055 Accountants Selling Software To Accountants

Back from Engage and feeling inspired by the number of lovely former accountants that I talked to that now sell software to accountants. What? And are absolutely crushing it. Should you sell software to accountants? It's one of those things where we all react. That's not a thing that I know anything about or have ever even considered. But it's also one of those things that's like, you don't do it until one day you actually just. Just do it. I don't know that any of the accountants selling software to accountants now are like, oh, yeah, I set out to do this someday, so I will share my experience. I've built three apps. I've sold one app. I can tell you what my experience building software has looked like, how to, like, dabble into this a little bit if it interests you, and just kind of some tips, cautionary tales, all that. Let's do it. Come on in. Happy Monday. It's Jason Daly. As I was going around engaged last week in the expo hall, talking with folks, I just, like, almost everybody I talked to or a ton of the folks I talked to are former accountants and they're now transitioned into software. And I get that doesn't feel relatable. But the consistent theme I picked up from, like, almost all these people was like, I could have never imagined doing this. It started as maybe this little side hustle or like, they're just regular accountants like I was, who never expected to become software people. And I've done a little bit of software development myself. And so I thought, let's just dedicate an episode to, like, what it looks like to start wading into that a little bit, developing a product or developing software for our industry. Because I still think ultimately accountants are best suited to do this. Doesn't mean that you can't have a big engineering team and a whole bunch of traditional tech bros on the team. But we've all seen the inverse of that and how badly it can go. The whole trope of previously, I founded this tool and I just was frustrated that there just wasn't a good accounting solution for my startup. So we decided that we're going to go out and make one ourselves. How many times have we heard this? How many bodies are on that road at this point? Accountants, we chronically sell ourselves short. I think you have a lot to add to the software space. So a few, like, specific folks that I talked with at Engage. Ashley Francis. This isn't really a software product so much. I guess in some ways it is. Ashley Francis is doing her 30 days of AI AI as a way to just like introduce accountants to AI and help all this stuff feel approachable. She rolled it out as like this kind of little thing and like the first two months of it sold out like almost immediately. And Ashley's a great example of somebody who's like, I'm just an accountant. I've got a little like solo practice. Never thought I'd be doing something like this. I'm going to put links to all these things in the show, notes Audit Miner. Another example, Kelly Mann, the founder behind Audit Miner, a bootstrap software company where they do like automation for benefit plan audits. Super cool lady. Like an example of someone who's like, they've got like a team of, I don't know, 10 plus people now. They're absolutely crushing it. It's like these people are accountants. They could have never imagined doing this stuff. And then tax plan iq. Jackie Meyers behind this one ran a tax, she ran a tax practice for years, just sold it at the end of last year. And now she's got like a, a platform for helping you engage clients to do more profitable tax planning. And these are all like, I mean like totally, I don't know, normal like accountant folks who like decided, I'm gonna start doing a little bit of this, but then eventually got like pulled into it more and more. And one of the things I feel like we do on this show is like, try to push technology in a helpful direction for us. Like, what do we need help? How do we get the message out there? How can AI make the stuff that we're building better? And so if we can inspire one more person to go out and make that killy thing, that will make our lives easier. Great, let's do it. Not that we need more people to leave actually the doing of the accounting, but we need really smart people like building solutions for us. And that is like a really cool aspect of building software. And being involved in that process is just the scale on which you can help people. So like, you think about the tax offers. We use QuickBooks, stuff like that. These tools impact like the productive output of our profession on like an unbelievable scale. On a greater scale than we do when we're doing one to one work with clients and stuff like that. As much as, like, as easy as it is to like dunk on product people, it is really important stuff that they're doing. So why would you do this? A few things I've kind of like encountered in my experience in terms of like what a few whys were for me, obviously there's the money making aspect, the fact that software is in the grand scheme of things, very low overhead relative to like, you know, if you're going to devote half of your time away from firm running to something else. Software theoretically is just about as high leverage as it can get. Depends on like the model of that software company. There are softwares that are like more hands on and just kind of have more inherent overhead. But it goes without saying a successful app can be a big time moneymaker. But that's not the only reason. So second, it could simply be to add legitimacy to what you do. And I think when we think of software we often think of these big ambitious things, big teams, raising money, all that stuff. But there's a lot of versions of software that are like come