Integrating Accounting Firms & AI: GPT & Zapier Central Use Case Ideas with Isaac Perdomo

Isaac. Hi there.

How are you doing?

I'm doing all right, Mark. Thanks for having me. It's an honor to be here, to be presenting a practical use case for your community.

It's great to have you here because I know that the stuff that you talk about that you're an expert in will really resonate with the people that come along to these sessions, which is all things AI, Automations, API, that kind of stuff.

I'm really excited by the live demo you're going to do because I'm going to learn as well. And for everyone watching, that's going to be in a little while time probably. We'll do a few questions for 10, 15, 20 minutes and then we'll get into the live demo. So let's start, Isaac, with a few questions before we get there. You might want to at some point when you've got a minute, when I'm asking questions, Isaac, just turn your screen share on the green room just so I can make sure it's working. It won't come onto the live stream just yet. Just so I make sure I've assigned that screen share. I think it'll work.

Perfect.

Okay, so Isaac, just to kind of kickstart things, what brought you to kind of this space? Tell us a bit about your past and what it is about your past that prompted you to get into this marketplace to serve and why in particular accounting firms. Just give us some background.

Yeah, more than happy to give you a little bit about my story. So, effectively I graduated from engineering. I was always a nerdy kid and liked technology. But funny enough, I did everything you're supposed to do in college and graduated at the top of my class and couldn't find a job. So I was quickly reinforced by reality that businesses don't really care about grades, but about your ability to do the job right. So I decided to actually just go on my own and I started a business. I figure I want to learn something that's going to be useful for the rest of my career. And at the time I started writing for a living and I did it specifically for the software industry because I know I wanted to learn about that, that world. And effectively I forced my the first three to four years of my career doing that, just getting better at it, until I realized, you know, I kind of like software more than I like writing. So a now partner of mine talked me into starting to join his agency and offer these automation services to other businesses. So that's what we did then. And over time we started to recognize patterns of who Got the most value out of the things we did. Who had more opportunities for automations in their businesses. And I've always had a personal interest in accounting. My actually my dad was a cpa. He used to own. He's from own business for is now retired actually. My mom worked in tech as well in like banking and other financial institutions. So I always had like that, you know, empathy and like exposure to all the manual work that needs to happen for an accounting function to be run in a business.

Well, cool stuff. Really interesting. You start off as a writer, as many people know. I've written lots of books myself but I hate writing. It's really hard. Yes, it is difficult. I still write occasionally but I prefer to figure stuff out and just think about stuff. And probably a bit like you with the automations. As many people know, I love the prompt engineering and figuring out what prompts get the best possible results.

It's funny because that's writing ultimately, right?

It is, but I think it's more than writing. I think it's kind of solving puzzles really. I was talking to somebody who lives close to me in Portugal here and he's into, you probably know, know this, he's into RPA's Robotic Process Automation. And we were talking, we had dinner on Friday and he was saying that the thing he loves about that is it's about solving puzzles. And I think that's where my brain goes. Prompt engineering to me is puzzling out what's the form of words that gets the best result which is more interesting than just sitting down and writing a book.

Yeah, it's kind of like at the edge of technology and writing. So I see why you like it a lot.

Let's stick with the start with the definition so people are clear because there are all sorts of definitions floating around. When you say automation.

How do you define automations?

Yeah, I think automation have different levels but for me it's effectively anything you're doing to make a process automatic in a firm. So the way we like to see it, we have like five different technologies usually involved in that definition. One of them is just apps that you use on a regular basis, off the shelf apps that you can buy. There are also integrations, which is usually the world we live in. A lot of like connecting one app with another, making sure that kind of communicates with each other. There's also sort of like data and data storage. There's often a needed component to make certain more advanced workflows operate properly. And there's also like robotic process automation which you can mention mentioned a little earlier, which is often where you don't have an integration capability but you still can mimic what the person, what a human would do in a computer of click here, click there, type here. So those.

Technologies combined with AI really are limitless things that you can potentially automate in a business.

Yeah. And when, if I go back 18 months ago, two years ago, thinking of automations, to me it was tools like make and zapier. Is it zapier or zapier? How do you pronounce it?

It's actually zapier because it's supposed to rhyme with happier.

Okay.

At least that's not what the company says.

Yeah, see I would think it has to have, it should have two P's in then. But anyway.

So anyway the question where was I going with that? So kind of that's where to me I thought of as automations with tools like make with zapier. So how has things changed in the last 12 to 18 months with AI? What's AI? How has AI changed things with the automation space?

