Adam Tzagournis – Avoiding Frankenstein FP&A with FlowCog + ChatGPT

Adam Tzagournis – Avoiding Frankenstein FP&A with FlowCog + ChatGPT

Episode - 016: - Where we learn about Adam’s move from finance into software development, his view on low-code tools, why you don’t need a finance background to be a good business partner, AI autosuggestion, how Adam’s building ChatGPT into his tools, the future of FP&A AI predictions and much much more.

Adam is the founder of FlowCog who make SaaS financial projections easy. Adam is a developer himself, and writes about everything SAAS related, modelling metrics, FP&A and more. And in his spare time, he likes to spend time with his wife and 15 month old and has been studying Brazilian Jiu-Jitsu with his wife for the past three to four years. So watch out everyone.

Audio Podcast Links

Show Notes

Where to find out more about Adam

Tech Mentioned

[Note] Tech for finance are a Notion affiliate, but we also use and pay for their services


[00:02:07] Adam’s move from finance into software development

[00:08:46] The evolution of low-code tools

[00:17:38] Not needing a finance background to be a good business partner.

[00:23:42] Finance + Sales & finding the right customer fit

[00:26:07] Why Saas companies?

[00:31:57] Auto-suggesting industry specific accounts structures

[00:39:11] AI Autosuggestion + future user interfaces

[00:45:57] Building ChatGPT into FlowCog

[00:53:08] How long before AI can make predictions from external variables.

[00:59:02] AI seeing context vs prompt engineering

[01:01:25] Failing building ChatGPT into Google Docs

[01:07:14] Adam’s favourite tools



[00:00:00] Adam T: And so you have this kind of Frankenstein, piece of software that you're building and, and trying to maintain. And that inevitably will lead to, errors in the structure and the integrity of the spreadsheet, which are gonna give results that are misleading when you go to interpret them.

It's, really interesting this kind of concept of auto complete or have an AI guess the next word or the next action, that should be done in the form of a suggestion.

Let's say that, you're a manufacturing company and you do a lot of business at a port and that port gets flooded. The AI that you're using should be able to deliver a detailed explanation of this cascading set of effects.


[00:00:45] Adam S: Hello and welcome to Tech for Finance, where we help finance professionals leverage technology to level up their lives. I'm your host, Adam Shilton, and in this episode we're chatting with Adam Tzagournis, founder and CEO at FlowCog, who make SaaS financial projections easy. After starting his career as an associate at PWC, Adam went on to become a senior accountant at Halo, then the finance director for Stack Overflow, who are the largest online developer community before then founding FlowCog.

Adam is also a developer himself, and writes about everything SAAS related, modelling metrics, FP&A and more. And in his spare time, he likes to spend time with his wife and 15 month old and has been studying Brazilian Jiu-Jitsu with his wife for the past three to four years. So watch out everyone.

Before we start, if you like what you hear, what's, if you like what you hear today, please make sure to subscribe to Tech for Finance on your favourite podcast platform and on YouTube.

So thanks for joining me today, Adam. It's great to have you on.

[00:01:47] Adam T: Yeah, great to meet you.

[00:01:49] Adam S: Fabulous. So over to you then. So tell us a a little bit more about where the inspiration from FlowCog came, came from. I've not heard of a huge amount of people that used to be finance directors, but then go on to be developers and, and develop sort of SAS tech companies.

So do you want us to tell us a little bit more about your, your journey?

[00:02:07] Adam T: Yeah, I was kind of scratching my own itch over the years. I had taught myself programming in various capacities, even as far back as, when I was at PWC and we're in busy season and the audit team is there late at night. If you can find a way to save everyone, even a little bit of time, ev that's, you're doing a lot for your team.

E everyone gets a little bit more sleep that night. So I started, creating macros in Excel and Visual Basic. And from there, over the years, I kind of snowballed into me teaching myself, Python and JavaScript and, programming language called R, which is used with, statistics. So it's kind of, an unexpected journey or path to becoming a developer.

And along the way I had, various side projects and, even in, during my kind of, full-time roles throughout my career, I had always done consulting on the side. And a lot of it was financial modelling. And most of my clients were SaaS companies, a few manufacturing companies here and there, but for the most part SaaS.

And I very quickly realized that. Whenever I had either inherited a model or was asked to create one from scratch, there were a few problems with it, right? One problem is that whenever you inherit a financial model and you start digging through what the previous person did, chances are that you're gonna find either some errors in it, some hard coded numbers, some circular references, formula errors.

I saw this interesting stat that said that 88% of spreadsheets have errors in them. Mm-hmm. . and so that was my experience every time I inherited a financial model. And the kind of other problem with it was, even if there weren't any explicit errors with it, it was hard to get insight from the spreadsheet.

It was hard to actually get something out of it, to, to really kind of, Have it be actionable and to, to help in strategic decision making. Right. And so I realized that, okay, well, the problem is that if it's that difficult and there's not much, the, the fruits of your labor are not that valuable and it's clunky to update, you're probably not gonna go through that exercise very often.

And then what ends up happening is someone builds a financial model, they don't revisit it for six months, a year, what have you. And the projections are so stale that they're basically irrelevant and you're not using it to make, decisions. And you're kind of, you're kind of flying, in the dark there.

So I very quickly realized, okay, I'm encountering these same problems over and over as a consultant and I've done enough, development work at that point. And I kind of realized, okay, let me try my hand at running a startup and, and seeing if I can build a tool that makes it a lot easier for people to do financial modelling and actually gives them more insight than a spreadsheet would.

And, a couple kind of challenges that I had come across with that was, one of them was that, at first the product itself was a spreadsheet with kind of a software layer in it. But when you go to set up a product like that for someone and they see a spreadsheet that they didn't build, now you're asking them to inherit a spreadsheet.

And if the person hasn't spent their entire careers in spreadsheets and they see more than a few tabs, it's, it's a challenge. And so, Instead of having someone, either rebuild the logic from scratch or have, have to understand all of it, I figured, okay, what is the kind of easiest and quickest way to get someone up and running with a financial model and to actually get insights from it, right?

Because that's the, the big picture that I think a lot of folks, when they go through the exercise, regardless of what tool they use, they sometimes get lost in, in the details and forget the overarching goal of, actually getting insight and interpreting their projections to help them make decisions or to help, that finance person support the founder, the c e o and them in making decisions for, how to best grow the company in, in an efficient way.

[00:06:52] Adam S: It's interesting, and I, and I think the, the, the relevant point here is not, not just, you know, Making it easier than a spreadsheet to build a forecast. Appreciate there's a, there's a big business case for that, right? But we're also talking about pace as well, and, and speed. Because if we look at the way that things can change, whether it's a global pandemic or a recession or fill in the gap, it's almost every other month now that we are having to, to re-review what, what we are looking at.

Right? And, and we don't want to have to unpick a spreadsheet that took six months to build, to be able to make a quick decision based on, on a variable. I guess a slight tangent, the question I have for you on, on moving into becoming a developer, do you think working at Stack Overflow helped you there, or do you think you would've done it even if you hadn't worked at Stack Overflow?