up well short of that. They don't ever necessarily need to directly make you a pile of money, but they can also add a ton of, they can create a bunch of legitimacy for you and maybe an adjacent business that you're building. So we've kind of been talking about building stuff for the accounting industry, but what are the problems that your clients have where like a micro product there could be a helpful solution and yes, it may make some money, but is there a way that that application could add legitimacy to your main business? Maybe that's the firm or something like that. Maybe that's a great approach to marketing. And then third, simply to learn, like to learn the nuts and bolts of how to develop something from scratch, maybe to learn how a new technology works through the lens of like this specific application, this problem that you want it to solve. So in my time I've developed three apps that have kind of touched on each of these wise. So the first one two or three years ago, fin daily. I'll put a link to all this in the Show Notes. I also linked the previous apps I mentioned in the Show Notes. There's an aspect of this podcast like today that is self promotional. And just to acknowledge that I don't really know a way around that, I'm just gonna tell you my experience and like what I enjoyed and didn't enjoy about building products. So Finn Daily I developed that two or three years ago. It's basically a automated email that goes out on a schedule that you pick. And before that automated email goes out, it goes out to your accounting API, your plaid banking API even zapier to pull in like variables, like live data for variables that you set up in these email templates so that each day or each week or whenever you want these Emails to go out on an automated basis. It pulls in live information to those emails right before it sends them out. For me, this was helpful for cash reporting that we did for our clients because I could pull in the bank balance. I could pull in like, upcoming bills from QuickBooks or Xero. I could even load in custom amounts from zapier. And this was just a way of doing a service we were providing for clients in a more automated way. And when it came to building this product, I learned what I enjoyed was the building of the cool thing, like making that thing. So I didn't develop it myself, I was working with a developer, but I enjoyed like, breathing life into that idea and like, getting it to market. What I didn't enjoy, I learned, was just how much work goes into, like the ongoing development of it. Like that is sitting on calls with accountants all day gathering feedback and deciding what you want to do next. I joke I'm too much of a know it all to like, enjoy sitting on calls with accountants all day when I'm like, no, like, I want it for this. And I have this very specific thing in mind. Like a big thing you learn right away in software is everybody wants different things from that product and ultimately it's up to you to like, define the direction of what it will be, which. And just importantly, what it will not be. And I learned, like, having gone through it, that I didn't enjoy that part of running a product. And then what does marketing look like for your product? Like, I think there's like this perception of software that, like, it can be this amazing kind of exponential thing that just happens when the reality is like, it's a lot of work and it's showing up every single day. So like, how do you sell a thing on a recurring basis? Like, how do you keep getting that in front of people? How do people discover it organically? It's just a lot of work. And at that stage, that was when I was still running a firm. That was when I was investing in content like this. I was like, yeah, no, I just, I realized this is a really cool thing that we've built and I really like it, but it's not the project for me right now. Like, I learned what aspects of building software I liked and I didn't like, I didn't like the, like, continue running that and growing that. Like, that was just not the project for me at that point in time. So for me, coming into that, that was probably just straight up like a make money sort of like initiative. If I think about what my why was maybe there was a little bit of adding legitimacy. This was like very early in my YouTube days when I was starting to create content. So maybe there was a little bit of legitimacy there. But it was mainly like I wanted to like wade into product development and be like, is this the thing that I would actually enjoy? Cause I'm very obviously a tech nerd and I really enjoy software and all of that, but would I enjoy being on the other side of it? And what I found was there were aspects of it that I enjoyed and there are aspects of it that I didn't. Another app that I launched, when was that? Beginning of this year. Scan Straight. So what Scanstraight does is it takes smartphone images of documents and it removes everything but the document. So like, it auto crops it, it like de skus ocrs it so that regardless of what the images that you have, what you're left with is a PDF that is OCR'd and, and tidied up. And it's an elegant solution to a problem that you shouldn't have in the first place because it's like kind of the last like the worst case. It's for handling the worst case scenario where somebody didn't use a scan app when they captured those things and sent them to you, or they didn't use like your portal app to like capture those things appropriately. But like we've all been, especially if you do tax, we've all been in that situation where a client sends you 50 smartphone images of all these documents and you're like, what in the world am I gonna do with these? Now that's what this app is for. It's a desktop