I mean I think one of the biggest things about AI that's really changed the way automation professionals like us, it's really the possibilities and also all of the conditions that you used to have to manually code but now an AI can kind of figure that out for you. You have now like human intelligence almost that you can integrate into your workflows and we're going to be able to see an example. But before you had to say, you know, if this email had this text in the subject line versus now you can just say, hey, is this a complaint from a client or. Right. And you can now integrate those things so much more easily than you would have been able to otherwise.

Yeah, and I'm fascinated by the demo you're going to do because I've had a, I know you've told me some of the things you're going to do, I can't wait. And that'll I think bring it all alive. When we think about the things that you do building these automations and so on, what sort of firms are the best firms for this?

Well, we personally have found good success with sort of like small mid sized firms, typically, you know, small teams, five to 20 people usually have a lot of work to do. They usually have a combination of both compliance and advisory services. Of course there's usually more opportunities for automation than the compliance side, like tax prep, you know, building organizers, taking that data in, sending it to the tax software. And also there's a lot of opportunities. So like in the accounting and bookkeeping side as well, where it's much more about, you know, you know, coding transactions and classifying them and analyzing the data.

That's primarily who we kind of best serve.

So just to clarify, because I'm not sure so I can understand why the automations would work with things like the compliance work and bookkeeping and so on. But you mentioned business advisory. In what respect?

Well, part of our job sometimes is also to help clients build sort of like data and dashboards. So sometimes that those insights can at least inform the advice that some of our clients give. Right. And there's a lot of automation required to make the reporting happen in the first place that sometimes automation can kind of aid that particular process.

Yeah, absolutely. Makes sense. And one of the things I've talked about in the value pricing academy when I've talked about advisory is that there is a, there is an opportunity for firms to help their clients with some of these automations. With AI I teach prompt engineering, I talk about GPTs and I know you're going to talk about GPTs later. And I think there is a genuine opportunity for accounting firms through their advisory to help clients with some of this stuff.

Absolutely, yeah. And actually we've, I've seen, I know personally accounting firm owners that have offered tech consulting and automation as a service to their clients and can an additional revenue stream potential for people that can do it.

Right, Yeah, I remember seeing, I think it was about five years ago there was some research done by amongst the top four accounting firms and they predicted that the biggest growth area in the profession certainly for the big firms in their firms was going to be tech helping clients with tech automation which at the time smaller firms might have thought, well we can't do that, we don't have the expertise. But these days with things like ChatGPT, AI and so on, it's now something that's available for the smaller firms to take advantage of not just for their own within their firms, but also to help their clients with.

Yeah, definitely. And that's something that I feel like this low code and AI thing has enabled business users to actually provide what would traditionally been like a technical solution on their own with a little bit of help, some consulting help potentially as opposed to trying to rely on developers which used to be the thing.

Yeah, talk a little bit about. Obviously there's so many things you could automate but what are the best automations that we should put in place that's going to get a return return for the firms? Obviously there's an Investment in time, money and so on to make these automations. What sort of things should be think people thinking about automating?

Yeah, One of the best.

Resources and kind of like tools I found to answer that question is just going through the, the client's journey because typically if something has to happen for every client, it's something you'll have to do on a regular basis. So there's this concept, it's actually from a book called Systemology recommended read called the Critical Client Flow. And it's a super high level flow chart essentially of the steps a client has to go through to engage with your firm. So at the top it's like marketing and everything you do to gain attention. There's a lot of things in that world that you can automate. I'm sure you've talked about creating a client profile, your content, all of those things. Right from there you have the next three steps. Usually I see it as one bundle which is like your sales, your onboarding and your billing. So the process you go through to convert that attention into a client, there's usually a lot of administrative tasks around the sale that can and potentially should be automated. There's the onboarding process as well. When you're capturing that initial data from the client, potentially tax organizers or getting access a lot of administrative work that happens there as well. And then you have billing which is kind of automating the process of collecting payment which many softwares like the one I know you own, Mark, can do very, very efficiently and beautifully.

From there you have right marketing, sales, onboarding, billing, then you have sort of deliveries, the bigger, the biggest bucket of course because there you have like task management. How do you automatically potentially create your tasks so people know what to do, what and who needs to get the work done. But also you can start looking into the individual departments or service lines which might be your bookkeeping, your tax prep and see what are the kind of double clicking there and seeing what are which. One of those steps could potentially be streamlined or automated. Well.

