Go back

Everyone’s Using AI Wrong in Their Practice | The AI Watch #01

61m 4s

Everyone’s Using AI Wrong in Their Practice | The AI Watch #01

The podcast introduces a new AI-focused series that emphasizes real-world, practical applications of artificial intelligence in accounting practices. Instead of promoting AI as a replacement for human professionals, the hosts advocate using it to automate mundane, time-consuming tasks—such as extracting and organizing meeting notes or cleaning financial data—so accountants can work more efficiently. A key example is an automated tool that scans meeting notes from a platform like Plaud, summarizes key points, and posts them directly into the practice management system, freeing time and improving client communication. The hosts also share how they built a custom payroll onboarding workflow using AI and API integrations with secure, compliant systems, saving significant time and reducing manual errors. They stress that these tools are designed for internal use, not for sale, and are built to solve specific, personalized problems. The process involves iterative testing, clear prompts, and structured workflows using AI assistants like Claude and Lovable. Importantly, they emphasize data security, compliance with GDPR and Guernsey laws, and the need for human oversight. The core message is that small, incremental AI projects—starting with simple, manageable problems—can lead to meaningful improvements in workflow and client relationships. The creators recommend starting with a single pain point, such as organizing external links or flattening financial files, before scaling up. This approach enables practices to gain control over their operations, boost team efficiency, and build stronger client trust—all without requiring major investments or replacing existing software.

Transcription

11495 Words, 59891 Characters

English
This is us doing an intro. Hello, hello, hello, hello, very much to ask the account of what series we are doing a new brand new series. This could technically be the second one of the AI. I know it's the first time me and you've done it, but me and Johan went off and did a little AI update once. It's come through. But we want to turn this on its head. We want to do an AI watch series. So clearly we're going to be talking about AI. But we want to keep it practical. We want to talk about the things we're doing in our firm that's got AI goodness on it. Not just software, but other solutions and everything else that goes with it. And we thought, why, how, how's the best way to tackle it? And obviously it's got to be doing a podcast and if I'm going to do a podcast. I need the person who's basically looking to take over my seat. As if you go by the last, as the accountant appearance, it came through. Hashtag, give, give, give Tim the job full time was going mega at the time. So Tim, how are you? And why are you here? I'm good. I'm, you know, like everyone, I've been far too hot. And the, you know, the weather has been painful. I'm incredibly tired. We've had staff holidays. We've had people sick. We've had all sorts going on. But apart from that, I'm fantastic. And I think I, I would note that the person shouting the loudest for Hashtag Tim for host was you. I don't know if it's so much that I'm coming for you on the podcast or if you're actively trying to give it up. But I think, you know, the, the pattern number 200 falls on a bank holiday. He's not going to, not going to help or, or, or, or advance anyone's prospects. So let's just, you know, let's see how that goes. Yeah, looking forward to that. I mean, are you going to be involved? Have you decided? I'll be free. I'm like, let's face it. It's a bank call in Monday and it'll be the morning. I'll be at work quite nice. So, you know, the stop date. So, you know, I'll be in the office on my own. And what else am I possibly going to do in my time other than, you know, joining the, the, almost like podcast marathon, almost. That's what we're going for. Yeah. We're going to be talking top trumps. We're going to be talking all sorts. So, yeah, definitely get yourself on there. Let's just do a bit of housekeeping while we're here. We're live to the world. We've chosen six o'clock on a Thursday. I'm not sure that's the date that these watch series will continue to go for. Maybe go back to the dinner time slot. We'll try and figure out. We'll try and find this slot that works for you. But obviously, it's evergreen content. You're going to be able to comment down below. If you are in the lurking in there, the six or so of you there, then do use the comment section to say hi. And I'm all made sure that we bring your comment on. But the whole point of this is. We're going to put these out on semi semi regular basis just to give some updates on how we use night AI, what we're using it for. What good signs we've had, what bad signs we've had. And it wouldn't be a podcast for a controversial start. Let's say controversial tip. Just explain what your view on the use of AI is at this point in time because I feel like you're you're you wanted to say a bit of truth, right? The sort of stuff that we're not. It's like we're not we don't want to admit we don't want to talk about. And hi to the lady, but keepers just said hi in the comment, but there was definitely there's definitely something you want to get off your chest about AI, isn't it? Go for it. Yeah, and before I do, I think it's worth really being specific about the date that we're that we're initially having this conversation. It was 20th of August, 2026, and the reason I'm saying that is that if you go back and listen to the last six months, where for us the accountant even and other podcasts in the in the sector, yeah, everything is changing so fast at the moment. And there's an analogy to that that I always find quite interesting, which is back in the early 2000s when I was first, you know, in the workforce, brilliant. I put the phone on silent and it didn't work. And when I could have had a live you answered and clearly. Yeah, exactly. So you've got AI looking after your telephone done. I will take that box. Whoever's doing the AI bingo out there. Because I've got a new phone and I like I thought I'd put it on silent and I haven't. So the reason that I mentioned this back when I was like first at work, it was computing power, computing power, approximately doubled every. And that meant that you got, but it was really noticeable because it was doubling and doubling and doubling and doubling every single year, everything got more powerful, everything became more impressive. And that obviously slowed right and the speed at which things doubled became more incremental and you can't really see what was going on. AI is now at the point