[00:07:50] Adam T: I think I would've done it even if I hadn't worked at Stack Overflow. Obviously, I love the product. It's, it's, a fantastic community that's been super helpful in my own journey as a developer. It is encouraging to be around other extremely intelligent folks that, that worked there. And that encouraged me to, to, and, and help nurture that curiosity that I had and, and to continue kind of building out that skillset.

And it was very helpful in, in, various capacities as well, in terms of being able to retrieve data myself and analyze it and, present it and kind of pipe it into a, a consumable format for, for other folks as well. So, yeah, I, I do think it, it certainly helped, but not strictly a requirement to becoming a developer.

There's, there's a lot of, different paths to get there, so there's, there's not one right way .

[00:08:46] Adam S: Very good. Very good. And, and you mentioned a point previously as well about. You know, not, not everybody having the experience or the, the amount of time building models in Excel to be able to interpret somebody else's Excel model.

There's, I'm, I'm, I'm hearing more and more now about low-code. Mm-hmm. , you know, and that, that whole principle of in theory, leveling the playing field, you know, so, so, so my 30 years developing models in Excel shouldn't. Somebody who's maybe more junior in a role, being able to generate similar results just with a, with I guess an, an easier to use platform.

Right? The trade off is that, you know, platforms are gonna cost money and maybe people with 30 years worth of Excel experience, you know, ex excel's free. Right. But if it's not accurate, then there's a, there's a question mark over how free it actually is if it's not saving your time and, and money, I guess, at the end of the day.

So do you, do you think we're gonna see more of this sort of low-code approach given sort of current skills shortages and all of that sort of stuff?

[00:09:52] Adam T: Yeah, it's, it's a great question and kind of the, the problem that I see is if financial modelling is not your full-time job, if you ha, if you're a finance person at a, at a software company or at any company, you have tasks and, responsibilities aside from just updating and maintaining and, and building the model.

And that means that you don't have a thousand hours of free time to be able to put into a model. And the problem ends up being that. A spreadsheet is in reality, just a piece of software. It's kind of like a, a watered down programming language.

And so if you're not thinking like a software architect, when you build a financial model, you're probably gonna get a product or a result that's not easy to update. It's, there might be technical debt in it, right? That's a concept from software development where you built a part of the, the spreadsheet to apply it to spreadsheets, a while ago.

And now that that module or that sheet or tab in the spreadsheet is no longer relevant or doesn't fit well or flow well into the rest of the model. And so you have this kind of Frankenstein, piece of software that you're building and, and trying to maintain. But because you didn't approach it from, you know, first principles of software development, you're on a shaky foundation, right?

And that inevitably will lead to, errors in the structure and the integrity of the spreadsheet, which are ultimately gonna give, not only no results to interpret, but even worse, they're gonna give results that are misleading when you go to interpret them. And that's kind of the, the big danger in it.

So I do think that instead of asking everyone to become experts in spreadsheets, just like you wouldn't ask everyone to become experts in writing their own accounting software every time they needed to go, you know, record a journal entry, you should have a tool and doesn't necessarily need to be my tool or, any tool in particular.

It could be a spreadsheet that has this kind of low-code, layer in it where the building blocks that you're using are a lot easier, to, number one, to build and, and implement, but also to interpret if someone is, is handing that, that product off to you. And so I think those are the kind of big issues that I see.

How existing workflows, are in, in companies where you're asking someone that doesn't have a thousand hours to go and build basically a software product, right? and that's, that's an unfair ask for them. So it's no surprise that as a consultant, when I go in and I, I look at, a financial model, I see, either explicit errors or a spreadsheet that's just not very user friendly and, and difficult to use.

But yes, I, I do think that eventually, with low-code you can abstract away some of the more challenging aspects of building out something like a financial model and, and get to a place where it's a lot more user friendly to either build it or interpret.

[00:13:22] Adam S: I, I, I asked these questions and it's, it's as with you developing your company, it's to scratch my, my own itch because I, I don't have a financial background quite transparently, you know, so, so when you look at my sort of potted history, you know, it started in music, it went into selling houses.

Then it went into, you know, I've been, I've been all over the place, but, but tech and finance was, was the bit that stuck, right? But I don't have the 10,000 hours of, you know, posting credits and debits into a, as you say, an an accounting system. So it, it amuses me asking these sorts of questions because at, at some point, it'd be good to be able to do some sort of challenge that says, right, well let's, let's take three people that just don't have a, a finance background, right.

And let's, let's give them, you know, best in class, low-code tools and, and see, see what they end up with. And I, and I don't think we're million miles away and I, and I think a lot of finance teams are probably thinking, oh no, nobody will be able to do that because it's, you know, it's, it's a threat and it's scary.

Right. But I think, you know, you, you give people the right tools and it's, it's funny what results you end up seeing.

[00:14:30] Adam T: Yeah. It's, it's funny that you mentioned that because I think as technology progresses, I think we should raise our expectations of the user experience of, of software tools in general, right?

Not just finance, but you know, finance, FP&A tools, financial modelling tools in particular as well. You shouldn't need to be an expert to use these. Right. And when I built FlowCog, I. Have the end user in mind as a kind of this customer persona of a founder who is busy. Maybe they don't have time to build a model from scratch, but they need projections to help run their business, let alone for board meetings and, and, pitch calls with investors.

And so how can I make this tool easy enough for them to use and to actually, get through pretty quickly without investing a ton of time? Right? And so I think the, the quicker and kind of concept of time to value, right? How long does it take someone to use a tool, and do they need to be an expert in that field in order to use the tool in the first place, to actually get value out of it, right?

And then there's this kind of concept of, not only the time and effort and brain power that you need to put into the tool, but what do you get out of it on the other side, right? So in principle, as. Technology advances and we get better and better tools and they, they become more useful to us. We should be able to put in a relatively small amount of data, effort, brain power, time, and on the other end, get out more of all of those things, right?

It should, it, it should be, you know, this, this machine where the inputs are, or the outputs are, are greater than the inputs. They're greater than some of the parts. And so you should be, expecting, and I, I think this will happen in the next few years as especially with, kind of generative AI and, and all that, being able to have an output of a product, whether it's a financial modelling tool or any software tool where the amount of time and effort that you put into it, pales in comparison to the actual insights that you're getting out of it.

And, Again, it shouldn't require you to be an expert in it. So, you know, I, I like to think of, this, this kind of funny concept in, in product development of like, the user is the final boss. And so, when I'm building my product specifically, I can speak for myself here. I want to build a product that's easy enough for someone like you to use that is, is not, that doesn't have the 10,000 hours of modelling experience, right?

You should be able to look at the tool and understand it in plain English and enter in your numbers and, and do what you need to do and get something out on the other end that's useful for you and, and not require you to be an expert, not require you to give a hundred hours of your time in order to get something valuable out.