The last step in that sort of like critical client flow, flowchart, it's called repeat and I think it's often overlooked but it's how do you now make sure that the client, I like to call it the three Rs which is either the client repeats or recurs, right. Stays with you, refers or reviews your business. So there's also some automations you can put in place there. So so that at the moment of maybe highest satisfaction, where you completed your first project, you can automatically prompt your clients to leave you a review, potentially refer you and have system in place as well, to make sure you are able to capitalize on potential opportunities, if there are any. So at least that kind of. I think it's a great framework to simplify kind of what are the typical processes we're looking at and trying to automate within our firm?

Okay, that's really useful.

I was going to ask you to confirm, but I think it's Dapo Depot has asked the question, can you repeat again the name of the book? I think it was Systemology, but yes.

Systemology by David Jennings.

Okay, awesome. Yeah, that's right. Confirming. So that flow that you talked about then, does that suggest that you, when deciding what to automate, do you start at the top then do you start with marketing things first? Is that kind of an order?

No, not necessarily in order because it will depend on the firm. Personally, I like to start with sales, onboarding and billing. And here's why. In marketing, you usually have multiple marketing channels. In other words, clients have multiple ways of finding out about you. In delivery, you also have multiple buckets, multiple services. So clients might choose different services. However, typically your sales, your onboarding, your billing processes are or should, in my opinion, be standardized across the board. So they're the same for every client. So more often than not, we starting there and then we're diving into either marketing or delivery, depending on the firm. But typically we like to start like in the middle and then we kind of branch out from there.

Okay, cool. That makes sense. Yeah. Interesting. I was just over the weekend, I sometimes can't stop with the AI stuff. I was writing a whole set of chat GPT, really long, detailed prompts for helping accounting firms either build their onboarding process or they would feed it their onboarding process and it will review it and give feedback. So it's one of the really powerful, important processes, systems, automated automations that every firm should definitely have on board. One of the first ones.

Yeah. Yeah. Well, whenever we're ready, I'm happy to start shooting the live demo. Mark. And I'm excited about it.

Okay, that's gonna be cool. So let's do it. Let's have a live demo and let me just see if we've got it on the screen. You probably just check. There we go.

Yep. Okay, let me know. You guys can see my screen on QuickBooks right now. And you guys. All right, looks like we're good.

I can see it. Yeah, I'm just going to leave you to it, but I'll come back in if I got any questions or whatever?

Yes. So high level process of what we're going to be doing. The goal of this particular automation is to automatically get and analyze the financial statements of your clients and tell you in a summary what the automation found. What's interesting about this, I know Mark, that you talk about using ChatGPT and custom GPTs, but here in this case we're going to do it actually with a GPT assistant which allows you to actually do that automatically without any human input. So you don't even have to go to ChatGPT to actually give it the data. Okay, so four simple steps to get this done. The number one, it's getting the data and getting the financial statements. We're going to see how to do that within QuickBooks with actually the built in feature. Number two, we're going to be creating the prompt for the AI. We're going to be looking at the kind of platforms OpenAI interface and how you create an assistant in case some of you guys are not familiar. Number three, we want to be looking how to automate that prompt within the sapier environment. The SAP your tool we're going to be using kind of like to put it all together. And number four, it's sort of like after the data, what do you do with it? In this case we're going to put all that data in a spreadsheet. But you could technically take those findings the AI gave you and put on your task management system, whether it's Trello, Asana, Notion, Carbon, whatever it might be, for simple demonstration purposes, I'll be doing it in a spreadsheet. Let's go back to step number one, getting the financial statements. This is a demo account of QuickBooks, but if you go to reports, actually you're able to schedule a report to go to an email on a regular basis. And the way you do that, for example, let's just say I want a PNL by month. Usually I like to do this year to last month to make sure you don't have kind of partial data for a particular month. And if I save this.

I should be able to then come back to reports.

And now here you see email on schedule. Oops, sorry about that.

Here's where if you click on Edit, you're able to set an email schedule. So in this particular case I'm working with a sample company, I don't have that feature enabled. But whatever report you want to do or want to analyze automatically you Simply go to QuickBooks, run it, save it, click on Edit within custom Reports and Then you'll be able to send that to an email. What happens from there? We're going to be now using Zapier to process that email, send it to ChatGPT, and then doing what we want from there. So let's go to Zapier. This is Zapier's interface and I'll just simply click on the Create button. Create Zap, just one of their automated workflows. And this is Zapier's visual interface where you'll be able to create stats step by step. What you want the automation to do. In this case, we are ready set here that we're going to be using an email specifically to get the data. And here you can see that I'm able to actually say when a new inbound email comes in. And I'm able here to specify the email address that Zapier creates automatically to process it. And this is the email I will use kind of within the QuickBooks environment. So here I'll just put Automatic Review as an example. I'll click Continue and let me go ahead and send an email to that email address as a way for me to test if Zapier is actually receiving that. What I'm going to do here is I'm going to be attaching that P and L and I'm kind of here simulating what QuickBooks would have done automatically. Okay, so hopefully people are following up until now. I am sending an email and I want to see if Zapier receives it in this trigger that I just created. It usually takes about 30 seconds before I can see it come in. But it's important that I start testing step by step to see that everything is running correctly. Here I can see email August 19, 1124 from Isaacpser Co and all of this other data that we're going to be using later on.