app. There's cloud services that will do something like this. But ultimately like am I gonna go out to a cloud service and just chuck hundreds of my clients docs into it? Probably not. So this is a desktop app that runs locally. And so my thinking with Scanstraight was like this is like a cool little micro product that I can set up and won't require a ton of like the ongoing work that I had to deal with with Thin daily where it's like eh, I don't want this to be a project until the end of time. I wanna create this useful thing and then like move on with my life. Upsides to that is like it was a simple enough product where it just isn't always creating work. Downside to that is when you build something like this, you gotta think about like how are people going to discover this on an ongoing basis? I Have like content and whenever I mention scan straight it'll get like a little bump in sales. But there's nothing like organic about its discovery right now. Like there is no team getting it out in front of people so that more people find it. And for in order for it to be like a proper business like you have to think about that. My advantage is I have an audience and so I can like at a given point in time get it in front of people. But like that's not a long term strategy. It is not a business that has legs really on its like on its own unless I just keep plugging it. Interesting things about launching a desktop app. I think most people these days expect continuous improvement from all of their apps because 90% of the things we pay for now are like on a monthly recurring basis. But a desktop app where you like plop down some money once and that's it, people still think you're going to continue improving upon it even though they already paid for it and they like agreed up front. This is what I'm buying and I'm not paying anymore. Like I think our brains as much as maybe sometimes we miss the days of buying desktop software and that just being like this clear cut thing where you make the purchase once and then you never have to pay for it again. Like as much as we long for those days, what we miss out in that environment is like the never ending improvement cycle that now on the best apps is such a big part of buying software. Like you're hitching your wagon to the company that you think will continue developing faster than all of the others so that from the time you buy it to two years down the road, it's actually way better than when you started buying it. And so like I do, I am also exhausted by the number of software subscriptions that I have and all that. But one of the upsides of that is kind of baked in. There should be some expectation of continuous improvement. Being on the other side of this, developing a desktop app, I did learn that like people just kind of expect that from software now. Like if they buy something from you, even if it's a one time purchase, there still is like this expectation of like you're going to keep improving it over time, which I thought was kind of funny. So the why for me on this one was honestly probably legitimacy more than anything else. If this app like isn't going to generate recurring revenue, fundamentally like it's not going to be a big money maker long term. Right. So developing this app was almost like as a Service to the profession. I think because it is a big pain point, I saw, like, a relatively short path for how we could solve it. And it's something that, like, in my mind, is outrageously affordable for the amount of time that I spent in my years doing accounting, like in Acrobat, cropping these janky smartphone images to try to get to something that was useful, usable in a work paper file. But even then they were, like, all different sizes. Like, you have images that will come through as PDF that are humongous, and then you scroll between pages and they're all different pages, whatever, huge pain in the neck. But for me, like, this was, like, I can see a way to get this done. I'm frustrated it doesn't exist. I'm just going to, like, take that privileged position that I have and will it into existence because I think it will be helpful for people. And that, in my opinion, is reason enough to build software. Maybe your audience isn't the accounting profession, but you have an audience. If you have clients, if there is something really painful there, like, think about how you can. Like, maybe there is a way to build a solution for those people that will add legitimacy to what you do. Not necessarily be a huge money maker, but, like, build on that level of respect that you have with your audience. And then the third app that I built earlier this year with a developer I haven't released, and I don't know that I ever will. And what it is is a tax. An AI tax research app. We finished it a few months ago. And my main motivation for going through that was learning some of the stuff we've talked about the last couple of weeks. Embeddings, vector search. If you haven't heard last Friday's podcast, it's huge. Like, it's redefining the software that we use, how that technology works, what we went through in that podcast. But I got to learn all that stuff really, really early days, and that experience for me has been invaluable. I do some AI consulting with software companies, and most software companies are still not turned on to this new paradigm of how software is being developed. And I wouldn't know all that stuff had I not stumbled through it and, like, just, like, muddled through the really fiddly aspects of developing for AI right now. Ultimately, I don't know what happens with that app if we end up shipping it, if we sell it down the road, I don't know. But basically what it was was a research app that has ingested, like, all the IRS publications and