How to bulk-use AI in spreadsheets for fun and profit

Stan: Have prepared a few slides Okay. Can you see my screen?  
Ben: Yep.  
Stan: Alright. I was gonna move this around. Okay. And there's good positioning your camera. Okay. No. This doesn't work. Okay. I go. So, yeah, the our products are called, for work, the most recent one, which I'm going to speak about today. GAM or get another mail merge with which some of you may know. It's a mail merge for Google Sheets, Form Publisher, and Awesome Table. And, we have some of the top, products on the Google Workspace Marketplace. So the the plan today is to talk about, more specifically, GPD for Work. Is, a series of 4 extensions, 4 Excel Word, Sheets, and Docs. The most polished are for Excel and Sheets. With Word and Docs is for now a little bit simplistic. And, well, the goal is to enable you enable the users to make TELGPG or other AIs or their models, powered through repetitive tasks, super fast. Literally a 1000 times faster or more than you could So just showing a few screenshots before getting into the actual demo. This this is a screenshot of Excel. Here, you have let's say you have a catalog of Bites. And you open get on here, q p for Excel. You can type a prompt with taking your data and into account from columns and write, for example, a description, and it will write it in bulk for you. The same in sheets. I'll show it after. And in a word, it's similar, although it currently doesn't work in bulk. So popular, use cases for spreadsheets spreadsheets are the first one and the easiest is translating in bulk. Example, if you're, an ecommerce business and you need to open a country or a new locale, can translate your whole product catalog pretty quickly. With our tool. Of course, you can also create some SEO content or add copy at at scale. Categorize your product information or your CRM, analyze free form answer of surveys, etcetera, etcetera. There is really many use cases. I'm sure there will be questions, but and I'll run through a few of them. Why should you use PPT for work? Well, the number one argument is that you couldn't stop mission copy paste from ChargebeePT back and forth to your spreadsheets. And JPD 4 can run on Excel, up to a 1,000 prompts per minute. So this is pretty fast. And it's also reliable. Right? We can actually do 1,000,000 on Excel. And on Sheets, you cannot have 1,000,000 rows, but you can go up to 200,000 it will take a few hours to run. We offer a large choice of models, so Ben you're familiar with a specific model, you can use one of those. We recommend, GPD4O OpenEyes model. If you don't know particulars of each of those, and to the most reliable and, yeah, we provide good support. We have, extensive documentation. Why would you do that again? Instead of using ChargebeePT or Copilot or Gemini? Because their UI is not made for bulk processing, They have often no daily limits. They offer their own they don't only offer their own models, so you will you will not have load inside or inside Copilot and and so on. And, yeah, there there we go. So I can jump right in now or do we already have some questions maybe?  
Me: No. I think, yeah, diving in would be great.  
Stan: Okay.  
Me: Actually, just got one question on I know that ChatGVT has like Google Drive integration now. Do you have anything that lives outside of the sheet 6 out? Like, do you have anything that could plug it? I know they had GPT plug ins previously, but have you looked at anything to integrate within the chat activity interface at all? Is that possible or not? Given the features.  
Stan: It's it's not to my knowledge, and it's not something I mean, we've actually gone the other way around is integrating, DI inside the spreadsheet where inside the document where people live, and where the data lives. That you can see the data and what you're doing and the outputs all at once.  
Me: Yeah. Makes sense.  
Stan: K? Alright. So I'll show my demo on Excel. It works on both Excel and Google Sheets, but it works better on Excel. For many reasons. But Rusticel is faster and more stable and more reliable than Google Sheets in general. And so, our add on also works better. But once you've installed it, called gpt4excelword on the Microsoft marketplace, you have you will have a button here and you can open it Ben with a one click. And the, so just maybe for those who use who have Heard of this add on before. It started with function, like, this one GPT, and you could just say Hello? To the and it would send hello and send you the answer back. To GPD 4 o in this case. This works fine for many peep for people who are very savvy with formulas, It's actually not easy to use for many people. So we've developed this new interface, which is much easier to use and also more powerful anyway. So in this case, we have a  
Me: I mean, you could zoom in a couple  
Stan: first of all, is it is the font is it is everything big enough, or should I zoom Maybe. Alright.  