So step one, done. We are now getting the data automatically by setting up a schedule report in QuickBooks and setting up this triggering Zapier. Now we want to run this through.

ChatGPT.

I know, Mark, you talk a lot about GPT and I mentioned this a while back, but within GPT you have like different levels as well, right? You have the ChatGPT environment, which everyone can use and it's more generic and you have to kind of write your prompt all the time from there. If you kind of graduate to a custom GPT that's more specialized and you can have a prompt, but you still have to go there as a human to give it the data or interact with it. The next level at least in my view, it's an assistant that's very similar to a custom GPT. You just happen to also be able to call it automatically within zapier. So this is a assistant that I went ahead and created. And here you can see the prompt. I'm just telling it, hey, you are an experienced accountant reviewing financial statements. Please analyze the attached income statement and provides a concise report of any anomalies that need investigation. Here you could potentially give it a little bit more details of what specifically you as an accountant look at, so that it kind of captures what you would have captured on your own. Right, what I proceed to tell them, then it's effectively the format in which I wanted to give me the result, which is effectively a way for the computer to better process this particular format. So you can see here effectively is a series of strings of the different headers I want and then the different rows. The reason I wanted that this way is to then populate it into this other spreadsheet in that same format. So I already have this assistant created. If you guys wanted to follow along, you have to go to platform.OpenAI.com click on the playground button, and within Assistant, you'll be able to create an assistant, give you the instructions. Here we're going to make sure the file search is enabled because we're going to be given the attachment that came from the email. We want to make sure it has that functionality built in.

So now that we have this reviewer demo assistant set up, if I go back to Zapier as step two, now I want to look for the ChatGPT app and as an event, I'm going to look for conversation with Assistant. So rather than being a generic conversation with chatgpt, I want it to be specifically, as you can see here with the assistant I just created and what I wanted to do, it's to process this financial statement. I also want to give it the attachment that came from the email. Every time we're automating something in something like zapier, we want to always make sure we are passing on the data that came from previous steps if needed. If I type here attachment, you can see it tells you it exists but not shown. I'm just going to put that there in the new files and I'll be able to test this step to see if it works correctly.

Mark, I don't know if you have any questions at this point. Anything you want me to clarify for the audience.

I'm just taking lots of notes because.

I'm in learning mode. I probably will have questions, but it's a good time to remind people watching if there's any questions you've got, remember, stick them into the comments and we will get to your questions. Because, because for some people this is familiar stuff. For others it might be. Wow, what was that step again? So far I think I'm kind of following along, but it's a good reminder for other people to think of their questions.

Perfect. Perfect. Well, if you learn something, Mark, this is the mark of success for my webinar because I know you are the GPT guide in the accounting world, so that would be my I've never used.

Success, I've never used Zapier, so the zip I've seen Z, but I've never actually used it and perhaps one day I should learn to use it.

Nice, nice. All right, well the output has come up. There's a lot of things here, but I already know the thing that I'm interested in is this response. Okay, so this is the response from the ChatGPT that came up from the PDF I gave it and this is what is returning back and what it found. Now this format is not as readable to a human being, but this makes it very easy for me to format it appropriately to then put it on the spreadsheet that we have now. I'm going to go a little quickly here.

To format that response in the way I want and we're going to be using some built in features within Zapier as well as a little bit of code to process that accordingly. So within Zapier I have an app called Formatter by Zapier that allows me to do certain things within for a particular, you know, string of text. As you saw in the full response that came through, I have what I want, but I also have this like triple back text and JSON that Zapier provides kind of tells me that that's in that format, but I don't want that. I'm not going to use that. So I kind of want to remove those things. In order to remove those things, I'm simply going to use the replace function within the text formatter functions of Zapier as usual. And as I said earlier, we always want to take.

The data from steps above to process those.

Here I have the response that came through here telling me, hey, what do you want to find and what do you want to replace it with? I know there are here some triple back text that I want to find and here it's telling me, hey, leave it blank to delete the Found text. So if I test that out, you can see that now I have the same output but without the triple back text. So I'm going to go ahead and do the same for that JSON text that I also don't want.