that we were in 2002, 2003 with just like what's the latest Pentium is it a three or is it a four. Okay, so the reason I mentioned the day is because if you go back and watch this in three months time, everything may have changed dramatically. And that's why I mentioned it. So my view on AI at the moment, because I surround myself with a lot of people were very pro AI and a lot of people were very anti. And there are some bits of it that I hate like I really hate the way that every single harvest got the exact same or had the exact same World Cup posters because they've all done them on like. I think we using the wrong thing, okay, we we the industry is trying to use AI to replace us. It's trying to replace bookkeepers. It's trying to release accountants and it's saying, ah, you know, I built this whole team of agents and I've given them all keepsy little names. Doris does the credit control and you know, barrel does the payables and Ian does the bank cracks and you know, they will give them human names and I find that weird. But they're your agents called and what you want. But I think what we should be using AI for really is to replace the non accounting and by keeping things that we do all the time, okay. So that doesn't mean we're replacing all the administrators that doesn't mean we're I'm replacing Emma right it means that I'm replacing. I need to scan this paper and file it somewhere. Let's how can we use AI to speed that process up and make sure that information is extracted in the right places. So that when you're searching for stuff, you can find it, okay, how do we make sure that, you know, how do we check over an incorporation before we file it. How do we do this, this is those are the things where we can use AI not to replace people, but to save us time and turn a one hour job into a five minute job. We're still doing the checking at the end of it, but we're not, you know, we're not trying to replace a human with that. So we're having not a second pair of eyes, but a third pair of eyes because I tell you what, when you run a draft set of accounts or something through something that's built for it like Claude or whatever. Making sure of course, Johan's usual caveat, make sure you're subscribed to a plan that does not mean you're preaching GDPR, that's your responsibility not mine. It will find things that you might have missed and another human might have missed. Maybe they're in consequential things, but it's flagging those up, right. So this is what I mean, we're not, I don't think we should be using AI to replace any human beings. We should be using AI to make some of those human beings more efficient and make some of those boring tasks that you do more efficient. And the second thing, the second point of that, which is a much shorter point, which is more I've been doing is using AI to do something that I can't do without spending tens of thousands pounds, which is build internal tools, some of them very, very small. Yep, so I'll me build something that helps me do what I want to do. I love that. And we're definitely going to go into that second point more detail, but that first point, what I also love about that is. It's so much more relatable to client, look, if we went to clients and said, oh, by the way, I have just automated all of the bookkeeping blah blah blah blah, the first question can come out of their mind, their mouth is. So what's that going to do to my big because they're going to be fee sensitive, right? Like that's a conversation that's going to come up. But if I go to a client and a great example is this week, I love my note taken out, we all have our own right, I'm using plaud for mine, that's what it record it through. But the end of every single day, I have to go through the plaud and it's champagne problems in there, but you go through plaud job or meetings and we have a policy that all meetings, we have a client, we record it on our parchment solution, so that everyone can see it and we get a nice way of being able to have accountability and all that sort of stuff comes with it right. So today, for example, I have had what was it six meetings today, I'm doing a podcast now, I won't be able to even comprehend looking at that until whatever time it's going to be right. Or I've built in this nice all agent that will go in, it talks to plaud, it finds out what's happened, it looks through all the things there and it goes into my practice management taught and it will post it there for me. And then what it will do is just ask me, is it happy with these bits and kind of give me a chance to confirm. So instead of me having to copy paste just blah, blah, blah, blah, I'm in a process now of just approving what's come through and move to that now. there to take away a really or in task that no one benefits. it from, but I'm able to automate it, make it more simple. Hopefully means it's done quicker as well, right? Like you're in a world where you've got real time information because it's happening on a more regular basis. I can't forget, you know, there's all those sort of controls that come in, but well, but I think more importantly than if else, everybody had a conversation with a client about that today because they were talking about their note taker and how they're frustrated with their note taker because it was, you know, they were just going to avoid and didn't feel like it was doing anything, but we were going to have a real in-depth conversation about, well, why are you just pointing to a point? If you're just pointing to avoid, how's your team going to know about that? And I spoke about the fact that we have this management solution and everyone has access to it. Surely you'll see our own solution. Oh, yeah, my solution has a same sort of thing. And I told you about the fact that you can automate it and bring it all in. Absolutely loved it, right? That's a proper meaningful conversation that a, they now have even more kind of faith, whatever it is in me and in my firm, because we've solved a problem for them. But we're building that relationship, right? We're giving them a reason to continue spending the money with us, continue being our client. You know what I mean? Like, these, these little things, we're, you know, we're becoming experts in those fields. And that's what Kanton's book keepers always were. We are problem solvers at the end of the day. That's what we're, we're best at. And I feel like this is just an opportunity for us to flex those muscles, you know, show off some of these ideas. And generally it's helping maintain client relationships. Maybe you win clients later down the line. Have you had any examples where you've been able to do that with clients? Well, because of our client base, a lot of them are obviously hospitality. Yeah. And