[00:17:38] Adam S: No, I, I'd, I'd agree. And I mean, talking about your persona there, you know, the, the, the company founder that's wanting to, to, to obviously model some projections and, and, and that sort of thing. I'll say something controversial now because there's a lot of talk about, you know, finance, business partnering, you know, how do we communicate financial information to non-finance professionals and so on and so forth.

I say it's controversial cause I, I don't necessarily think you need a finance background to be a good finance business partner. I think you need a good grounding in business, you know, and I, and I think you need a good understanding of the different component parts that make a business work. And yeah, a hundred percent you need to know how finance works.

Don't, don't get me wrong, but I don't think you need 30 years worth of prior experience in finance to be a good finance business partner. If you are able to demystify and provide guidance to other members of the, of the company, and it's all like tools like this, you know, that, that, that help you do that, right?

Because you can immediately start producing results based on imagine scenarios. I. anyway, so I'll, I'll, I'll let people fight over whether, whether my words are true or not, but we can ,

[00:18:45] Adam T: I, I actually, I, I do have some thoughts on this as well. I, I agree with you. in large part, I think. Yeah. To frame it as a, as a modelling problem, right?

The effort that you are putting in and your technical chops in a finance role, can only get you so far before you've run into the other inputs in the model, which are the, conversion rate of your efforts and technical expertise into some sort of actionable sound bite for management or for whoever your deliverable is going to, right?

So you can have the, the best spreadsheet in the world that's doing the fanciest things, but maybe that's not how you should be spending your time at all. Maybe you should be spending it talking to those other departments in the company to figure out what problems you can help them solve, what roadblocks you can help them, you know, remove.

And I think that when finance folks get excited about going deep into, modelling, there's nothing wrong with that fundamentally, but, I think it's important to contextualize it and to understand that, again, not losing sight of the bigger picture. Your goal in doing some things in exercise, like modelling or other types of ad hoc analysis is to actually get some insight out on the other end that you can then take to one of the departments in your company and, and help them improve a process or help them, develop a better strategy or approach.

And so, yeah, I agree. I mean, sure. You need, you need some understanding of finance if you're in that role. Ultimately though, you know, and, and, and, you know, especially as, as, technology progresses and we get better and better tools. Hopefully that abstracts away a lot of that, that legwork that the finance person is doing anyway.

And so then the focus is even more on, okay, now that I've freed up a bunch of my time, cuz I have better tools at my dis disposal, how do I then communicate the insights more effectively to actually move the needle on the, the trajectory of the company? Hmm.

[00:20:59] Adam S: And, and it, I guess it comes down to, to the way that you, you, you describe it as well, and as, as you say there about focusing on the, the message that needs to be to be communicated is, you know, fi finance business partners only a step away from, I guess broader business partners.

You know, so, so how do you define it? Is it predominantly a finance business partner that's focusing solely on finance information and communicating that, or is it a business partner? Is also finance savvy. You see what I mean? And, and I do, you, you can't, you can't answer that question without understanding the sort of business and, and the goals.

But, you know, so, so again, a another loose definition for people to decide whether they do or do, do or don't want to get into the camp of anybody can be a finance business partner, but , there we go.

[00:21:44] Adam T: No, but it's, i, I I think it is part of a kind of, a broader perspective of, of looking at the finance role and how it's changing, right?

Especially with technology. overall, you may not even be spending your time wisely. Doing modelling right, regardless of the tool spreadsheet, my tool, any tool out there, it may not be a good use of your time depending on the, the size and the maturity of the company, right? If you're really early on, and, and I tell prospects this even a lot of the time, if they're under, call it 500 k in, in a r r, you can use my tool.

You're not gonna get as much value out of it as if you were already at a, a few million in a r r. it may still be valuable and you may still need it for your, your, board meetings or for investor calls, but ultimately don't spend too much time and don't, don't kind of overemphasize the outputs of that model because it may not.

Relevant in another six months. Right? And so you want to minimize the time that you're spending on, on an activity like that because maybe there are other parts of the business that you should be working on and focusing on. Right? Maybe for the finance, business partner, they should be working more closely with marketing to help understand their, their marketing r o ROI on I'm their ad spend, right?

And that may be infinitely more valuable than, you know, being heads down in a financial model for the business, at, at different stages in, in a company's lifecycle. So that's, that's another thing to take into account. You know, the, the act of kind of modelling in and of itself, is not valuable unless it leads to some actionable insight.

Assuming that that's the right time and place to go through that exercise and, you know, again, if the company is early enough, you should probably be spending your time doing something else.

[00:23:42] Adam S: Yeah. Especially. You know, younger companies that are growing quickly, the amount of variables and potential for change in the early stages of a company's growth would be, I think, incredibly difficult to build into a financial model just because there are so many unknown unknowns at that stage.

so I'd, I'd agree what you're saying there, and, and coming back to your point about sales and marketing aligned with finance, really interesting conversation with, with Chris Ortega, where we, we got into the weed quite heavily about that sort of, customer acquisition cost and all that sort of stuff.

so yeah, a hundred percent you need to, you need to pick your battles at the end of the day, don't you? You know? Yeah. But it's good. You probably win more business by saying to people it's the wrong time. Come back in six months because then you're trusted. You're not just pushing a product, are you? You, you're actually looking at their goals and making sure that it's the right fit rather than something that's not gonna generate that value.

[00:24:39] Adam T: So, yeah, I, I think it, it's better for everyone, right? If I think that. A prospect is a little bit too early on to get real value outta FlowCog, I may explain to them, Hey, it's not that the product won't work for you. Come back in, in six months, it's gonna work a lot better for you. Or if you wanna set it up now, we can talk about maybe the, a discount or something.

But ultimately you wanna be careful with how you're spending your time overall. And if you can spend a minimal amount of time using my product and, and get a result from it, great. But you really do need to understand that at different stages in a, in a company's life cycle, modelling's, usefulness is going to fluctuate, right?

It's, it's gonna vary. And yeah, ultimately it saves both me and the other party time, if I can tell them, Hey, Like, I would love if you gave me money right now for my product, but ultimately I don't think either of us, it's, it's gonna be a good, use of our time. Yeah. Because if you're unhappy with it and it's not actually giving you value, you're gonna end up churning anyway.

And so I've invested the time in getting you set up with the product, even though a lot of it is, automated and, you know, I've, I've gotten it down to a pretty short amount of time in terms of onboarding. Yeah. it's still not a good use of both of our times. Right. So if I can save them time and help them understand upfront what the expectation is, I think everyone is just better off overall.

[00:26:07] Adam S: So you mentioned the, you, you just put the figure out there, you know, 500 k ARR, which is obviously an, an annual recurring revenue, right? Mm-hmm. leads me to, to a point that wasn't on the list of questions that I sent across, but obviously you've decided to focus quite heavily in, in sass, so, so SASS modelling.

Is there a reason that you specifically went down that SaaS route as opposed to just creating a, a, a, a modelling tool that had wider applications?

