate that in a way that they’re really familiar with. So, like, the… Insights tab that Savia showed earlier. Is built off of that. Other tab that had all of the, kind of, like.
Not raw data, it’s already done some calculations to it, but that’s a really nice way to… provide transparency and visibility so other folks can kind of validate, the numbers that, like, AI is outputting. Because obviously, you know, when you run through any sort of agentic analytics system, you’re opening yourself up to risk for it. Doing some sort of calculation or pulling data that it’s not necessarily what you, intended, or what the stakeholder intended, and this is a really easy way for people to basically, like, actually follow the numbers back.
So, even if you’re presenting it in a different format, like a doc or a deck, this is such a good kind of companion… Resource to share with those, to share with those readouts. Like, a lot of the work that we do when we’re Trying to validate Agentic Analytics is just, make as much visibility as possible. When you’re a data scientist or someone who’s, like, working within it yourself, a lot of that time, it’s like, oh, go read through this log or this JSON of all the queries it ran, or all the tool calls it did.
But this is, like, kind of on the other side, where it’s like, hey, maybe someone doesn’t know how to read SQL, but they can still go through this sheet and see how we derived those numbers.
Sravya Madipalli: That’s a very good point, Shane. I think from now on, every time I do an end-to-end analysis with Google Docs or something, I could create a Google Sheet with it.
Shane Butler: So fast.
Sravya Madipalli: I know, yeah.
Shane Butler: Before, I’d be like, oh… like, imagine if you were before, like, you’re already having all… spending all this time doing analysis as an analyst or a data scientist, and then you… share that. That was already weeks of work, and then imagine if you had to be like, okay, now I’m gonna try and redo this entire analysis in sheets, so someone else can kind of, like. understand what went on. But now you can just do it like you did in, like, seconds or minutes.
Sravya Madipalli: Absolutely. And even, like, I’m also talking about, let’s say, if… We are going to create end-to-end analysis pipeline, and we trigger, like, end-to-end analysis. We could also ask it to, like, log every step of the way data in each tab. So that it… we could follow it through. I know we were doing it as part of the queries, but also data could be saved within Google Sheets. And, you know, pointed within the Google Doc that, hey, at this step, this was the query used, and this was the data, you know? And at this step, so that almost we have logging of every step in the way, so that if people want to cross-check the data, and also validate it, they could just go back and run with it as well.
Awesome. Okay, any, any other questions by the group? Regarding this, Did you have any, like… like, vicious about, oh, this would be great to work with Google Sheets, and, you know, can we, have you tried that? Any questions around that, too? That’d be great. Maybe I can work with it, or play with Google Sheets a bit, and get back to you guys on that, if you have any of those ideas. Okay. Yeah, I’ll share a bunch of these, like, you know, some of the giveaways as part of the email, I’ll add them to the email and share it. You’ll also get access to the recording soon. So, yeah, you could also reach out, to any of us. We have a Slack channel, you could join the Slack channel if you haven’t.
I’ll share that in the email as well. And you could always ask these questions in Slack channel, or, you know, ping us on LinkedIn, or any of that. Would love to see you in future, like, you know, free workshops. We have, next week we have a workshop on Clod 101, about basics of how do you create agent skills, and a bunch of those things, so, and you’re free to join that, for sure. And also check out, like, our… a bunch of other offerings that we’re trying to do. We have free email courses and so many things.
Shane Butler: Thanks, everyone. Appreciate the time.
Sravya Madipalli: Thank you, everyone. Bye.
Shane Butler: See ya.