is, like, A universal way better version of Control F. So it embeds all of the publications we have, humans that have conditioned the context. That is, if it comes upon a table or upon a worksheet. An actual tax pro has explained that semantically so that it's discoverable when it's. When you search. Because right now, like, if you use a service that, like, chats with your documents, if there's a picture or a table or something like that, it just doesn't understand that we had tax professionals, like, explaining all of those worksheets and tables and all that stuff so that all of it was queryable. Is there a way for me to show it without showing what the URL is? Hang on. Okay, if I show it, you need to figure out what it is. But basically what it would do is you can type in a question about U.S. tax rules. It'll provide what it thinks the answer is. Every single sentence is cited, deep linked into the source PDF. So you click on that citation, it will take you to the exact page of the publication, like down to the paragraph. So literally everything it's providing as an answer is a direct quote from an IRS publication. So it gives you kind of initial, like what it thinks the answer is. But then it gives you another tab that is a list of all the exceptions. Same thing there, citations to the source for all of those exceptions that may impact the right answer. And then a third tab that's just other considerations, that is a bullet point list of what things may have impacted the correctness of this answer that maybe weren't provided. Because oftentimes the questions with tax research is like, you didn't give a complete information for it to give you a correct answer in return. And so it outlines, like, all of the adjacent things that could change the initial answer that it gave. Again, all those are direct citations, deep links into the actual publications. And it's why now I'm so opinionated about the fact that, you know, if you have somebody doing technical research, whether it's an accountant or tax preparer or an attorney, they don't want an answer from an AI. They want the AI to provide all the source documentation to help them get to their own answer in the most efficient manner possible. Because there's ultimately no level of smartness that AI will reach that will make it an authoritative source. So, like, whether it's today, whether it's 10 years down the road, when it's way better and it knows way more, just getting an answer from AI will never be enough, will never absolve Me from liability. So, like, helpful AI within that context is not a really compelling answer. It is maybe a really compelling answer backed by here's why this is. Here are all the reasons why this is correct. And so ultimately, like, everything, there has to be citations based, has to. Has to pull from a huge. Like, what are all the possible, you know, I say authoritative sources. I guess a pub isn't an authoritative source, but, like, what are all of the most relevant sources that it's pulling from? Not just the Internet. How do you actually build that into an app today? And honestly, like, there still are not any research apps out there doing this in a meaningful way right now. And we'll get there. But for me, like, what made this worthwhile for me, why I went through the work of building it was just to learn the nuts and the bolts and of how all that stuff works. And as a result, I've been able to, like, transfer that learning into the other domains and other types of tools that we use. And we talk about it on this show and we get it out there and we inform other developers and, like, we move the profession forward as a result, because we're having those conversations. So, like, there's a version of this and software development where, like, it may be helpful enough for you just to get a better understanding of what's happening behind the scenes. Obviously my role's a little weird in the, like, the thought leadership stuff that I do. But for you, if you're wading into this stuff, like, for most people, these projects are like bootstrap kind of side hustle sort of things. Like, there may. You may have your own reason for doing this that is completely legitimate. Maybe it's. How does AI impact this specific part of what my clients do? Like, maybe that's reason enough to learn more about this stuff. But I love seeing people wade into this stuff because I think there's a bunch of reasons to explore it. And it's another example of one of those things where nobody does this, nobody identifies as the type of person who does it until they do it. And then you get there and you're like, oh, this is kind of weird. We're like, we're now on the other side of this. Like Finn Daly, I ended up selling to Luke Templin. It's still an app you can go out there and pick up. I think I sold that for, you know, 25 grand or something like that. Like, I never would have imagined that I would be, like, selling software. Like, that's. I would have never expected that. Now how, like how to wade into this? I think the a big blocker for people is like, well, I'm not a developer, like how do I even kind of start going into this? Most of us like, and this is an example of where developing your professional network is really helpful. Most of us have those people in our networks who are open to like doing a side project to being like a technical co founder. I will say the more explicit your expectation of what it is that you want, the more targeted that partnership can be. So like the more clear and more explicit of a vision you have for exactly what you want, the easier that development journey will be. The price to do all this stuff has come down a lot. Like consider, is there a no code expert out there who could build like an mvp, like an initial