[00:26:35] Adam T: Yeah, this gets back to what we were talking about before about low-code tools and, the difficulty of having someone either learn a new tool or inherit a model and, and learn all of the kind of building blocks and the, the kind of atomic units of it from scratch.

I think that at least for modelling for SaaS companies, it should be a lot more automated and you shouldn't need to spend that much time going through that. Right? Because there are a few kind of fundamental principles when it comes to modelling SaaS companies and where the emphasis should be, so, Once you narrow the tools, kind of industry down to just sas, you can get pretty, automated and opinionated about the structure of the product and what that looks like from, from the customer's perspective.

So instead of showing them all of the guts of the product, which, they can see if they, they dig deep enough in different parts of it, a lot of that is kind of purposely tucked away. The idea is that they should be able to see the inputs into a model that are relevant to them. enter them in with, you know, Making sure that they understand the inputs and giving them enough documentation and resources to really, come up with, with good estimates.

And then on the other end of it, to get actual insight out and, to not have to, to be able to do that, that whole journey and to minimize the amount of time that it takes, right? so the kind of more narrowly focused in terms of the industry, your, your product is, and it doesn't just go for financial modelling tools.

It goes for, for a lot of tools out there. You see this with CRMs these days. A lot of, There, there's kind of this, this move more toward, vertical SaaS companies where they're serving a particular niche and their tool ends up being more useful because the product is not as flexible. It is more opinionated.

Doesn't mean there's no flexibility in in some of these CRMs, but because they know that they're a CRM for barbershops, they can abstract away a bunch of the kind of, legwork that a user would need to do to get up and running with their product. Right. so that's what I've tried to do with FlowCog is to make it so specific to SaaS companies.

And this is another area where I, I will actively turn down a prospect if I. Shortly into our conversation, realize that they're not a, a true SaaS company with recurring revenue. that breaks a lot of fundamental assumptions, and they should go and use another product because FlowCog is so specific to SaaS companies that, you know, even coming down to what the, the p and l looks like, right?

It's gonna be very opinionated. I don't let users add or remove line items from the P&L. There's actually two P&Ls. One of them is a more detailed one that gets allocated into a SaaS formatted P&L SaaS forwarded, SaaS formatted. P&L is five or six line items. Revenue, cost of revenue, sales and marketing, R&A, G N A.

And that's it. And that's because that's how the entire industry does it. You need to fully allocate your expenses between your cost of revenue and your, your operating expenses. The reason for that is because a lot of those, metrics that people wanna see. Your gross margin, your lifetime value of a customer to customer acquisition, cost ratio, burn, multiple rule of 40, things like that.

You want those to be comparable to other SaaS companies, both public and private. And so there are just some definitions that have emerged over time in terms of what those metrics, how you calculate them, and the kind of p and l format that you need to be able to calculate those reliably. And so, For SaaS companies, if their tool is not giving them, the output in that format, when they go to do their comparisons or benchmarking or anything like that, it's not gonna be as relevant.

It's gonna be apples to oranges. And so you want to give the user, the kind of best fitting product and result, for their particular use case or company that you can. And it's hard to do that if you have a more generalized, tool. And I would include spreadsheets in that as well. Right. There's, you can think about it in terms of, degrees of freedom.

How much flexibility is there with the, most fundamental building blocks of this tool with a spreadsheet, it's nearly infinite, right? It has so many degrees of freedom because you could do anything in a spreadsheet. But the problem also is that you can do anything in a spreadsheet, right?

Yeah. Yeah. So that can lead to a lot of, a lot of issues. you all of a sudden have to re, you know, reinvent the wheel and, and start to think about data structures and everything. And, you know, if that's not your full-time job, and you should also be spending your time elsewhere, it's gonna take you that much longer to get to, a point in time where the, the thing that you've built is actually serving you and giving you, more insights back than, than kind of the data that you're feeding into it.

[00:31:57] Adam S: And, and it's interesting and, and I'm very keen to see how things progress. So, so a lot of my experiences with sort of, I guess more mainstream ERP systems like you, you know, your Microsoft’s and your Sages and your SAPs and, and that sort of stuff. And of course there's been a, there's been a big shift recently as they took old, old style on-prem systems and they've moved them into their own SaaS platforms, whether it's on Microsoft Azure, Amazon Web Services, what whatever happens to be.

But what's interesting is obviously now that they've got data, what huge volumes of data in, in public infrastructure, i e you know, mul, multi-tenant, shared, shared server environments, when anonymizing data, these big providers can start analyzing trends. And, and some, some stuff that I've heard floated about is the concept of stuff like, GL as a, a service, you know, and, and all of those sorts of things.

And it's exactly what you're saying there. So, so your tool, because it's SaaS focused, will only have one structure because it is. Industry best practice to have this structure. It's what everybody else is doing. So it takes that thought process out of it. I think we are gonna see a shift with that in more strength mainstream solutions, not from a forecasting perspective, but from a general accounting perspective, not just in the enterprise space, which is really, you know, a little bit where that's, I think gonna be aimed to begin with in the SME market.

Wouldn't it be Ace if you, you know, you logged onto a, a Xero or a QuickBooks or whatever the platform is, you entered your industry and immediately it just came back based on best practice from other industries, how to structure your accounts. And somebody might correct me and then please do, if anybody's hearing this and they know of any of the smaller solutions that are giving a preset industry chart of accounts, for example, based on live data, I'm not talking about just a, you know, a pre-canned format based on something that was developed 10 years ago.

I'm talking about shared data from public, public, public infrastructure. So I think that's definitely the way that the market's going, but the end result is, Time savings and obviously, you know, being able to share best practices without having to go through that 10,000 hours piece that we've, we've been talking about today.

[00:34:08] Adam T: Right, exactly. And, a couple companies come to mind for, on the accounting side of things. One of them is called, NS Sheets. I know that they're doing some interesting things when it comes to kind of automating the, the accounting journey. another one called SaaS Ant, they, they do something similar.

But I, yeah, I, I do think that that's the direction that we're trending, right. If you have. A, a big enterprise product and you know, you're a multi-billion dollar company and you see all of your customers doing something over and over in the same way, there's a good chance that when you combine that with like best practices in the industry, you can come up with some sort of layer of intelligence that you should be delivering back to the user or to a new user when they, start to use your software to say, Hey, you should be setting up your chart of accounts in this way.

Don't add this account because you're gonna end up removing it later. Because that's what a thousand companies that we've seen and that have started to use our accounting product have done. They, they start off with these chart of accounts, they converged to this, set of chart of accounts. And I think ultimately it's not trying to, shoehorn everyone into a one size fits all solution, which is where kind of the nicheing down comes into play.

As long as it's kind of. Industry specific, and it serves those customers well. And there are, you know, kind of shared principles across different, companies or, or, business structures in, for that industry. Then I think it lends itself well to something like this. I don't know if I would call it AI necessarily, but this general idea that when you are setting up and, and implementing a new system, whether it's an ERP or or accounting system, whatever, you should get a lot of help with that.