Let's go here.

Take the output from the third step, find JSON and delete that from that text. Lastly, I still have some space there that I want to remove. To make sure that doesn't create an issue later on.

I'm going to use another function that I have available within Zapier that's called trimming, let's see.

Trim, white space. So what this does, it simply takes up a value, deletes any space that might be before or after it. And it's something we often want to do to avoid issues processing this later on. So these three steps are effectively removing those triple backticks, any space and that JSON text that we have there. Lastly, I'm going to be using a little bit of code that I already came prepared with. So I'm not going to be coding in front of you guys. But the whole goal is to turn that format into a format that I can see much more clearly, that Sapier can understand much more clearly. We're going to be looking at here taking what came out of step four, five.

Again here, a code that I already had prepared. And I'm happy to share this with the audience after this.

And here's where selective magic comes in, where we'll be able to see in a much more human readable way.

The result that ChatGPT gave us above. So here you can see, okay, we have this particular account, there was a significant drop of income this month.

And it's telling me, hey, prior, in the month prior we had this negative amount, now we have zero, no discount we're given this month. It's giving me a lot of these, a lot of its findings in this case in this format. So last but not least, I want to put that into this spreadsheet and I want that after I publish the automation to happen automatically once Zapier receives the email. We've run through all of this. Okay? So in order to do that, I don't know how many findings it's going to deliver.

And we're going to use another built in Zapier function for that. That's called looping. You can see looping simply as iterating for each thing that you gave it. So, so in this case I know that I have a number of findings of account names and categories. So I'M going to have and simply kind of map those fields. Remember that I have here account name, category, current balance, prior balance finding. And then I'm just asking it to tell me if it's positive or negative. So I'm going to go ahead and simply map that out real quick.

So if there are any questions in the meantime, I'll be happy to hear them. Just take me a second to get this done.

Yeah, while Isaac's doing that, think about your questions, stick them in the comments. I've got a few questions myself. I'm just making a note for later.

Yes, perfect. So I'm just mapping out the account name variable in this case to what it delivered so that then I'm able to put that into the spreadsheet and we're going to see that in a second.

Finding.

Sentiment. Okay. So I'm going to be testing that out and again you can see like a little kind of grayer area here. And this is sort of like zapier telling you, hey, for each one of those items that you gave me, I'm going to run the following step. In this case, what I want is go to Google Sheets.

And I want to create.

A spreadsheet row.

Here. I'm going to select my account.

And I'm going to select specifically the reviewer demo spreadsheet, which is what I'm using here as the demo, the first sheet. And you see that it automatically tells me, hey, here are the current headers within that spreadsheet. And I simply have to, as a last step, populate those from the seventh step and we should be done.

So let's map that out.

Current balance.

Prior balance.

Finding.

Then.

The sentiment.

We're actually able to test once to see if that actually goes through and populates the, the spreadsheet. We can see it here. It's telling me, hey, this particular income account, it's called Design Income, had a significant drop this month. It's kind of not a positive thing, maybe something you would want to investigate for your clients. So I'm going ahead and publish this because now we're effectively done. And the idea here is that every time that that email address receives an attachment with any statement or your client statement is going to go through Ping chatgpt, format it accordingly and add it to the spreadsheet and we can test that right now. So let me go ahead and pull this up.

Automatically. Was it again? One second.

Okay, so automatic reviews. I'm gonna take here a sample PDF of a profit and loss statement.

Say testing live.

Let's just Name this accordingly. Automatic Reviews.

Demo Life.

And I'm going to send that out.

Now by me sending that out. You can see here that this app is actually checks every two minutes. It's going to take within two minutes for it to go through automatically and add things to this spreadsheet. But in the meantime, Mark, you're free to come in and share your thoughts and.

Yeah, we can keep the conversation going.

Yeah. So just so I'm clear, so we will see in the next few minutes this spreadsheet is going to be populating.

Yeah, Correct. It's going to be populated with the findings of that particular PDF. And the idea here, you can see how it starts coming up now without my involvement. The idea here is that if you could do this for Argo, clients literally have right now a client that's doing this and they have like 60 clients on the roster. And this is what they wanted to do. Right. They want to make sure that rather than they having to go to their clients financial to try to find the issues, trying to see if having them run through AI is able to at least bring up those issues to the team so that they can investigate further. And here you can see the things it found. Of course, I think that one thing I should say is that you can update the prompt over time and you don't need to update the automations run through the new prompt and you can through prompt engineering make sure that the outcome or the output that AI is giving you, it's kind of better and better over time. But here you can see a couple of examples. Hey, this were a significant increase in this particular fuel expenses account.