really basic version of this on the cheap for mobile apps? Like that's definitely the case for desktop apps in most situations. That's the case these days. Like that initial janky MVP which is where you should start, can be billed in a much more cost effective way than ever before. You know, Scan straight, that was like a landing page that would cost me like $9 a month and it just connects to Stripe. And then when somebody buys it, like it emails them a link to download the program. Like it is simple as can be and you should absolutely start there. Like, like everything else we do, I think we kind of over engineer and we think of all the things it could ever need and oftentimes that becomes the stopper to just getting started and waiting into it. Think about like what is a really janky MVP version of this app? Is it something that could even like live in Excel or like some, you know, other really basic form? That's the place to start. But also think long term, is how does this thing grow? Like is it tied to you? Like do you have to like roll that boulder up the hill yourself? Will people find it organically? Is it worthwhile as a thing to just have like adjacent to your accounting firm, which will make you look like bigger experts and will attract new people to like assign a different level of expertise to you? There's honestly there's some really great like non financial reasons for wading into this stuff, you know, from adding legitimacy to what you're doing or for you to just develop a better understanding of how that problem could be solved behind the scenes. That's kind of been, that's kind of been my journey. Like really only the first time I did that was it like this is the thing that could make money and be fun to do. After that it was just like, I want to learn or this is a thing that will help my audience and like add legitimacy to all the things that I do. So it's perfectly okay to find your own reason to wade into that stuff. Curious if you've had like, if this is something that you have thought about, like developing software for clients or for accountants or something like that. Drop a comment. I would love to hear kind of what your mental model is for whether this is worthwhile for you or not. Ask around. I bet you got developers in your network. I've got a buddy that's a developer. We basically would just work out a deal to where like the first X dollars in revenue would go direct to him. He had an appetite for, you know, building something from scratch, something that he had equity in. I think with a little bit of like shaking that network tree to see what falls out, there's probably somebody there that could help you with that. Or if they're here, even if you're a developer, like drop a comment below. Parting Thoughts Interesting paper that came out a week or two ago. It's kind of been catching up on my reading since getting back from the conference. You might have seen this. I will put a link to it in the show notes. Basically it's called Voyager. It is a AI model that learned how to play Minecraft better than any other AI models before. This is not a Minecraft podcast. So why are you telling me about this? Really interesting. The whole idea here is they develop this model that will basically like master the application like in a self guided way where it will go out and learn things it'll discover. It'll build kind of these libraries of actions just based on like testing and kind of this reward feedback loop and video games. Like there's a very explicit kind of reward feedback loop where maybe that's a little bit harder thing to create and like business software for example. But this is a different approach on agents which we have talked about in the past. Agents that can use your software and like are one of the really exciting kind of near term AI developments. They basically act as human users and go in and complete tasks for us. But this is like a different approach to developing agents that will themselves learn how to use an application that can ultimately be leveraged by any of us. Right. And so some of the applications we've talked about for agents in the past is our accounting software. Like is it really that hard for AI to navigate a software program and reconcile a bank account? Something like that? Like what are the little specific individual skills that agents can begin doing for us. Because ultimately I think the future of what we're doing looks like. If you think about your task management system right now we have ultimately then agent users in addition to human users. And you can assign tasks to these agents who will go out and complete the tasks in the same way that we assign tasks to human users who also go out and complete the tasks. So interesting paper here. I try to really pay close attention to any kind of new agent methodologies because when we think about how accounting gets done, when we think about using like tax apps, especially in the US we're using legacy tax software. Think about an agent that sits on top of your tax software. Everything that needs to be known to like, know, like, where are all the inputs for all these different things. This is an approach to modernizing what is a really hairy, really ugly legacy bit of software. Interesting study. I'll put a link to it in the show notes. But something else I'm really hot on that I've been doing more of is the ask your PDF plugin in ChatGPT. I was kind of sour on like PDF chat applications for a while and it's because all the ones that I had been using were using GPT 3.5 because they were like free apps and GPT 4 is expensive but now built into GPT plus if you have the paid version of ChatGPT. You may have seen me talking about this because like I pulled the PDF for, you know, the new legislation