You shouldn't have to go out and read a bunch of resources on the best way to set up this product. Right. That that product should give you a lot of the intelligence of previous users of that product or in general best practices from that injury. You know, in the best case scenario, both of those, and you should really be guiding the user through that journey of, of implementing your onboarding in that way.

Right? So it's almost like I think of the concept of, bowling with bumpers. And so you're, you're not just out there, you know, bowling for the first time and you're, you're gonna throw the ball in the gutter every time. You have bumpers on both sides that say, no, no, no. Like you, you may wanna reconsider how you're doing this thing and, and maybe do it this other way.

[00:36:54] Adam S: ‘Bowling with bumpers’ love that…

[00:36:56] Adam T: I think that, eventually we will, we will see a lot more of that, you know, and, and I think that's the, kind of progression of what, what it means to write good software is like how much does the user need to think and reinvent for themselves, or go and hunt down the answer someone else, somewhere else?

And how easy is it for them to just get up and running with the tool? The tool feels intuitive to use, and they're getting. Feedback or insight from the product itself that tells them, Hey, do things this way. don't do things this way. And so I think that's where, again, the, the nicheing down comes into play.

It's really hard to make those suggestions if you're just a general purpose tool. and don't get me wrong, you can, you know, you can think, you can imagine, let's take spreadsheets for example. there are still, even though it's a tool with infinite degrees of freedom, right? You can do anything in a spreadsheet, there are still best practices for writing formulas, not hard coding numbers, inside formulas, things like that, right?

And so you can imagine a world where Google and Microsoft, over time, they start to build in that level of intelligence, into their, into their tools. And they say, Hey, you know, Clippy pops up and says, Hey, we noticed that you just hard coded a number in a formula. We recommend against that for these reasons.

Here are some resources if you wanna learn more about it. Oh, and by the way, here's a better way to write the formula that you're writing. Yeah, right? Yeah. And I know that there have been. AI tools that have popped up to help with, you know, writing Excel and, and Google Sheets formulas. I just think that we'll see more of that over time, and it won't be something as, disjointed from the product experience itself, where you'll have to go to a, a side panel and say, Hey, write me the formula for this.

It'll just, the tool itself will know. And when you go to, write a spreadsheet formula, it'll know, Hey, you're not doing this in, in a way that, industry experts would recommend, spreadsheet experts would recommend. Right. There's a better way to do it.

[00:39:11] Adam S: Yeah. and it will, it, it will come, it will come with more data.

So, so the, there's a couple of examples there, but I think I still, I think we're still very much in suggestion territory i e high level suggestion rather than prescriptive. No, you want to be doing it this way. So the, the two examples are, and you mentioned one there, Google Sheets. Right. so for, for my work, I use Microsoft for my business, for my personal, I use Google Sheets, right?

It's just, I'm not greedy, it's just that for whatever reason, my Gmail email started using Drive and then it just evolved from there. And obviously work has always been Microsoft, but I was doing some, some calculations on, on Google Sheets, and it started doing just that. And I've seen it before. with, with Excel it's that, you know, you start writing a formula, like a, like a sum and look at the spreadsheet and it'll say, oh, because you've started writing this, we can assume that you want to sum the total of the, the columns above, you know, and that's where I say it's suggestion rather than this is best practice because it's just inferring from, I guess, a common set of interactions that it has.

The other thing that I saw recently that was quite cool, this wasn't to do with spreadsheets or finance or anything like that, but it was actually in hubs. So obviously HubSpot as a CRM tool, again, has been in the cloud for, I dunno how long, but, but a long time. Right? And they are working from a, a shared infrastructure and we got a popup the other day with an email that somebody had incorrectly entered.

And HubSpot wouldn't allow us to message using that email because it said HubSpot has reported a high volume of bounces to this email by other companies. So again, this is kind of hints at what's to come, I guess, you know, and, and it's nice, nice to see. But the thing is, to most people that are just using these systems day-to-day, it's kind of a Oh, that's nice.

That's pretty cool. They don't necessarily understand the stuff that's going on in the background. So I think we'll be in a position pretty soon. And, and, another podcast guest that, that I won't give away cuz it's not released yet. he. Forecasting that the next step away from the likes of ChatGPT, where at the moment it's still very prompt orientated.

You know, write a better prompt, you know, get a better result. The future is even further than low code. It's almost as if just do stuff without thinking about it. And then you'll have all of this platform in the background supporting your actions without you even knowing it. So I think that is definitely, and then we're all obviously into that territory of, you know, human versus machine.

Are we a hybrid? Are we half robot? And all of that sort of stuff. But no, I'm fascinated by this stuff.

[00:41:49] Adam T: Yeah, it's, really interesting this kind of concept of auto complete or have an AI kind of guess the next word or the next action, that that should be done in the form of a suggestion.

Right. And, you know, hopefully the, kind of the UX of a lot of software over time, evolves to the point where things are happening in the background that the user is not aware of, and it's exactly in line with their intuition and expectations.

Yes. And you know, like the, the HubSpot example is, is perfect, right?

HubSpot should never let you send an email to, an email address that they know has a, a super high balance rate. Right? Yeah. That's easily avoidable. You don't need AI for that. That's a, you know, simply store that in a database somewhere. And, you know, HubSpot, should be able to do that and retrieve that when the user goes to, to send something and immediately just surface that intelligence, right?

Yeah. Yeah. but you know, you, you see this kind of auto complete too with, GitHub co-pilot, which is, I I absolutely love it. I think it's a, a fascinating product. It's right most of the time. And if you know how to, Kind of prompt it. It, it does, but two things, right? you can prompt it by writing a comment in your code base to say, Hey, ums, you know, select this item.

If it has this attribute from this list of, of objects right in, to give a program example, it'll go and, and write the code out based on the rest of your, the code in your code base or in that, in that particular file. And it does a very good job of it. I'm, I'm extremely impressed by the kind of, not only the, ability for.

Co-pilot to come up with an answer at all, but then to come up with a very, performant or, afi an, a snippet of code that actually, executes in an efficient way as well. So I've had a, a really good experience with it. some may disagree with me, but if you know how to prompt it by writing in a, a comment saying, Hey, I need you to do, X, Y, Z, there's a good chance that that co-pilot's gonna be able to deliver on that.

and then even separate from that, it'll know if, if, depending on the kind of structure of your, your code, what the kind of next line of code, could potentially look like. And again, these are suggestions, but, vs. Code and, and GitHub co-pilot being integrated into it. For instance, to be such a kind of, a tool with so many degrees of freedom, right?

You can write any sort of code and, and, a text editor like that, to still be able to deliver value based on this kind of automatic recognition of the context that you're building the software in. that's what allows it to then say, Hey, this is, this is a really, likely or high probability, snippet of code that you'll probably need next.

Here's our suggestion.

[00:45:01] Adam S: Do you know what technology is being used in the background?

[00:45:07] Adam T: Yeah, it's, it's GPT3.