a lot of things that we will talk about regularly on podcasts and, you know, at events and things like that. They maybe don't apply so much to that sector. What, what has been useful is being able to process their information more quickly, right? So their information. Okay. And I would make one point about what you just said, which is that by having an AI tool extract all your meeting notes deal with them and put them into wherever you want to put them. You've had six meetings then you're now doing a podcast, right? And then you might actually be traveling. You might be going home, right? And you might guess that kind of trap your ham and you might get home. And then the thing you were going to have this chicken for dinner, but it turns out it's gone off. So you've got to go to the shop and get something else and then it gets to like half past 10 and you're off to sleep and you still haven't done it. And then you're not in in the morning because you've got three full meetings and somebody else then doesn't know about your meeting from today because you haven't had a chance to do it yet, right? That happens all the time. That's the sort of thing, especially as a practice owner or a manager of any type, but across all levels of any business, oh, weirdly, I didn't have time to do that last night because circumstances. Okay. So that means AI is helping us overcome irritating circumstances, right? And a human factor circumstances that happen. Sometimes you might be late, you might be sick, you might forget. It's stopping that bottleneck. Because otherwise then someone's calling you in the morning going, oh my God, I really, really, really need your notes for last night about X and you're like, ah, I'm driving right now. I can't send you that for the next hour. I'm really sorry. I'll try and find something to pull over. I've got no signal. This is all the real world stuff. If you've got something overnight just going into your applaud account and taking everything and knowing what client it is and posting a summary, even if it doesn't get it 100% right all at the time, that call from your team member, then is Aaron, did you really tell this person that their fee was going up to a billion pounds? No, it's misheard. It's misheard. That was a joke. It's 1200 quid. Oh, right. Okay. It's minor adjustments that you can verbally tell them. I think this is some of the prime use of this technology I really do, especially for those super, super busy and trying to do 12 things at once. Exactly. We're all trying to do it. We're all trying to build our practice in the best possible way. We're all under so much pressure from our professional buddies, from our government, whatever reasons and problems and excuses you want to put out there. We're all in the same boat. We're all super busy. There's so much like red tape for us to do at this point in time. We need to have these quick wins. Otherwise, I jokingly opened this conversation with you before. I'm asking you, remind me again why we're in practice. Sometimes we need that. We just need that kick off the backside. We need that reassurance. We need something else. But at the same time, we also just need those quick wins because you just don't switch after you. Never. Never. Yeah. All right. So that's just an example of how we can just use them to be there. And we've got those more examples. We'll definitely bring more up and we'll talk about them. But I do need to move on to that second point you made because you did a presentation at a recent accountant advisory group over in sage. And it was an amazing presentation like the idea was you were going to stand up on stay at the front of the room and you were going to just explain how you've been able to use AI to be able to help you fix some of your client's biggest issues, right? And find ways to be able to go in there. And I won't build still your fund on how and why I never know. But what really kind of struck me on that was it. It just got my brain back into gear of I've been wanting to do that for a long time. But I kept putting it off and I kept doing it out. And the reason I'm bringing the story up is because on the way home from that accounting council. So the next day, jumping on a train, I kind of had that kick off the backside to do exactly what you're about to explain and how it's done then for now. And I went, I've been I've been such a long time that I've been wanting to build this particular product for my practice, which is work papers, like I've been looking everywhere for work papers, right? And I was able to on the train journey home build the framework for my work papers that we now use in our practice today. And I just needed that one little bit of kick up the ass to go, right, let's let's actually go ahead and do this. Now for full disclosure, that one train journey was the kind of the framework of it. And most of that framework is still in play now. It took me three, four weeks or so after that initial bit of spark to go through the security and setting up like permissions and making sure the users are right and all that sort of stuff. And I have had to also get external help just to make sure that it's definitely, you know, it's kind of data and they write want to be responsible and everything else. But that initial train journey home was all it took for me to have that kick up the backside to go, right, let's build it. And now we are in a position where we, you know, we feel like we're way more in control than we've ever been. We feel like we're able to take on more work. You know, there's all these positives come with it. And our team of all like energized by this new piece of software that's coming because we've built it to make it easier for them, etc, etc. So I just wanted to give you kind of the, or give you kind of the start the starting point of why I think what you're about to say, like explain how and what you've been doing. It's so impactful and how honestly anyone out there can do it. Anyone out there can find the solution, find it, find a problem, shall I say, find a pain point and use this technology, have a play with this technology to see if you can fix it as well. Because it's generally changed, changed our practice. I think that there's obviously a couple of things to just put in at the beginning. Number one is that I am not building any software to sell. Okay. You want to be a software to sell. You can do it like this. But then you really do need to speak to someone who knows what they're doing to get that production ready. Okay. One of the things that falls like Claude and loveable are perfect for prototyping. Okay. And for building internal tools. Now, I also work with, I've got a couple of developers that I know that I do a bit of work with who