Me: Yeah. Sure.  
Stan: Let's zoom in a little bit.  
Me: Alright. I was just gonna ask, how does the  
Stan: Okay. So this is what you if you would if you were to sell bicycles online, you would have something like this. Okay? You have, a catalog of bikes. Here, I have a 150 or so. And you would have, yeah, 160. And you would have, so some metadata columns like manufacturer model name, the year, type, frame, weight, price. You could have more specifics like colors, for example, and then, let's say, a description. You would present on on your website. Okay? So this is all in English. Let's say that I want to open, France as a market. Okay? The first thing I'll want to do is translate, the description to French. So I'll do that here. I'll just put Ben French label on this column, And for this case, we have a special tool that is called translate. It's very easy to use. So the way it works is you just select what you want to translate here. The column, 1 k, in this case, you can translate to French. And I'll put results in home l Okay. You can test it out. On, let's say, the first Right. On the first three rows, Hello, launch. And as you can see, it is bringing the translations to French. Lower than usual, the demo effect. There we go. 3, subscriptions. Don't know if some of you speak French or not. I do. I can translate in other, language that you guys speak if you'd prefer, if you wanna check the results, but it's basically, you're testing telemetry in this case. So it it works. You can also add specific translation instructions community or audience, and you want the language to Ben turned in a certain way, you can provide this here, and it will do it this way. So that's one of the huge advantages compared to Google Translate where you cannot give specific instructions as to how the translation should happen. Okay. If you are Yep.  
Me: the 2 variants per row I saw that toggle  
Stan: Oh, no. I can I can, actually talk about it now? That's  
Me: does that look like? I might be jumping to another part of where you're gonna talk about this, but No. Not on that. I think that's  
Stan: do so variance is going to ask, attach EPT for 2. Translations. In this case, and you'd select 2 columns instead of 1. So I'll just enlarge that a little Bites. And let's run 3 more. Alright. So you see you get 2 2 possibilities, and then you can judge. Actually, I've noticed that when translating it's not necessarily, super useful, but it can be. It's mostly useful when you do some actual writing, or when you ask for a description based on some I'll show you that later. And, actually, oftentimes, the second one is better. I've noticed. The second one is better. Although, gonna make this, a bit easier to use for now, and we're it's what you were experimenting with this, but it's it's it's useful already. That's how it works. And once you're I'll just close it for now. Once you're satisfied, let's say, with your settings, You can just run, all rows, and it will run the rest. Over here, we have 154 And it's, the counter is going there. It's not necessarily appearing in order of the rows, but it's gonna do all of them. Pretty soon. Okay. So, yeah, we're doing over a 100 in less than a minute here. That's, I'll just stop it. Almost finished anyway. So that's the gist of it. That's, the, let's say, the easiest example. It's translation. Translation very well. Much better than Google Translate. Especially for languages that are maybe not the most common ones. Like, we had a customer that said, for Scandinavian languages, Google Translate doesn't work at all. But TADGPT translation worked It's Ben me well. Okay. So I can show you another, another use case. Let's do, extractions. So in this case, I'll delete this Yeah. That I don't have to move around too much, say that we don't have this, these columns here, weight and price. And we only have the description. This sometimes happen when you have, a catalog, from a provider that doesn't contain all the metadata, dutifully extracted, So basically, we're gonna extract structured entities from unstructured text. Okay. So in this case, I'm gonna extract the weight enterprise, and I'm gonna extract from column k and put everything in l and m. Alright. I'll start this little thing here, first row, is to tell, the add on where to start. Is not everybody starts their spreadsheet at row 1. So it's helpful. So in this case, let me just check Oh, yeah. I'll just delete this for now. Alright. So this is pretty fast. It's much faster than writing descriptions because the output is super slow. Super small. Sorry. So in this case, can see that Bites worked, 9.1 kilos or £20. This one, we can verify. Let me enlarge this real quick. Yeah. 6.1 6.8 kilos, 1 point £15. And here's 7.8 kilos. So it worked, but it's not perfect. It's not super usable because of what I would have preferred. Is to get only the numerical values. Right? Right now, it's this. I can't really sort it or filter it easily or maybe, do any arithmetic on it. So I'm gonna ask I'm gonna add extracts extract instructions. For wait, I'm gonna ask, okay. Return only the numerical values of kilograms. Okay. So I'll delete this And I'll run it again. And now I get good numbers. That I can sum or do anything with in in my spreadsheet. But that's really important because otherwise, in many tools, you get data, but it's unusable in the spreadsheet. So, the the fact that you can Really? Rify makes specific what you want and the tool is very powerful. And then when you're happy with it, you can just, again, run all the rows. And you'll get your data pretty quickly. Oops. Sorry about this. That was very sensitive. Okay. And sometimes the cell is empty. It's because there is actually no data. Any questions so far?  