[00:45:09] Adam S: So it is, it's GPT3. Right. Okay, fine. So, so they, they've just done what the, the various other developers have done and they've said, look, you know, this is my prompt language.

Effectively this is the data to look at, you know, when somebody asks you this, then, you know, bring back this response.

[00:45:23] Adam T: Right. Okay. Yeah, and I'm, I'm sure it's some sort of super custom implementation of it and lots of fine tuning and, you know, there's this layer of being able to get feedback, from the user when they either take the suggestion, that copilot is offered up, or they reject it and say, no, no, no, I needed something else.

So there's, there is that nice feedback loop that they get as well. which is incredibly important when it comes to collecting data to then feedback into, these machine learning algorithms. Okay.

[00:45:57] Adam S: So staying on the topic of GPT3, I haven't been able to get into it for like two two, two weeks now.

Cuz the, the, the demand for it's so high. I have clicked the button to say I'm interested in premium for whatever reason, not, not available in the uk. I don't whether you've got the pro version in the US yet, has it come out? It has, yeah, it has, right. Okay. So I'm, I'm still gonna, I'm on the wait list, but, but either way, but no, coming back to you and what you've been doing, because you were one of the first that I saw, on LinkedIn, I think it was to start building GPT3 into your, your solution.

So can, can you tell us a little bit more about that? Obviously we've talked about having low-code options to, to build, financial models and, and forecasts. What, what's the advantage of building in, I suppose, additive AI, which is what we're saying there to a, a forecasting tool?

[00:46:50] Adam T: There's, there's kind of a couple approaches to this.

Again, kind of the, the North star here is this time to value for the user, not just when they first open up your product, whatever software product it is, doesn't need to be a modelling tool, any piece of software, right? When they are engaging in it, not just for the first time, but when they pop open a browser window and they, they go to your app, and they're looking for a specific answer, how do you get them that answer as soon as possible?

Right? so, you know, for instance, maybe the way the data presented in a tool, doesn't do all of the… Kind of infinite permutations of the calculations of year over year growth, or quarter over quarter growth for different quarters or aggregating different line items and then doing the comparison, right?

There's, you can imagine that the tables or dashboards would grow infinitely large if you tried to capture all of that data in, in one spot. So you should have some sort of assistant or a helper, to be able to ask a question in plain English and say, Hey, I don't wanna go digging for the answer of how this variable or how this line item changes over time.

What does it look like? Where's the inflection point in its trajectory? what makes up that number? And so you should be able to ask a question in plain English and get a response back that has a combination of, oh, okay, well we know what you're looking for because we've gathered enough context both from the product itself and what you've prompted us. Us being the AI here. And it, it should be able to go in and, and return a result.

That's a combination of English words on the page with numbers in a kind of easily, understandable fashion, right? And that's, that's the goal is to not, require that everyone using your tool is an absolute expert that spent 10,000 hours, in that field, right? You want software tools that are easy enough to use that deliver you insight back.

As long as you can give the, AI a halfway decent prompt at what you're looking for. Right? And it's really cool to see because like in, even in FlowCog, if you are using the AI feature to ask it a question about your projections or even your historical results that are in the tool, you can misspell words, you can, you know, come up with, there are, there's a lot of room for error when it comes to you asking the AI a question and the AI being intelligent enough to understand it and actually answer the, the question that you were really asking.

And so as long as you can kind of implement something like a a GPT3, ChatGPT in the right way for your product, you can have it be pretty specific with the responses that it gives so that it's doing enough explaining so that the user… It's not as much of a black box for the user of how the AI arrived at an answer.

You can tell it, Hey, I need you to show your work. Right? So that's what I kind of did with FlowCog is I made sure every time I'd give someone an answer, oh, you're, you know, how did your gross margin change in your projections going from, you know, the year 2024 to 2026, something like that. it'll say, “Hey, it went from, you know, a gross margin of 75% to 82%.”

“And the reason it did that is because these line items changed over time. And the reason it did that is because these inputs have these compounding effects in there that, that will change these outputs in the model.”

Right? so you, you wanna be able to kind of give that insight overall, and you also, in theory and eventually this is where I would like to take FlowCog, to be able to do that, but also to kind of integrate it more closely with the experience of when you're using the product itself, right?

So not just for querying, outputs, but also for, explaining to the user, Hey, you should be focusing on these inputs. Hey, these, inputs or this kind of idea of like sensitivity analysis or what if analysis goal, seek, things like that.

Being able to kind of take your set of data and assumptions that you've entered into the product, synthesize some sort of, insight and surface that to the user and say, Hey, you should really be focusing on this in your business because out of all of the inputs that you've entered, you know, no matter how, Efficiently.

Your, marketing spend gets your conversion rate to go from an opportunity to an onboard customer is so bad that it doesn't matter how, how well your marketing team is doing, ultimately the bottleneck is your conversion rate here. Mm-hmm. . So a tool like FlowCog should be able to eventually surface these insights to the user and to steer them in the right direction of what to spend their time on.

Right. so that's, that's kind of the, the ultimate overarching goal of this is to be able to, answer a question that the user doesn't know they have. Right. And so this gets back to the idea of not having the user or not needing or requiring the user to be an expert and to have this. expert level intuition on what they need to dive deeper into to be able to surface that for them, that I think is, ends up being really powerful and, and saves the user a bunch of time.

And again, this whole equation of how much time and effort and expertise are you putting into your tool, how much are you getting back out on the other side? so making sure that kind of, that equation is very favorable for the user, I think is, again, that's why it's, it's my kind of north star, making sure that they get those insights quickly and, and easily and without needing to come up with a perfect question to ask the tool.

[00:53:08] Adam S: Yeah. And, and how far do you think will we'll be able to, to take it? And I know sometimes this is a bit pie in the sky type stuff, right? But let's, let's say that you've got. Three, four years worth of data in FlowCog or whatever the tool is, to the point where we can build up a pretty good picture of performance.

How stuff has changed based on variables A, B, and C, how close we were to predicted scenarios A, B, and C, and so on and so forth. Now, obviously, coming back to the point that we mentioned earlier about unknown unknowns, when I first got access to to ChatGPT, I lost my Christmas to it, by the way.

I'll be perfectly open about it. I was, I was chatting quite a lot to, to ChatGPT and I was asking questions like, “What effect would a global pandemic have on the profitability of a company in the manufacturing and distribution space” for example? So I was just, I was trying to be as awkward and as complicated as I could possibly be as I think a, a load of people were.

I mean, I was trying to get it in and construct, whereas I think a lot of people just tried to break the machine and ask silly questions. But as, as I said, I didn't get down that rabbit hole, but I, I did try and test the limit of usable information that would come out of it. And of course it came back with the standard.

“Well, it could be this, but you've gotta bear in mind this,” you know, being that sort of politically correct, not wanting to actually make a recommendation and just, just, just put some information out there.

But going back to the question, so let's say that we've got three or four years worth of data.