before any of this becomes like live and we're actually using it live in our practice, you know, they're going to, they're going to give it a once-aver. They're going to tell me that it's terrible and they need to be building. And they're going to then rebuild it. But there are some things that are so, so small and the small bits of the starting point. Okay. Yeah. And I also think, look, it will never hurt to just mess around with it and see how you go on. The time that you need to start thinking about security, excuse me, proper security, and penetration testing, all these things that might sound scary, but you can learn about them quite easily. That's what you do before you put any real client data in something. Okay. You can. People say, I'd never build my own tool because I really want to. And I've got this great idea, but I'm really scared that it's going to leak all my clients information. Great. Put fake client information in them while you're while you're messing around with it, right? I have got a system full of Tim T_test0091011ZZZZZZZZY, etc. And actually, because I've done, I'm actually beyond that. And I've done a lot of that security work and everything. And I have got real client data, which I'm obviously not going to show. But the initial part was just like, there's about 700 of me in that system, right? right? Because you set the email address up with the Tim Quartz, something else at TCR. And, and, and that's how you get through that. The other thing that I would say is this ain't a workshop. I'm not teaching you anything on this. I'm giving you an idea, basically. And my preferred method of working is that I use Claude and I use Lovable and I use those together. But I haven't increasingly connected them together. And what I do is I spec things out in Claude and I build the prompts in Claude and then I pay copy and paste that into Lovable. Lovable does this thing. And then it sends back a response to Claude and then it helps me interpret that result. That's because I am not a Coda, right? I cannot write a line of code. I cannot get to a Hello World. I've never been able to. It really doesn't stick in my brain. That's OK. Because no, I don't need to. Right. And some of this stuff I've been wanting to build for years. Because it's possibly never going to ever be a commercial product. There are things that I want to do in our practice. I've pre-wound our friend Johan about this. I am building my own practice management software. It's only for me. Right. It's only for TCS and 65 is. I'm never going to sell it. I still think the Engager is one of the best practice management tools on the market. I would still encourage smaller practices, especially but of, you know, any reasonable size. To seriously consider using Engager if they're not currently not a paid ad hashtag not an ad, right. But we have some really, really niche requirements that I would never expect Engager to even entertain the thought of building them in because it's not commercially viable for them to do so. OK. And I'll give you some examples. But first, I want to give you one example of just a really, really super, super simple thing that I don't. OK. And it took using the methodology of Claude Plus Loveable. This took about 15 minutes. OK. So we're going to attempt. I've never screen shared on Streamyard before. So if this goes horribly wrong, then I blame Streamyard, I guess. You know, I'm worried about your preferences on, I think normally you have to come in and out, but we'll see how it goes. We'll see how it goes. We're going to find out that it's worked already. And hopefully you should see, there you go. Right. Ignore my side. This, I was getting frustrated when you export nominal account transactions into a CSV from sage. If I was sent to load pad, I'm ready to share your life. If you export nominal transactions, you get formatting, right. You get you get stuff at the top, you get stuff at the bottom, you get totals, you get all sorts of nonsense that you don't want. Sometimes I want to do other stuff with that data, and I need it to be not formatted. And if it's one account, right. If it's, if it's just the one, that's absolutely fine. OK. We can all we can all manage to reform up a download for one account. But my problem is that it's never just one account. Is it it's three accounts or five accounts or ten accounts. And if I, if I want to format those and then I want to pull them something else or to pivot or whatever I want to do. Our most common example of this is for the States of Gernsey for our payroll control accounts. We generally have three accounts. ETI payable employers, social security payable employees, social security payable. And I want to really easily see all of the transactions for a whole entire year. So I made a flat file converter and all it does is it makes the file that you have exported flat. It can do it with a normal export and do it with trial balance export. And the way I did this is I went to Claude and I said, I have a file that looks like this and I showed it the file. I manually made a flat file version and I said, I want you to use loveable to build me a tool that turned file a dot XLS into file B dot XLS. Ask me the questions that you need to ask to find out what you need to know. And it did it said, OK, zero, zero. You can practice. Are these all going to be financial information. Have you considered this? What about this? How do I make it flat? What do you mean when you say flat? What do you want in there? Do you want any options? Do you want XYZ? OK. And so it now will do exactly what I want. And that's why it will never be a commercial piece of software and I'll never sell it because it's doing what I want and it might be that absolutely nobody needs this. But I can drop a CSV file into this into this and hit download and it extracted 901 transactions there in about half a second and it's a flat file. OK. And now if I I have to change screen share, don't I? That'd be really annoying. It exports the file in a format that I can have multiple accounts over multiple years. And guess what I can just grab that and pivot table it and do what I want with it. Love that. Love that. So great example of something that only took a few minutes. Now, in Claude specifically. The way I've done that, there is some organization here. OK. And this is the bit that this is what I really wanted to sort of show a bit on today's episode right is the organization behind it. OK. What I've got here is a in Claude, I've set up a project to deal with this. OK. And in that project, I've got different chats and they work on different things. I also have and this is super important once you start doing stuff once you start working with Claude like this. I've got a project overview. I've