Me: pretty easy to understand. There is a question from Al Bores. Which is a concern I've run into around using AI at work as the access to proprietary information by the tech companies or platforms in case of a hack. So can't upload high sensitivity documents, does this still access chatty b t in the same way that sort of passes data over or is it secure and private in in any way?  
Stan: Oh, we use the, APIs from the various model, AI providers that we support. Mainly OpenAI and, Anthropic. Google Sheets, we also have, Gemini. And So we are bound by their terms, basically, and their terms are very, privacy friendly. So they do not train models on inputs or outputs that are sent through the API. So that's the first thing. K? Secondly, the data that you, process through the software has to go through our back end. So it goes through our back end. But we are hosted on GCP. So it's encrypted at rest and in transit. It's protected by old, GCP, infrastructure. And we are, as a company, certified ISO 27001. And you can see how we process and handle the data in our all our documentation terms, privacy policy, data processing agreements,  
Me: Perfect. I think that's great. I think that  
Stan: k.  
Me: answers, all of that. David's question, does it call the API every time the page is refreshed?  
Stan: Every time the page  
Me: Yeah.  
Stan: is refreshed, so you mean every time you open the Excel file?  
Me: So  
Stan: That I'm guessing that's that's yeah. So no. We first of all, as you can see, that's also an advantage of using these both tools instead of the functions. Is that the the results are directly written in cells. This is actual text. I mean, I can I can modify it, if I go inside, I can modify it with my keyboard and just edit it? So it's it's as if you had inputted by yourself. Okay. But there is no refresh button. When there was refresh button is when you have a formula, like, you were using this one, for example, say hello in Spanish. I'm just gonna change it a little bit. In this case, you have the result, but it is a formula as you can see. When you use that, can be some advantages if you are the view is referring sales. We actually cash everything in the spreadsheet itself. So if you open it again, it will not recall the API. Okay. In Google Sheets, it's a little bit different. The the so we have the same both tools. And we have the same formulas. The both tools behave the same. But in with the formulas, are auto refreshed by Google, regularly, like, every 4 to 6 hours even if you don't use it. So this, used to cause a lot of problems. It's impossible to stop and now we have put in place some measures so that, it's it doesn't affect you. So We have a cache in the back end that will reply with the results so that it doesn't recompute and resend your credits all the time. Okay. You can also my advice, if you use these formulas is any way to always some point when you're happy with the results, to replace them with the values so that you've got the the results in raw text and you you have basically no way of losing them, anymore.  
Me: Perfect. Yeah.  