How far away do you think we are off being able to ask whatever the “ How do you think our profitability is going to be affected by a recession or by a pandemic?” Do you think we are gonna be able to get to, to those sorts of answers or do we still think that that's quite a way away?

[00:55:07] Adam T: I think probably in the next year or two.

I know, I know for my specific case with FlowCog, I think it'll come a little bit sooner. Again, I have the advantage double-edged sword, right? I, I have a smaller target market just focusing on SaaS companies, but it does allow me to get more specific than with how I am implementing something like GPT3 to give the user, an answer that at least has a more narrow context to it.

And so then it can, you can kind of direct the AI to return. different sets of parameters. So instead of just giving a, a general response of, oh, this is, we think it may affect profitability in this way, or it could affect profitability in that way, it really depends on how it all plays out. You, you can get it more, to give you a more detailed answer by saying, okay, GPT3.

You know, when you give the response to the user, when they ask how something is gonna affect the profitability, first take into account what the inputs of the model are and how those are related to the substance of their question, and then give them this kind of range of outcomes. Montecarlo analysis, sensitivity analysis.

So you can see how changing a few of these inputs in the model, either isolated changes or in tandem with one another, are going to affect the outcome and how much it shifts the curve. And then you should also explain that to the user as well, and say, “Hey, we think that it may affect profitability by x, y, Z percent. And we got that by changing these variables by this degree.”

And hopefully still, of course, give the caveat, this will vary based on your company and, and what you know, even if you are a SaaS company. Not all SaaS companies are identical to each other. So, of course you wanna give that caveat.

But, I think the, kind of, the takeaway from that is as long as you can get, the, the more specific you can get with the context that you're, that you're giving the AI, the better answer you're gonna get, the more useful or actionable answer you're gonna get.

So if you. I don't know. Let's, let's make up an example. Let's say that, you're a manufacturing company and, you're, I don't know, you do a lot of business at a port and that port gets flooded. you know, the, the AI that you're using should be able to deliver a response that says, okay, well, “Taking into account how long it's gonna take the flood to get cleaned up and all of the damage that it's done. If it takes x number of weeks, your inventory will be backed up by this much. You'll have this much extra carrying costs. you won't be able to, you, you know, you'll have delays in terms of your cash collections because you won't be able to invoice the client as soon.” So it should be able to kind of give a, a detailed explanation of kind of these, this cascading set of effects.

But again, that's only if the, the tool. Has the specific context of what manufacturing companies look like, how they behave, what are the external factors that affect companies like this? so I'm not an expert in the manufacturing industry, so . Yeah. Yeah. I I may have butchered that, that explanation, but I think you get the point.

Yeah. there's, you know, I, I think, over time as AI tools get better and are more integrated into the products that we use, I think that context will be there. Hopefully the, the developers that are implementing these models, will do it in a kind of, smart enough way so that the user is actually getting some sort of, actionable insight out from it.

[00:59:02] Adam S: Yeah, and that's, I think that's the biggest limitation at the moment. Coming back to the whole sort of question about prompt engineering, and obviously people are joking that, you know, they'll be full-time prompt engineers pretty soon. I'm, I'm hoping that's not the case. Because I'd like it to get to the point where the AI does see the context.

It can already infer a lot of what you're already asking it. So the example I gave was, I, I use Notion, which is great. It's my to-do list and my notepad all in one, no affiliation. So I don't, don't get any kickback if I, if I recommend it, but, you know, it's free, it's easy to use. But that is my brain, that's my virtual brain at the moment Notion.

Right. And there's hundreds of pages worth of, you know, book notes for my reading. You know, there's, there's, customer conversation notes in there. You know, that there's all sorts. So if I were to plug Notion in something like ChatGPT, it already understands my language and, and the context. So I think we're, we're talking about the same things there.

It's a, it is about the ability to fill in the gaps without you having to be overly prescriptive on the question that you ask it.

[01:00:04] Adam T: Exactly. Yeah. You, you nailed it. And I think. When, and don't get me wrong, I think there are place for prompt engineers at some point in, down the rabbit hole in terms of how close you are to, you know, fine tuning the model itself.

Yes. But certainly I think it's a, a lazy approach to put the onus on the user to all of a sudden become a prompt engineer. Right? Yeah. Just like, I think it's lazy to, if you have a tool, doesn't need to be a modelling tool, but if you're asking a user to do something really complex and, and learn something from the ground up and, you know, hey, write this complex Excel formula in order to, to get the insight that you need, not that the folks that build Excel are lazy in that because they've built a tool that has the, those degrees of freedom and that's the expectation of when you use it.

But, eventually you wanna get to the point where you're using software that doesn't put the onus on the user to bring this expertise to the table. And it. Assuming that it has that context should be able to, like you said, from a very, basic or maybe even overly simplistic prompt from the user, be able to interpret the real in, question there or kind of the, the crux of it, and to deliver some sort of, more intelligent response.

[01:01:25] Adam S: That's it. And, and, and I'll use it, a really simple example. I mean, it's, it's sim simple in, in, in, as simple as I can get it, I guess. But what I did is I, I built ChatGPT into Google Docs.

Yeah, so again, I'm not developer, you know, I, I basically stole some code off GitHub, and I, I went into Google Docs.

I did the, what is it called? The, the integrations or something, I can't remember the button. But anyway, it opened up the panel. I pasted it in, and then I just changed the prompt. But the, but the theory was that I could just select the text in the document, click on the prompt, and get ChatGPT to carry out the action on the, the small amount of text that I selected.

So the theory was take a podcast transcript like we're having right now. It's an hour's worth of conversation. It could be so many pages of a four in terms of text. And what we've got, because we are human is filler words. You know, we've got, ums and up, we've got repeat words, you know, we, we've got all of that sort of stuff.

So a major headache of mine is taking that transcript and actually making it readable. You know, and, and tools like Descript that we were discussing before we started recording make it easier, but they're still not perfect. Yeah. And you've gotta pay for the really expensive platform. If you want to get the clever, you know, AI that takes out all of the repeat words and all of that sort of stuff.

So I just haven't done it. So I thought if I could get ChatGPT to, to basically remove all of the, the filler words and stuff from, from my transcript. Hey, I'm winning, right? I don't have to pay for, for the, for the hefty D script subscription. and I've got a bot that's working for me and I started the prompt and I can't remember exactly what I put, but I said, you know:

“Cct as though you are an audio transcription service. You are to take this text, remove words like, uh, um, as well as repeat words.”

Now, obviously because the instruction wasn't good enough, it ended up removing repeat words that weren't in the same sentence or next to each other. So I then had to change the prompt to say, remove repeat words that are next to each other and then that are next to each other, separated by a comma.

Do you see what I mean? And, and immediately I had a prompt that was, you know, a, a, a paragraph, two paragraphs long, and the results that it produced by the end of the exercise of, getting to, to that large prompt, were actually worse than the, than the initial prompt that I'd given it just to do a bit of a tidy up.