got a structure document, which is just like, hey, this is what this system does. And then I've got memory. Now memory is the super, super, super important bit memory and instructions for anyone who's doing this to stop Claude trying to be your friend. And the reason you want Claude to stop trying to be your friend is you need it to work with your working style. So for me, for example, I've given it a rule that's in the instructions and in the memory that says never, ever. Give me a brief to pace to lovable and then continue asking me stuff, always put the brief at the end. That's because I know that if it keeps asking me a load of waffle, I'll ignore the brief. I'll forget it's there or worse still. I'll download the brief, send it to lovable and then read the rest. And it will say in there somewhere now, of course, you should only only paste this brief into lovable if you're sure you want everything to be pink. And it turns out I didn't want it all to be pink wanted it to be red, right. And then I've wasted credits and I've got to go back and there again. So that's and you can edit this as you go along right. So this is. These are my working style notes. It's very straightforward. Who are you working with? I do not write code, but I do know what I'm talking about don't give me walls of text because I can't read them, give me bullet points and so on and so forth. And so this is one of the things when people start to do work like this and then they go, oh, God, actually, it was too difficult. I gave up. It never did what I wanted. You need this. You need the memory. You need the instructions. And then just be clear with it as well. A while ago, I decided that what I really wanted to do is do the quick xenon connect API integration, right. So I said to Claude, a few chats back. I said, I want to build a xenon API connection. I've spoken to xenon and I'd got the API docs. And I said, he grew to his API docs. How do I do it? And it then gave me all the reasoning and it was like, OK, what do I do? What do I do? What do I want to do? And then it asked me a question of what I wanted to do first and I told it. And then it built out from there. Now that helped me focus on the order in which I wanted to do things. Because if I had to simply go off and read xenon's API documents and then work out which order I needed to do things that I would never do it. I would have given up. This is one of the primary uses of AI because it can go and learn and it can work out. You can tell it. Work out from me what order I should do things in should I do this first or should I do this first is what I'm trying to achieve achievable tell it to be honest. It will be honest. It might occasionally lead you down a bit of a garden path. That's fine. That happens. It's still less time than you learning all this new stuff by yourself. And I think the best way I would like give people a way to just trial this right because all of this. We should really start with this is a solution where we're told that it's called loveable right. If I just quickly give you a bit of a share screen of what loveable looks like that sort of thing just to give you a bit of a like a bit of a bit of an insight into what we're talking about right. So this here on the screen here is loveable. Left hand side is very much like any other chat box you've ever seen in your life right. It's on the left hand side you're talking with et cetera, et cetera. And on the right is a live demo of that particular thing that you're just that website you'll build it. And there's a fancy little publish button so that you can publish it to be an actual website living on the on the internet for you to enter out with right now that bit that Tim was talking about there. So important because you could just use loveable and you could just put prompt in here you could say building here's a here's the. an interactive version of our guest. our top-traum game, right? I know highly recommend finding a little project like this where it's not going to break, you know, you can't break this, nothing's going to go right or wrong, you're just having to play, right? You're just playing with the tools to see how comfortable you are with it. So try and build yourself a little game or a little something, right? A little something interactive that you can go with. So I chose to build the interactive asset counting game, and I could have just wrote a one thing. In fact, I think I might have done this at the very beginning, where I just wrote my own prompt and just sent it off and yeah, it came out with an okay solution, and it seemed okay, it seemed fine. But if you take Tim's advice and you go into Claude, we can have a back-forge conversation with Claude, and you can start asking and it will ask you questions, right? Tim, it will ask you, don't you think you should consider this? Don't you think you should consider that? And it's kind of going back and forth and I've announced. And like, Tim showed you, like, wouldn't you've got that set up in a project that you're going to make it exactly how it is? You put the information here, but then don't stop that. Take a screenshot of maybe there's a problem missing or something not quite right or something like that. Go back to Claude and say, whoa, whoa, whoa, this isn't working or I press this button, it doesn't seem to work. Claude will then figure that bit out for you, figure out the answers, figure out the questions, and push it straight through. I think it was that bit for me, Tim, when you kind of, like, really hammered home the idea that, well, let's use Claude independently and let that be the prompt to be older for you, that change the game for me in terms of building this. Because if you just do things and actually the Lady Bookkeeper may say, I made my own practice management platform available, not nearly as complex as engage with it. And that's absolutely fantastic to hear. And I think that the difference in doing it, jumped in lovable and doing it in lovable and Claude is, if lovable gets in, all of these AI tools, all of these vibe coding tools, they have a tendency sometimes to get themselves in a bit of a loop and think that they've done a good job when they haven't sometimes, not all the time, sometimes. And it then depends on you trying to pick that apart and work out what it's done, whereas if I know that lovable has actually misunderstood, it's done something weird, I can be in Claude and say, right, here's the outcome of the game, ask lovable why it's doing that, probe it, get the source, tell it not to make any changes, find out what it's doing. Claude will then build a prompt for lovable that's a diagnosis prompt. It will find out what's gone wrong and it will fix it. I had an example where I was doing some