Stan: Okay. Okay. So now let me show you something, a bit more powerful. The, the custom prompts. Alright. So in this case, let's assume that we do not have this description. Okay? I'll just hide this. Calling And let's delete and delete this. And this clean up a bit. And I'll say a new description. Here. So you can, let me just So you have a prompt box here, basically, that you can use to type a custom prompt. I'll just copy paste in here, and you can refer to existing columns either by their letter or by their header. Depends. Sometimes typing a letter is easier. It's less error prone than typing the full header. But if you're moving your columns around, it's better to use the headers because they will stay put. Okay? So let's say that we want to write a description for an ecommerce website about this bicycle. Here, it's model model name, column b, manufacturer, and address. Mix it up a bit and here, put in it name of the header, do a little bit of, Standardization so that if you make a typo will not appear in blue, so you should know. But just in case. And price actually is in column j. I'll type price. And then weight is in column h. And frame type, column e. Okay. That looks good. When you use such, custom prompt, I advise to specify how you want the format to be returned because if you leave it to the model, it will sometimes return will oftentimes return markdown, which may not be what you want. In a spreadsheet at least. Can also ask for a HTML if you want, if you intend to paste the result directly in your WordPress site, for example. You've got a little preview here. Which little bit small. We're gonna make it harder, soon, but it helps you check if you've made a mistake or not. K? It takes each row replaces the tags here by the values. K? Alright. Then choose your output column. In this case, this one. And let's also say I want a 100 word descriptions that doesn't write too many words. I'll try And there you go. You've got 3 descriptions. So the problem with that is you can see, they all start the same. Discover. Discover. Discover. K. So this is where you might want to Click the prompt a Bites. To say to give it some examples of, full descriptions Or to, give it some seed words, let's say, so that it cannot start always the same. And you can refer to, an example, via you can say, okay. Use this as this description as an example and then refer to if you have description in a specific cell, you can put it. So let's say, s an m one, a description, and I say something very different, like, great. Great bike. So very short. I think it will get confused, but interest just to, trigger the point. Actually, I didn't, do anything. But this is, this is, getting carried away in my demo. But if you give it a proper example, I think it will oh, actually, sorry. Ben one is empty. Let's try again. Let's try again. And 2. One second. Alright. No. Change. So the more effect, but if we give it a proper description, I'm sure it would work better. I'll prepare this better for next time. I got improvise here. But what I can show you, however, is, the Marion, which you were asking about, Ben, earlier. So in this case, we will get presumably different results on both. Okay. So here we have a bit more variety. Discover, experience, introducing. So it's starting to be a bit better. And then you can yourself select the ones you prefer. Okay? As always, since this is a custom prompt, you've got a you've got a really give it precise instruction as to what you want, and you will get, as much as as good as the as good results as you put work into the prompt. We are going, later on to make this easier and to have preset, prompt templates or, specific use cases, especially for ecommerce because that's a big segment of our customers. But for now, you're gonna have to do the work yourself.  
Me: Well, I think  
Stan: Alright.  
Me: there's a question that I have answered, but I want you to clarify whether I'm correct or not. So Abel has asked can you see can you sort of say in your prompt, to use unique wording for each description And I said, I don't think it would work because it doesn't have the context of the previous generations of the descriptions. We couldn't then it's not taking into account the rest of the sheet. To then redo it right or no?  
Stan: Oh, actually yeah. Hey. You know what? Let let's, let's try something. So I I I agree. I mean, I I it's a it's a great question, first of all. We have some plans to make this easier. For people to do That's but there is So here oops. Let's keep this one. Which is we have 3 different one. We have one that starts with discover, Ben that starts with experience, one that starts with introducing. So there is a little bit of variety here. Let's say, And and now we can say, okay. Example descriptions. Alright. Oops. Alright. I yep. We have all descriptions. And I'm gonna try this again. So we're gonna say, l 2 3 l 4. Okay. And right. Here is unique. Wording. Let's let's try something like this. Doesn't seem like it's working. So this doesn't work for now. Doesn't work for now, but with the variance, you get a little bit better results. And we're we will work on this specifically because it's been a nice out by customers. If you worked I will say though, however, if you work a little bit on prompts to make it better, it will be much better. Bites, yeah, you're absolutely right, Ben, that for now since Each description is written 1 by 1. Doesn't have the context of the others, and so every time Bites a new one, so it doesn't know that it's repeating the same pattern every over and over. Another another thing you could do, which I have not prepared, if you're a bit, technical savvy, you can train an OpenAI assistant so in the platform, it's like a custom GBT, but on the developer platform side where you can, feed it with some documents some existing descriptions, and then it will use this data to generate its output. And so in this case, can already put your guidelines some examples, in into a deep and AI assistant and you should have much, more custom and unique output. Okay. So  
Me: Nope. I don't think so on those things.  