So you can start.

Looking into these issues potential for your client and see what's going on.

Yeah. So in that Excel spreadsheet, the columns that you've got and what it's doing is driven by what you've put in the prompt.

Correct.

So in the prompt I told it to give me those things in a specific format. I map them in Zapier to make sure it align with those columns. Now that's happening automatically without my input moving forward.

Yeah, which opens up endless possibilities. But you made the points again for clarification, you said that you can change the prompt in the GPT, you can change the prompt and you don't need to change Zapier. But would I be right in saying that if you in the prompt though decide to ask for some different categories columns, you would have to, then you would have to, because you'd have to map those extra columns.

Correct. If that's the case, then you will want to add those in. Those would be kind of a small little update as this automation to add those in again.

Great. Okay, that makes sense. Yeah.

I've still got more questions, but I'm going to give some priority to a couple of questions that have come in, so go ahead. Teresa said this can only be done in ChatGPT or can you use Gemini or Spark?

You can use other models as well. I chose to use ChatGPT because that's what we pay for, we have access to. But you could use any other AI model and for that matter, any other accounting software and also any other either spreadsheet software or whatever it is that you want the data to go to. As I mentioned earlier, we have a client with this use case right now and we're adding them to their task management system for the team to look into there. So you have flexibility into the AI model, the accounting software and the ultimate action that you want.

Yeah. Just clarify for me if you could, Isaac, what's the reason why. So you're using a GPT as an assistant, using the playground, which is platformai.com.

Why are you using the assistant in the playground? What does that give you that a normal GPT wouldn't? In this case, yes.

So again, the idea is trying to think of it as at scale. Let's imagine Again, you have 50 clients and you wanted to run a prompt like this in a custom GPT. Even though the GPT will function appropriately, give you the insights, you still have to go there, give it the PDF.

And, you know, get the data from it. So you cannot call a custom GPT from Zapier, for example.

Yes, got it. Yeah. Yeah.

So in this case, we're building an assistant. An assistant is like a custom GPT, but not only it's not only built for a human being, it's also built for another program to call, as we saw here with conversation with Assistant, with this step number two in the automation we built.

So when.

When we go into the. So when we go into the playground.

Is. Can you think of it as. Essentially what you're doing is you're using a GPT, but in the playground you're extra functionality which you need to have to be able to do this sort of automation. Is that a good summary?

Right, correct.

Yeah, yeah. Because I saw when you went through the playground, a lot of the fields were similar to what you have in a GPT, but there was the extra bits as well.

You have a little bit more flexibility. Yeah. To some of the. You Know.

Yeah, yeah. And the important one in this case is the file search, is that right?

Correct. The file search allows to interpret can I download that file that I gave it within Zapier. And of course, the instructions, the promise. Also one of the most important things that you have to nail too, for it to give it your. The right, the right thing.

So for anybody watching who has no idea or never looked at the playground, it might be scary to people. Could you just kind of summarize a bit more in kind of simple terms, what is the difference between what is the OpenAI's playground as opposed to just going into the normal ChatGPT and GPTs?

Yeah, that's all right. Okay, so again, you have to think about what are these things built for? In the case of an assistant, an assistant is built not only for a human, but also for another program. So if you look at the normal chatgpt that you go in and type things in that particular GPT, it's.

First generic, it's not specialized. You have to type the front end every time. And that's usually where you want to kind of graduate to a custom GPT. A custom GPT is specialized, but it's still built for a human, and you can't really go in and automate it. So when you go into an assistant, specifically, you want both a specialized prompt, a specialized GPT, but you also want the ability to ping it automatically from another program, like we did with Zapier. So that's really where you would want to use assistance is when you want to graduate from custom GPT, you want to start, stop having to go in and manually give it the data to provide you with the result. And you just want that to happen without any human input at all.

Okay, great. Presumably, therefore, even with the GPT, you couldn't do the same thing with a GPT using actions and APIs.

Well, it depends on what are you trying to automate. Is it the trigger or the action? For example, here we want to automate the trigger, meaning I want just to send that email that comes directly to the GPT. Okay, now, in a custom GPT, I cannot do that. What I could do is to say, hey, when I tell you this, go ahead and maybe change something or send an email that can be done within a custom GPT. I can automate the actions, but I cannot automate. So like the trigger, I cannot automate it to.

I have to trigger myself as a human being, if that makes sense.