stuff with contacts and it was misunderstanding the concept of an external contact versus an external linked individual. So what I mean by that is external contact is like, okay, Tracy, who's the credit controller at the phone company, versus a linked external contact being somebody who does work for one of your clients but is not actually an employee or a director or anything like that. Does that make sense? Yeah, yeah. I'd explained it badly. Now lovable to fix what I'd done in lovable would have taken five or six prompts I guarantee it because I've been there before. With Claude, I was able to say, and I said to Claude, I'm not explaining myself clearly and I did a voice note to Claude instead and I rambled for about 10 minutes on what I meant and it thought and it summarized it and it said, I think what you mean is, this is type A, this is type B, this is type C. It got it right first time, sometimes it's second time and that made it a lot easier and that kind of brings us on interestingly to the cost of this in a way because doing that cost me in terms of money way less than trying to on kick it in lovable. I spent, you know, 35 p fixing that instead of four quid. Yeah. And that's where the strength of Claude really is because it's already kind of knows what the problems are going to be and it asks you those questions before and it's obviously learned from other people doing it before, right? And using it as a plot as a, sorry, lovable prompt generator and it's asking those questions for you and yeah, it works really well and back on money though, we were already seeing a return investment for us because we were spending so much time using other software as that was never really made for it and we were like doing a, you know, round, round shape into a square hole scenario where, you know, the work papers we were using before were okay, but they were, they only took it so far and and for just like, like, I'm happy to frame on the boss. It was QuickBooks's work papers, what we were using before this, right? And we were like, look, yeah, it's not perfect, but it's the best working paper solution I can find at the moment, but the team will always get frustrated because it can quite work and then they've done this update where they put things in the different place and when you downloaded it, it wasn't quite right and we're going to fix the download and we're going to put these, and we had, it was our work papers with an asterix, right? And then this asterix was costing us money, time, resources, whatever you want to put into it. And we were also trying to make it fit for zero clients and sage clients and free agent clients. We were trying to build it all in and we, we, you know, we were just using it, but we were kind of settling on it if that makes sense. And already the time I've spent building this out, getting this out, the train enough done to the team, the binding of it and everything else has come with it. We're already at a point where we've definitely 100% got a return investment now, right? And that's two weeks after we properly launched it. Like, if you do the maths and do the time and do everything else, yeah, there's an upfront cost for the lovable cost. There's an upfront cost for my time developing, there's upfront cost for the, you know, clawed at everything else. I guarantee I've already built that out because the team aren't having to, you know, compromise doing this, compromise doing that. And they're already coming back to me going, I really like this working papers because it does this and this. But why am I pressing this button to get to this bit and why am I pressing this one to it? So we're making these incremental updates on their behalf, which is just going to save X amount of time, right? So yeah, I highly recommend it. Yeah, it doesn't matter how responsive and how good. And we're super lucky, right? Because of the sort of circles that we move within, we're super lucky that we've got close relationships with the owner, the directors, the leaders of some pieces of software that we use on regular basis, where if we make a suggestion and they think it's worthwhile, they'll do it, they'll build it, they'll get it in. But it won't be that fast because they're building real software, it impacts their other customers and clients as well, okay? And some of the things that we want are just going to be daft, like a lot of software comes in light mode or dark mode, right? I want something sort of in the middle of that. So I've got dusk mode, right? I wouldn't expect in gauges then on or anyone to build me dusk mode that's ridiculous. That's just me going, I really like the dark mode, it's a bit too dark. I'm not that much of an emo. I want a sort of twilighty version, right? No one else is going to build that, but I can because all I have to do is say to Claude, right, we've built this with a light mode and a dark mode, but I kind of want a an in between and then it went through and it asked me and it was like, okay, I'm going to show you four shades of black, which one is it that you want to use the base color of this? I was like, ah, the second one's good, right, brilliant, that's what we're going to build on. And that's that's the sort of thing and that that's just a silly example, but a real one that I went through, when it's something functional, when someone's like, I, you know, that'd be better and if that button was huge, so I can't miss it, you can log in and go, make the submit button enormous, make the submit button fill the entire width of the screen. That's what my team want, that's what I want to see and it will happen without breaking anything for anyone else without affecting anyone's workflows, without suddenly logging in and finding out as we did a few weeks ago that certain buttons in a piece of software used to be rounded now they're square, none of that. And that's why it makes and cost wise again, it cost you a few pennies and you do have to really understand the cost of this and not spend thousands and thousands of pounds on it. And obviously depending on your jurisdiction, maybe it's, maybe it can be capitalized, maybe it can't, you know, maybe it's R&D, maybe it's not, but I'm not going to get into that because that's a whole mind field. But I think what I, what I wanted to just, you know, relatively quickly show for this week was, because this is live now and I've done it, we on board a lot of employees for payroll. We run payroll for many hundreds of people per month, most of them weekly, you know, we've got four, five hundred weekly, another four or five hundred monthly, it's a lot. And a lot of these obscenity businesses, they're part-time, they're casual staff, they're people that turn up and do three shifts and never come back. And a