Stan: any other questions? Okay. Just, just so you know, it it works on really a lot of rows. Like, this was also it's very easy to make this, we launched GPT for sheets, which was the first product, we Ben had a lot of other people who who copied the idea, and it's easy to make a shiny demo. But what we've worked very hard on for the past year is making it reliable at scale, as scale means over a 100000 rows. The, so far that we have one user the the record to beat is 2,100,000, executions in one day. That was not on the on one unique spreadsheet. It was on 5 different because in Excel, you can only have 1,000,000 rows. He did it on, I think, 4 or 5 separate spreadsheets. On Excel, I needed 2. 1,000,000 in a day. And we regularly have people that go above 100,000 executions per hour, doing this so you can really go through your your whole product catalog very quickly.  
Me: Yeah. It's a lot of those.  
Stan: With that. It's it's incredibly powerful. Okay. So I'll talk about something a bit different I say that you have, a list of companies. And you want to fetch, data about those. Okay? We have a tool which for now is, is what it is. It is perfect, but it works. So this tool is, basically calling the Perplexity API. Those who who who don't know, Perplexity is like a a chat JPG that's connected to the web a very good way. And they offer an API which is not as good as their web app. It's it it does less work. Still it will help you get answers about that are not contained in the knowledge by itself, that are contained in the web, into your spreadsheets We will improve this tool with other techniques, web scraping in the next few months. But still, it works, reasonably well for some, in many cases. It doesn't work when you ask for a very specific detail a web page, such as what color is this image in this web page? This this will not work. Okay. So, here I have a list of companies and I'm gonna do a few things. So the first thing I'm gonna do is fetch the URL So I'm going to copy paste this here Alright. I'll ask the question about the company's here I'm asking for the URL and to format it, in a certain way. I'll do a bit more this time. And I'll start, you know, row 3. Okay. So, usually, when you ask the URL, it works pretty well. If the company is completely unknown, it might it might fail. But you can have the sources here, because I Tossell this this little thing here for sources. So you have the sources. So in case you wanna check, the answer, you can check one of these sources. Okay. I'll delete the sources for now. Ben you can ask for a description, I'll put a 100 words, not 300. And put it in Con c Oh, again, here you've got the description that appear here and the sources that it used to create this description. So the website itself, it may not have been super interesting but when you start asking for something a little bit, more custom, then it's useful. Okay? Go here. It tells you how we created the description of Microsoft. It's actually didn't use the Microsoft website. It used some other websites. That's one of the limitations of complexity is you cannot tell it which websites you wanted to, you cannot restrain it to a  
Me: And that and that's  
Stan: specific website or a specific sets of specific set of websites.  
Me: that's different to the web.  
Stan: It will take whatever sources it wants.  
Me: Interface, isn't it? Because you can specify, like, just search Reddit  
Stan: Okay?  
Me: for this information or just search sit inside Yeah. I think so.  
Stan: I so I'm not familiar with this feature of the web app, but I, yeah, I believe you.  
Me: And with and with obviously, with this, your you're basically saying about each cell in column a, So is there any way for that input to be more than 1 column so that if these if sort of column a, b, and c are now all correct, so you're happy with that's the company. That is the URL. That is the right description. And just to make it more accurate, in terms of getting the next output. So that number of employees Apple as a different kind of company it's actually pulling in all of those inputs as verifying what the input is that you're asking about. That makes sense.  
Stan: Yeah. You wanna you wanna use the 3 the 3 inputs to get the next the answer to next question. That's, unfortunately, not really I mean, not easily possible yet. You can do it if you just, essay create a helper column here and then  
Me: Yeah.  
Stan: concatenate the 3? It's a bit stupid. We will improve this. There is no reason that we should only have one column choice book. So instead here, we'll soon do is have a multi select here instead of  
Me: Yeah.  