Yeah. Okay, great. Let me move on to some other people's questions. I'm just hogging it all at the moment and being rude. So Liz said if you have different automations for the same client, you use two separate emails and make the email addresses in Zapier.

Actually, that's where I would probably look into.

Adding another column to the spreadsheet so that then you can specify what client it is and you can actually change the prompt. Because if we look at that PDF that we had as an example, you'll notice that it says the name at the top by default, the report. So if I change the prompt, I tell it, hey, actually give me the client name as well, which is at the top of the PDF. I'll be able to have that in and I'll be able to just, with the same assistant and the same automation, be able to now run this for all clients. And I'll just have an additional column here that's going to say, hey, this is correct landscape design services. And then you'll see that not only for that client, but potentially for other clients as well.

So that's how I think about trying to scale it to do it, you know, across your clientele. Hope that helps.

Great answer. Thank you. Then we have a question from Karen which I think is about security, so I'll be interested what your views are. Karen says, do you delete the customer name before emailing it for security protection, before including it in ChatGPT? I know we had a conversation with our previous guests about security and. Yes, and what information you can safely send to chatgpt.

Yeah, I know that that's a big thing, especially given that GPT could potentially be training your data with it. However, they are being open about that. Any API call or assistance call, they don't use their models to. They don't train their models on that data.

So a couple of ways you could go about it. You could potentially try to work with client IDs to try to anonymize the data if you want it, but given that this is not trained in the model, I wouldn't be as concerned. But just caveat, I'm not cybersecurity expert either, so that's a little bit outside of our wheelhouse as well. We're kind of aware of best practices, but you might want to kind of consult with the appropriate professional there.

Yeah, that's a great answer. It's kind of my advice as well. And our last guest speaker, Igor Pagani, said effectively the same thing that these, the likes of OpenAI have said, that they're not going to train data. And so I have a fairly bullish approach when I want to analyze my own financial information. I just put it in there because I'm quite bullish about this stuff. But I know that with accounting firms and client confidentiality, I think it's an issue that each individual firm needs to have its own AI policy around this and different firms will choose to do different things.

And I think that's the big part. Making sure you are clear internally how you use it, that you're clear with the client as well as to if you use it or not and how. And I think those two things are probably more important than that. Whether this is going to be trained on your data or not. It's kind of clarifying the expectations with the people in the firm and the people that interact with it.

Yeah, absolutely. Okay. Going back to Liz again, she asked. I think you said this earlier, but just for Liz's benefit. How do we access the playground?

Great. So the way I do it. Platform.OpenAI.com and that will get you there. One thing to note, this is kind of like their, I guess, developer side of things. This is actually built separately from your chat GPT subscription, which I think it's important to, to realize. So you're charged like often like cents for each prompt you do. So you would have to kind of set up the payment method separately in order to be able to, to. To use this. But again, platform.OpenAI.com and just sign up for your account.

Yeah, Just to clarify, yes, you do get charged for every single prompt that you feed it, but it really is a small amount. You'd have to use the platform really heavily to be racking up $20 a month. So it's, it's definitely worth considering using it.

You can even look at the pricing if people are curious. This is literally like.

Now they have it like per million tokens. And the token, it's kind of like a very. Yeah, definitely.

For people who don't know, a token is approximately the equivalent of 0.75 a word. So a million tokens is about 750,000 words, which is quite a lot.

Yeah, definitely.

Yeah. Okay. I haven't got any more questions from other people.

We've got about five or six minutes left. I do want to ask. I will ask you in a bit to share how people. For you to share how people can find out more about you. We'll do that in a second. But a couple things I've got just for people because nobody else asks, but I thought it'd be one that people were Worried about can you define what is JSON, which I know is J S O N? What is JSON and why do we have to then use that formatter?

Jason, you can see it as a format. It's effectively what it is. Just like you have CSV, which is a format that many people. You can't hear me.

Yeah. Or markdown.

Exactly.

So it's just a format. I don't want to get into the weeds and the details, but it's a format that's used a lot into the APIs and software world for you to be able to send data in a structured way.

It's how effectively Sapio interprets things and allows you to see it in a much more human readable format.

That's why we use it.

Yeah, great. The bit, I think that probably was scary to most people. Certainly to me, the only bit I couldn't do would be, I think it was the step where you then put in some code which you kindly said you'll share and we'll put it in the description for people for this particular thing. But going forward, I guess what people need to do is unless they understand the coding, I guess their options are. If they don't, sorry, they don't understand the coding, one option is to go to you and we'll talk about that in a second. Presumably ChatGPT could help with that, could it? Because it's very, very good. It's code interpreter would help you code.