lot of these need on boarding. And we've always done this for a selection of forms and we have a workflow and we follow here and we get this and we verify it and then we, you know, set them up on the payroll system, set them up on whichever HR system wherever it is. And I was thinking about how we can streamline that. And at the same time, I was, you know, becoming absolute besties with contents there. Yeah. Number of events there has API access. You can ask for it. You give them a brief outline of what you're trying to do. They're gonna give you the API access. So I thought, well, I don't want to risk taking people's payroll information, people's social security numbers, people's bank details. I don't want to risk collecting that via any methodology that is not secure and data protection compliant, okay? We have a double data protection requirement here because we're in Guernsey, but we also process UK payrolls. We're gonna need to comply with UK GDPR and with the Guernsey data protection law. Which is very similar. In fact, it's almost word for word the same, but there are some nuances and differences. So I was able to speak to contents there and make sure that we're fully compliant. Review their policies, review our privacy policy on all our data protection and make sure that was done. And I just wanted to highlight that you need to do that because it is still your responsibility. You can't go and blame Claude or love a bullet if it turns out that you've caused a data breach. That's gonna be on you. But I've spent the time on it. I've done the research on it. I've spoken to external parties who have verified that for me and I've made sure it's secure enough, okay? We're more than secure enough, very secure. So what I then did is I said, okay, Claude, we're building a payroll portal, okay? And I said, here's our current payroll portal and here's what I want to achieve and here's the workflow that I want to follow. And what it came up with initially was our lovely payroll portal, okay? And obviously we've made it very pretty and it's very sort of TCS branded user aesthetic, okay? The UK one's not quite live yet, so don't try and make me click that 'cause I'm not gonna. But this is why I've been working now, okay? Now, we've always had payroll pages on our website so that we don't have to answer the same questions again and again. So first I incorporated that and I said, right, employ information, you know, what do I need to do if I'm an e-starter? What if I'm a student? How do I get a direction, notice? What do I do about tax and social security? What do I do about population management? These are quite Guernsey-specific, but that's why I needed to do this 'cause there is no onboarding system that already exists for Guernsey payroll, right? So that's why I had to make sure all these questions that they are literally frequently asked questions. So then after what's the best way of doing this? So I created a sign-up web play and once they sign up, it's gonna send them, now I'm already logged in. I've started a, I can start when you start from here and I can say, well, it's me. I'm gonna put a number of contents there that's like powering this, right? Yeah, exactly. So this at the moment is just what I feel, but I'm gonna create my sort of new start account and then what that's gonna do and you won't see this that's gonna happen in the background is it's gonna send me a magic link to a content and request, okay? And once it's in content there, I've got it. I can use it as a to-do list. I don't know, people who don't know content there, check it out. The reason I love it so much is they don't have to finish the formal in one go, it auto saves as they go along and you can pre-fill and you can see what they've done so far and see how far along they are with it, okay? But then once that's all done, I get my content for my request. It's secure, I submit all my information and it's then gonna send me in my sort of practice management side of things, a notification. Now here's what me and Zoe prepared earlier. Obviously none of this is real information, so just to make sure we're all aware of that, this is, you know, 1-1-1-1-1. I've been able to create, I've been able to pull by hitting fetch answers, it's pulled the information that Zoe's given me, okay? I also did some very specific names because I said, okay, I want first and last name, but in contents there, I'm only gonna put their full legal name. Why? Because we have a lot of non-western employees whose names are formatted differently and it might not be obvious what we call a first and last name isn't gonna match, okay? So we let them perk whatever they want and then we guess, but then I can override it if I need to, okay? And I hit fetch answers and because I've built that API connection which was very simple and to make honestly 20 minutes, that contents there request is now pulled all its information through to this employee record. I can then verify that and the way I've chosen to verify that is where players, I said, okay, we've put an inspirations contents there. What else can we do? So I've got a templating contents there for the employer to verify it. So once I've checked it and made sure it's not nonsense, I check that they've chosen the right employer and I hit, and the moment it's to send again, this is what we did earlier, but it's giving me the audit show here, it's giving me the request ID where I've then sent, I've generated a contents there request to the employer that says, hey, Zoe says she's coming to work for you. When did she start? How much you paying her? - Yep, yep. - Is she early or salary? Is she zero hours? Do you want, does she need to be enrolled in the pension scheme or is it deferred? Do you want us to issue your template contracts on behalf of her? Those respect per the employer. So for each employer that I create, they'll have their own template if they're different from everybody else. But all of that, that whole workflow, which is now working and which we've been using this week, in total, took about three hours. - Yeah. - Three hours on a Sunday afternoon with a little bit of frustration and a little bit of annoyance. The first workflow that I tried to build for this, I realized after an hour that I was going down the wrong direction and I rolled it back and I started again because I realized that I had a better idea. But the whole point of this is that I've got that how I want it, right? I can then create an employee record for this person and I can do the things, I can track the things that I want to track instead of the things that the HR software that you buy with shellpants to track. - Yeah, exactly. - I've got things on here that are, we have a lot of client employees who need a weekly