Stan: go select. Okay. But for now, yeah, unless you use a helper column to concatenate the results of the previous inputs, and outputs, then you can only apply the question to a single column. This is different from the custom prompt tool, which I just showed, which is made to, to use multiple input columns but it doesn't work with the web search tool. This is, product in progress. K? So, yeah, I mean, you get the you get the gist, but I'll just do it Yeah. Actually, for a number of employees, this is probably where adding the sources becomes quite interesting, because then you can check the results. And it didn't work. Oh, yeah. Because I forgot to put the Right, Carl. So interestingly, it sometimes, put a range where the the the bigger number is before the smaller number. That's something that's perplexity also struggles with is returning specific, good formats. A technique that you can use to, deal with this is once you have the answers, the data from perplexity, you can use at a much cheaper model like GB4 0 Ben to reformat the answers, and it will be very fast and very Okay. Oh, you can use you can use this to create, to enrich, let's say, your lead spreadsheets. And then you can use the other features like categorizing or reviewing to score these leads again against some criteria. Example, here I have an existing and not your list. Of companies. With I have or let's say I already have this data, the industry, the company size, a bunch of, enriched data that could have gotten from, lists or my CRM or or web scraping, and I want to grade these leads, according to, or score these leads according to several different criteria. Okay. So I can use the custom prompt tool here I'll delete this. And I'll put copy paste this one in. That's Let me just check the column, yeah, columns. Company c. Company's Bites is d e. Okay. For this kind of exercise, we're reviewing and scoring, KB for a mini won't work well enough. You really need a a stronger model, so I I advise. And we'll put it in cone n. Start at row 4. And you get your grades. Because then you could run all of them k. So let's say that we have those grades. Ben you can score it against, other criteria, like Ben was the previous engagement, what what is the content interaction, etcetera? I'll do all of those, and then I'll focus on the adjusted action on these 2, actually. Think they are the most, the most interesting. I'll do this one quickly. Alright. So here, I'm assigning a lead score. From 0 to a 100 based on, a few metadata. Alright. And then you can say you can ask it something that's a little bit more subtle. Like, based on all of that, Is it is it hot, warm, or cold, which can then help you prioritize, which need to reach out to or how to reach out to them. So you can type prompt like this one. Oops. Not the right one. Sorry. Okay. Take your eyes each lead into hot, warm, and cold. Based on its score and the grade. And Ben Oops. Yeah. You see this one was at 45, so it's called This 192 hot. 75 warm, So it seems like it makes Ben's. This one was an a, but 75 is a warm. And you can then so search, ask for it to suggest a next action to you. Based on everything. So just the next step or suggested next action. And this is where it gets interesting. So you've done this whole process. And You've got data that you can easily filter or navigate with the grades and the scores and temperatures, and then you get a more custom advice on what to do next. Presumably, you could filter old leads on, let's say, hot, and then you've got suggested action for each of those.  
Me: And even using that  
Stan: Okay?  
Me: then to you could say you could pull out The structured data on each to say, which is like a demo, which is cut like an e like, reach out  
Stan: Yeah.  
Me: email, I can almost use a bunch of that. So you can see I could see is what  
Stan: Yeah. Yeah. You could you could, like,  
Me: ends up happening.  
Stan: ask her to write an email based on everything. You could, do a a bunch of stuff. Something we're gonna do also is, doesn't exist yet, but it will release, in the next few months. Is a way to, have reasoning columns that at when you ask it for a greater score, this kind of thing, you get the answer in a properly formatted way like this, but also you get a reasoning column to the right so that you can see, like, why it assigned this score or this grade. And you can sort of,  
Me: Yeah. Makes sense.  
Stan: while you check all the scoring that's that's been going on.  
Me: I was thinking with the with the start of the sheet, There's plenty of columns in here that I assume could just use GPT for work to actually populate itself But when it comes to individuals often that's using so so we had a a workshop demo where it was it was a hacked together Google Sheets add on that had to pull from, like, LinkedIn API using the rapid API or something. I can't remember exactly. To then pull individuals into a sheet like this. Have you seen any way that people using GPT for work to actually find individuals and even email addresses and things like that to like, those suggested actions are great, but then it's like, if I don't know the email,  
Stan: Yes.  
Me: What am I gonna do with it?  
Stan: So you can try with, the search the web tool, which is connected to the Perfectly API. Your mileage may vary. Like, it's it it works a decent amount of the time, but not always. Collect specific data like emails or names. But it it it can work. It can get you, like, halfway there or 75% the way there, and then the the rest, you might need to find another way. Or if you already have some other tool that works fine and use this other tool, to create the input data, and then you can use, you can use other tool. We are going to make it better, make it easier to fetch data from the web, with  
Me: Yeah. It could be adjusting if there's, like,  
Stan: Qdya for work. But for now, it's just complexity.  