So the point that I'm making there is you can lose hours to trying to develop a decent prompt just down to the iteration process. So I caved in, in the end, I said, right, well, for the podcast, I need to get out. I'm just gonna pay GoTranscipt. It’s £40 an hour. You know, and, and if I'd done it, I would've saved myself at least two hours’ worth of prompt engineering.

Do. Do you see what I mean? So you've gotta pick your battles. I've gone off on a bit of a rant there. I'm sorry. But, you know, it's, it's all related, I guess.

[01:04:27] Adam T: No, it, it is, and this kind of, loops back into what we were discussing before is that, you don't want the user, you in this. To need to come up with Oh.

And make sure that the word, is next to each other. When you remove the duplicate, don't just remove it if it's shown up five paragraphs above it. Yeah. you don't want the user to have to come up with all of that internal logic. you, hopefully you want the, the tool to be able to, to do that automatically.

and it's hard to do that because there's a lot of guesswork involved for that tool to say, okay, well what is Adam really looking for here? You know, in the same way that in this example, if you had software doing this for you and they were thoughtful about implementing something like GPT3 here, you actually would want to also, Without the user needing to prompt the software, deliver the finished product in a way such that it's not even removing and applying the rule for every single filler word.

Because then you don't want the, the video transcription or the audio transcription to look or sound too choppy, right? Yeah. And so you do want this kind of intelligence and judgment calls, hopefully that the AI can make to say, okay, this was an or an ah, that took three seconds long. Yes, remove that.

Oh, but if it's below this, and we just removed an or an ah, Within, you know, five seconds of this snippet playing, don't remove it. Right? So you want these rules to be auto-generated. You don't want the user every time to reinvent the wheel, and all of a sudden need to become an expert in how human language sounds to the ear and what makes it sound natural.

Right? Yeah. hopefully the, the tool can abstract that away for the user and, and can handle those set of conditions or rules.

[01:06:24] Adam S: Yeah, no, a hundred percent. Yeah. Well, it's exciting to see where it's gonna go, isn't it? And I, and I, I'm excited to see how FlowCog develops. So, so, so we, we've, we've gone over time already.

But I, I will, if you've got a few minutes, just ask the, the questions that I always like to ask at the end of these sessions, and then you can, you, you can, you can tell people how to find out more about FlowCog. So obviously we talked a lot about tech in terms of modelling, ChatGPT obviously, which is a, a bit done to death, but it's an interesting topic, so I'm, I'm happy to keep talking about it.

But the question is whether it's in your work or your personal life, I mean, you mentioned the GitHub autopilot, which obviously is, is quite cool for you, but is there an app or piece of software that, that you just couldn't, couldn't live without?

[01:07:14] Adam T: I, GitHub co-pilot is definitely up there for me. Yeah. I would say, my second monitor is, is pretty useful. That’s a good one.

My, trusted, Microsoft Ergonomics 4,000 keyboard is, is my companion. it truly understands me. yeah. Yeah, I, I would say those things, you know, a, a good, a good enough microphone, hopefully. Yeah. hope, hope it's coming through. Okay. that's, that's a good one.

But yeah, in terms of software, I would say, I use Trello a lot. I, I absolutely love Trello. I think it's a, an amazing tool. I do use Notion as well for a few things. I, I do really enjoy, navigating and, and storing information in Notion. A couple of, a couple other ones. But yeah, I would say. I, I think the, the one that stands out is, delivering that, that magical moment to me consistently is GitHub co-pilot to be able to see it over and over deliver.

And it doesn't mean that it gets it right every time, but I know that if I have a task that I have mapped out in my head well enough, I know that I'll be able to save a bunch of keystrokes and not have to type out the 10 lines of code that are gonna be generated. I know that I can ask, get up co-pilot to produce that code and there's a good chance it'll, it'll deliver.

So it's, it's hard to find something else that is comparable to that just because of how. Amazing it is to see a tool, kind of into it, what you're, what you're looking for, right? Yeah.

[01:09:04] Adam S: We're, we're, we're in that territory and, and again, go, we, we are, we are a bunch of nerds, aren't we? Where it's actually rewarding to have these systems feedback.

It's, it's bizarre, isn't it? You know? It's, yeah, but no, I, I, I totally agree. The stuff that you can actually see the results pretty much immediately is, is the cool stuff, right? You know? Yeah. I completely agree with that. And somebody else mentioned to me as well that the notion of got their own version of chat, g p t, that they're building in from a generative point of view, which I still haven't switched on yet.

I have with three podcasts later, and I still haven't switched on.

[01:09:38] Adam T: Yeah. I, I haven't tried that one out yet, but, I'd, I'd be interested too. Yeah. Yeah.

[01:09:43] Adam S: And you monitor, you're extra monitor. Do you have it portrait so you can fill it, fit in, like thousands of lines of code or you just with your standard setup.

[01:09:52] Adam T: No, I, I have, I have a standard, setup side by side. Yeah,

[01:09:57] Adam S: Fine. Yeah. And, and, and for anybody that's, that's obviously not a developer and familiar with developer communities, I, I've worked with, quite a lot of companies when you go into the developer group and they've got a wide screen monitor that's tilted 90 degrees on its access, so that they can fit in more lines on the same, on the same screen without having to scroll horizontally that, that's why I asked the question.

[01:10:17] Adam T: But No, it's, yeah, no, or if you have the command prompt up or terminal can see what's going on. Yeah. Yeah.

[01:10:22] Adam S: Absolutely. Absolutely. So where can people find out more about you, more about FlowCog? I'll, I'll put the, link to your LinkedIn profile in, in the show notes. But where else can people find you?

[01:10:33] Adam T: Yeah, so, FlowCog on LinkedIn. Adam Tzagournis CPA on LinkedIn. I'm also on Twitter as well. Not as active as I am on LinkedIn. but those are, those are the main spots to find me.

[01:10:47] Adam S: Very good. And you've got the, you've got your hair tied up today. but if people find you on LinkedIn, you'll see, some, some long flowing luscious locks on the old LinkedIn profile, right?

[01:10:59] Adam T: Absolutely.

[01:10:59] Adam S: I used to have ringlets down to my shoulders when I was in the old, so no, I do miss it. I miss it quite a lot. and just so people can search your certain for people, on, on audio, sais is T Z A G O U R N I S, you. Yeah. Perfect. Excellent. So as I say, I'll link, I'll link everybody up.

Oh, you've got your hair down now.

I've just switched screens. I've, I've seen it come back.

Very good, very good. So, yeah, we'll link up in the show notes, but Adam, it's been an absolute pleasure having you on, and hopefully six, 12 months we can have another catch up and, and we'll see how, how well FlowCog’s doing, eh?

[01:11:38] Adam T: Absolutely. Really appreciate you having me on Adam.

[01:11:40] Adam S: Yeah, no worries at all. And Fab, thank you.


tech for finance

©2022 by Adam Shilton. Privacy Policy - Terms of Use

©2022 by Adam Shilton. Privacy Policy - Terms of Use