Yeah, definitely. You can literally say, hey, I have this in this particular format, I want to get it to this because I want to do this and this other thing. You can start, play with it and you can even, hey, walk me step by step how to get this up and running in Zapier and we start spitting things out. It doesn't always get it right, but if you really have the intention to do it yourself and you have the passion and the patience, then definitely can be done and can help you.

Great. Right, just two more questions. Penultimate question is from Meryl. Last one's from me. Meryl says, as we get started with Zapier, what level would you recommend for the level of automation that you demonstrated?

What level would you. I'm not sure I understand the question. If it's what's the level of the automation I demonstrated versus.

I don't know.

But presumably it means are there different subscription levels for Zapier?

Oh, subscription levels. If it's pricing and plans, you can start just on the lower plan. Often you'll have all the features you want there to Start with and just upgrade as you go as you need more often. If you want to do something like this, like has more than three steps and has certain, certain applications, you might have to go to the kind of second app, second plan. But I think it's like 50 bucks a month or 20 bucks a month. So.

It'S not pricey for what it delivers. And they've actually done many pricing changes recently to try to make it a little bit more affordable for people.

Okay, great advice. So my last question, assuming we run out of time, Isaac, is so tell people if they want to find out a bit more, how can they reach out to you? How do you help people? And what links can we share with people?

Yeah, I'll be more than happy to leave a link to our website, observe. And also I'd be happy to share some of those resources that we shared today. It's something I'll have to kind of get up and running, but we could have like an observer co mark that links to the prompt and the code that we use today and even a template of the zap for people to play on their own if they want it. I'm also someone active on LinkedIn, so if you go for Isaac Perdomo, you'll find me there and be happy to connect.

Okay, that's awesome. I don't know whether you send any links to Sarah or not, but if you did, she will put them in the description. And if there are any links you want to share, any further links you want to share, Isaac, we'll put them in the description against the YouTube, LinkedIn and wherever else. In fact, I can see already that Sarah's. There's one link already, so if you want a free consultation.

Then you can go there.

That was awesome, Isaac. I really enjoyed that.

Yeah. Did you like it? I think you learned a little bit, but I'm. Your gesture was a little bit, kind of like confusing a little bit. I don't know if it was kind of thinking a lot.

Yeah, I find it fascinating.

The only reason I look confused is because I don't actually use Zapier. It's been on my list to do a few times. Some of my team use it and I know many accounting firms use it and love it. And so I think that if you've played with Zapier, which I haven't, I would guess a lot of what you showed is actually makes a lot of sense. But. But it's also, for me, it's opened up my eyes as to practical applications of not just Zapier. The bit that I found most interesting was, which I'd not really thought about too much, was this whole way of integrating from an Excel spreadsheet into a GPT and then putting it into Zapier to do some stuff. The GPT bit I find easy because I love prompt engineering. It's the Zapier bit that was the step I'd never really thought of before. So I've learned a load. I love it and I want to have a, I want to have a look at Zapier now.

And what I will do is I'll ask everybody else if there's something here that you've really enjoyed. If you could just give the thumbs up. And then also, as we play out with some music in a little while, just put in the chat. What was the, what's the most interesting thing from this session from your point of view? What did you, what did you learn specifically? And I'll be looking at that as we play out with music.

So Isaac, that was awesome. I loved it. I would love to at some point do another session at some point in the future and talk to you more about these automations because they are fascinating.

I'd love to have you on our channel too. I think that would be an interesting interaction as well.

That'd be cool. Isaac. Okay, thank you so much and I will catch up with you some other time. Isaac and I'll I've got no other final comments to make to everybody here, but if what I would love it if you could just put in the comments. What's the, what's the thing that you've learned the most? What's been most interesting? And look out for.

Monday email next week because my next or next week? Next week I'm doing my three hour AI Virtual summit. So if you've not been to that before, I'm going to be teaching AI ChatGPT for 33 hours next week. We're giving tickets away for free and all the details are in in our Monday email newsletter. So we shared the link earlier. Sarah might put the link in again. And my next live stream, my next live stream here I think is on the 27th of August and on the 27th of August I'm going to share with you and show you how you can create prompts, prompts for chat, GPT or whatever LLM you use. I'll show you how you can create prompts where they become templates, which means you can reuse them over and over again for different circumstances, different clients. I love templatizing. If That's a word. Templatizing prompts and I'll be showing you that on my live stream on August the 27th. I will hopefully see you there. As always, I will play out with some music. It gives me a chance to read the comments and I will see you online soon. Bye for now.