Me: integrations or connectors you could add to GPT for work. So let's say if there was an API that I was using for linked for scraping LinkedIn, to be able to pull in. Yeah.  
Stan: Yeah.  
Me: Okay. I know that.  
Stan: The problem is scraping LinkedIn is notoriously  
Me: Yeah.  
Stan: difficult. They they yeah. So doing it  
Me: Yeah.  
Stan: at scale for thousands of users is, like, almost mission impossible. I prefer to not  
Me: Yeah.  
Stan: not really try, actually.  
Me: That's why I threw it to you, and I'm not gonna be working on it. So Yeah. Cool.  
Stan: Oh, yeah. I mean, that's that's pretty much it. Happy to answer more, more questions.  
Me: Sounds good.  
Stan: Or try some other use cases if you have some ideas.  
Me: Have has anyone in the audience got any questions? If you do, please raise your hand on Zoom, and I'll get you on to ask it live. We'll put it in the chat if you are unable to speak. Yes. We have one from Kristen.  
Stan: Hello? Can you hear me?  
Me: Yeah.  
Stan: Yes. Hi. This was this was wonderful. I've been in ecommerce for over 20 years, and do you imagine applying more advanced CRM analyses with this tool? Like, I can imagine doing some modeling with you know, based on past customer behavior, like, you know, frequency, recency, AOV, and then using the database and those parameters as a predictive model, you know, will this customer purchase in the next 12 months? That kind of thing. Are you thinking along those lines? So, at the moment, we're really focused on helping ecommerce people to to manage your, their catalog, to, format it properly, have proper product titles, descriptions, going to make it easier to, do all of that while respecting the different, constraints that, for example, the Amazon Marketplace or the Walmart marketplace, put on those, those Bites. We for now, have not planned to really go into analytics. We work on more on text or, let's say, unstructured data than structured data. Because, I mean, to our knowledge, we're not ecommerce specialist at all. We we sort of discover ecommerce through this tool. And there's already plenty of tools that that do analytics. Although, I'd be I'd be happy to discover, like, which are gaps in the offering that maybe we can do something like this. But but now we're really focused on, yeah, having people manage their catalog, clean it up, deduplicate it, enrich it, with, attributes,  
Me: Yeah. I wonder if there's a way to  
Stan: Right. Translating it, this kind of thing. Makes sense. Thank you.  
Me: using that Yeah. I was on mute.  
Stan: Thank you. You're on mute, Ben.  
Me: It might I wonder if there's a way if you have that data to put it into a sheet itself and have that sheet use a bunch of these like, add ons to create something that then you're referencing in another sheet actually, can you do that? Can you have references to different sheets within, like, different tabs of a sheet. I don't know what they're called. Ben we've got like, you've got your CRM enrichment on Ben's scoring qualified leads on the other, Can you reference different ones across  
Stan: Yep.  
Me: both?  
Stan: Yeah.  
Me: In your prompts.  
Stan: So you you you yeah. Yes and no. Mostly no. You cannot reference you cannot refer dynamic conflict is one that will be, taken row by row, because we think it's it's better to see what you're doing. If you start mixing things from that you don't see it, it becomes dangerous. You can, however, refer Teletic content like, what we what I show here in green, into into a prompt but I mean, that to to the to the original question, thing is once you have, let's say, I said that you have a product catalog with, descriptions, and you're missing columns with attributes. You can extract the attributes. And then with that, you can use, native tools like pivot tables You already start, analyzing what what you have. So that's Once you're in your spreadsheet, there is a ton of tools that are either native or come from other add ons that that you can use, to do all of that stuff. So what was hard until, I want to say, GPD for work is is rambling text inside your spreadsheet, but now you can. And so you can clean your data so that can be easily  
Me: Got it.  
Stan: sorted, aggregated, etcetera, with with other tools.  
Me: A question from Kashov is on our team. He has asked, does the tool allow you to use a few rows of the output column itself for in context examples. And I know you should sort of you just copy paste them somewhere else and use those as an input example. Right? But is there is there another way?  