that we have in the US and said like, are there any tax things here that would impact my clients? And like, it gave me a really helpful bullet point list that I could drill deeper into, learn that like it impacts IRS funding this new bill, but basically you have a big old PDF. This is a really helpful way to get information out of it. And right now with like new AI studies coming out every day that are like 20 to 100 pages, I'm not sitting down and reading all of those things. So I checked a URL, the URL to the PDF of this study into ask your PDF and just ask it some basic questions. Like in this study, were any references made to how this agent methodology could be used in other software? And they're using like Minecraft in this study. But like the exciting thing here is is this something you can plug into just about anything and the agent learns how to to use the software, Right? So the paper does not explicitly discuss the application of the agent's methodology and other software. However, it does mention several key components of the agent's methodology that could potentially be applied elsewhere. So five things here. Skill library. The agent used a skill library to accumulate knowledge which could be applied to other software to allow an AI agent to build up a repertoire of actions or operations it can perform. So this, this agent methodology has kind of built in long term memory where it never forgets how to do these things. Like not in the same way that ChatGPT has context limits. So it developed for itself a skill library where when it did a thing correctly, it now adds this thing to like this database of skill sets that it has. Obviously that could be a bunch of things, you know, in a, in a accounting application, from how to navigate to a specific page to how to reconcile a bank account and tax software for, you know, how to input box 2 from a W2 into the software. Like it could be stuff as simple as that. Second thing it lists out here, self verification. The agent uses self verification to assess task success. This could be applied to other software. So that is, how does it know that it did it right? That is one area where like video games are particularly suited for this sort of like initial testing is self verification, knowing that it did it right. Like I think here is maybe a little easier than business applications. Where did that thing it just do? Like, was that right? That may be a little bit more ambiguous. Third, automatic curriculum. The agent uses an automatic curriculum for open ended exploration. This could be applied to other software to allow an AI agent to learn and adapt to new tasks over time. So how does it decide all of the things that it will learn, all the skills that will develop? This is the thing that it's actually doing all on its own, like nobody's telling it, go run over here and do this, go run over there and do that. It is like for itself developing that curriculum to learn. Fourth, multimodal feedback from humans. The agent can integrate human feedback in two ways. One, as a critic, so that is a human user, like serving as the critic to the thing that it just did. And two, as a curriculum. So the human could outline, here are all the things that I want you to do, and this is how it learns. And then fifth, code generation, with execution, the agent's generating code for these, for each of these skills that it develops and then executing that code. So talking about Minecraft here, but like absolutely the sort of thing that can be transferable to not just Minecraft, but the other applications that we use as we develop these agents that are capable of using the same software programs that we do, working on the same desktop computer that we do, developing shared skill libraries. Because again, we are all doing the same things largely in parallel. What are the ways that agents can go out and develop these skills that we can then all share to make us more productive in the stuff that we are doing? Gang, I think that stuff is just around the corner. Like that is exciting. It is not scary to me because we did that thought experiment a couple weeks ago. That is, that was like, okay, what if I like worst case scenario for a tax firm? What if it actually will just do all of this stuff? And where we ended up was imagine there's this way better version of TurboTax that can do just about any tax return. If your client went through it themselves in the end, would they have a correct tax return? And I think where we landed was like, probably not. Either because they fell asleep and weren't willing to do the amount of work that it would take to get there. And second, because they bungled the inputs or they did something wrong that got them to a place where like they wouldn't be filing a correct tax return, they still wouldn't be having like an advisor to help them, like make decisions, like tax optimized decisions. All of these sort of things where like we thought through what is, what is that worst case scenario, the darkest timeline where AI actually does all this work in a meaningful way. What if that dropped tomorrow and where we got was, we would be just fine and there would still be a lot of work to do. So somebody drops an agent for QuickBooks tomorrow that will reconcile the bank and all that. We should do this thought experiment for accounting practices as well. We did it for tax practice, but right now I think there's plenty for us to do. So as far as I'm concerned, I can't get my hands on this type of agent soon enough to just take a lot of the grunt work out of the stuff that we're doing. That's it for today. Thanks for coming and hanging. We got Old Timey Tuesday coming up tomorrow, Q and A Wednesday after that. Thanks for hanging. I'll see you tomorrow.