Stan: Yeah. There's there's no other way. I mean, once you have some outputs, in theory, anything that's on the spreadsheet at any time can be used as input. Either you copy paste it as as you said, like, you can copy paste anything in your prompt. So if it exists in your project, you can just copy paste it. Or you want it to be a bit more dynamic, you can use, Sell actual cell references like this one. But I tried to do earlier with the Bites descriptions, it didn't work because I think my prompt was not good anyway. But  
Me: Okay.  
Stan: it should work. It really should work. Like, okay. Let's let's do something, like, really simple just to demonstrate it. Sorry. I'll just translate, this and then say p 4 2 Spanish. K. So p 4 is hot. Okay. Which is a bit stupid because this tool requires, a column to apply to. So I'll just sorry, I'll just put a column here that doesn't matter. Astridus. Okay. Actually, it doesn't work. Oh, yeah. May we may have a bug here. We may have. Value. Or maybe try this again. Oh, okay. Now it works. So the word hot can be translated to Spanish. And you can see the answer is the same all the time because p 4 is fixed. So it translates hot  
Me: Yeah.  
Stan: many times.  
Me: Yeah. I think it's  
Stan: Okay.  
Me: more like when you're often, if you're in a spreadsheet and you're referencing I think Bites is it for data validation that you've got a table of some things that you're using, and just have it's a case of just having that somewhere else outside of your main columns as your examples and then referencing that the prompt. Right?  
Stan: Yeah.  
Me: You're gonna do in context.  
Stan: Oh, you you can I mean, you  
Me: Yeah.  
Stan: here, I put just one cell, but you could actually put a a full range on it? So you could say alright. Let's do something a bit different. What can I do? Yeah. Let's let's let's do gg42 h5. E42h5. And we'll put it in n t That's that's the cool thing too with spreadsheet is you can just go to the right all the time via prompts for different things. Different operations, and keep all the intermediate results in your columns. That's something that only this spreadsheet is really well built for. It it's the most amazing software in the Stan to man manage columns or rows. So So there there you go. I used concluded, what did I do? G 4? So 3 of 3 of webinars, white paper download, one will be in our blog post Oh,  
Me: Yeah. It's just  
Stan: Perfect.  
Me: an example of a of a  
Stan: I would have to work on a better prompt. Bites it's just, like, I'm just  
Me: As of reference of multiple columns in a prompt. Yeah.  
Stan: yeah. If you are very savvy with references, you can use functions. This this will work perfectly. You can put a full range for example, I can say Okay. Translate this And here, I can just take Select it manually. And, had to put some commas.  
Me: Yeah.  
Stan: So if you want something super custom functions, are probably the way to go for now.  
Me: Cool. And then there's one more question which is do you have any use cases on for finance sometime is that, like, forecasting, for example?  
Stan: No.  
Me: Yeah.  
Stan: Short answer, no. We we don't work with numbers. What we do is is, arguably a bit weird. We put text inside a spreadsheet where spreadsheets were built for numbers. But yeah.  
Me: Yeah. Okay. Yeah. Because it seems like that might be similar to Kristen's question.  
Stan: Leave the numbers to, to itself.  
Me: As well where there's things you wanna forecast. There's Data Numbers The Yeah.  
Stan: Yeah. Excel is, like, the main use case for Excel is, of course, numbers and running them and forecasting them. We purposely didn't go there because we think, Microsoft Copilot is gonna do this. Actually, I mean, the re the recent demos show this. So they wanna steer clear of the I think they're gonna do this better than anyone else since they're they're Canadian native. But, yeah, we're just doing text.  
Me: Okay. Sounds good. Well yeah. Sam, thanks so much for your time and showing us this. I think yeah, there's there's plenty of use cases here that you do cover that couldn't do what would take you a lot a lot of copy and pasting. Otherwise, So, yeah, I really appreciate appreciate your time, and yeah, it's gbtforwork.com is where you can go to sign up. But, yeah, thanks for thanks for showing us. Yeah. Well, cheers, everyone. We will see you in the next one.  
Stan: Thank you very much for having me.  
Me: Bye.  

‍