deduction for that rent 'cause they live in a spa house, right? How much is that rent, how much are their bills? When does it start an end, right? I am saying that I don't know so his population management reference so I can make myself a note of that and then when I get it, I can set it and if she's called up and said that it's a temporary permit, then I can say, okay, I know that's only valid till the 31st of September, right? So then I can make myself a notification that on the 31st of December, I'll have to, yeah. Now that's not something that exists in any commercial software because it's specifically what I want, right? And that is the real sort of, what's the right word? It's the real like impetus behind this is that this can be specced properly to you. Do I think that everyone should go off and start building all their own tools for absolutely everything? No, absolutely not. - No. - Clency of the tools that we use today today are absolutely fine, right? We sage on a day-to-day basis are there things that are annoying and sage? Yeah, same as every other software. Am I going to build my own version of sage? Absolutely not, right? Are you silverfin to do all our year end accounts? Are there things that irritate me and tend to see in silverfin? Yes, there are, especially when it decides that I've made a mistake that I know I haven't made another story. But am I going to try and build my own version of it? No, God knows, far too complex. I'll make too many mistakes and it will be disaster. But when I need a basic document management system that integrates with SharePoint and allows me to say it belongs to this client, this is the external contact and this is the type of document. Yeah, I can just build that myself now because I don't need to go and spend hundreds and hundreds of pounds on a document management system because I've just built my own, do you see what I mean? Do you see what I'm getting out? I'm sort of rambling here now and I'm going to shut up. But that's just a quick example of something that I built in less than three hours, just under three hours that will in the next, and I'll report back on the next episode, it is going to save us time. It might save us time, it will. Because the final part of this that I'm building this week is that I can batch up all those new staffers and export them to a CSV file in the exact right format to import into the payroll system. So what have I eliminated? Zoe, Ryan, the payroll team are no longer manually keying in new start and information. They're not talking anything anymore. And you're the scope of how far you can take this is only limited to your imagination, right? So depending on what payroll software you use, maybe they've got an API and maybe you could one day just take it straight in. Like that's the sort of thing that it starts to generate, doesn't it? And the point that I'm trying to make with like just building a game or something, is just start on something that is a little bit fun, a little bit of, you know, just a little bit of a creativity spark in there. And that's where this starts from. And I promise you, Bill Smith, the work papers for me was like my big project and I'm really happy with it. And again, it's not finished. And it won't be finished for a long time because I'm getting feedback and everyone's coming in and everyone's giving me the feedback and we're changing things. And, you know, for example, now we now made it so that we put all of our categories down for our accounts production software so that when we export it out, it matches everything up and it's the numbers are done. Like, I didn't think that there's one of the team members. Like, well, why are we then just putting it in and doing it there? I was like, great point, right? Like, you know, and that's great. But then you start thinking of the stuff, like we were talking the other day, weren't we about capacity planning? Capacity planning is a solution. That I don't care what practice management software in the world or task management software in the world. They'll might make a good one. They might make one finding it that's going to be right for your practice. It's going to be really difficult. Like I would imagine every capacity plan in the world or every firm in the world will have a different metric for what they capacity planning to be around. We've got our metrics. We're happy with our metrics to build that capacity planner or to expect a practice management solution to build it for us. It would be kind of ludicrous because I'm not 100% sure they know or that they would be able to even get the bits that we need to put it all into place. Because every practice is so different because we've all got other stuff happening right so I'm looking at a capacity planning, which I am. But if I'm looking at capacity planning for certain members of the team, I need to think about, OK, how much time do they spend out of the office each day doing banking. That's not something that every practice does. Someday, that's not something everyone does. How do I plan for scanning paperwork, right? We have a lot of, we have a lot of paper in with the still because we deal with hospitality and they primarily get delivery notes and invoices on paper, right. How do I deal with that? Like that's not in any capacity planning tool. It's like, oh cool. Yeah, just put it down to happen. So it's not an admin. It's client work. They're spending four minutes scanning for client a and then eight minutes scanning for client B. If I want to take on another 10 clients that are similar to that, does that person have the capacity to do 10 more clients worth of scanning every day. Maybe not. There's someone else have to pass it. That's not something I'm ever going to have in a capacity planning piece. It doesn't matter how much I spend with them. Yeah, we used to use Excel and I love Excel, right. So very clock ups, if statement index matching all right visual basics. I will I will spend hours upon hours, but I was building a spreadsheet. I have done in the past requested plan like and it's it's it's fine. But it's just me that can interpret it. It's just me that can update it. It's just me that can actually get the most out and guess what happens when it's just relying on me. It gets left and before we know it when this big hole and that capacity planning piece of work that keep putting off because I know it's going to take X amount of time to get the all the information back into place or whatever the reason is going to be. Man that I actually caused myself an issue because I haven't been proactive in capacity. I'm just now just trying to backfill in. We all know what happens when we do that right. But we built this capacity planning tool with two two very distinct point of use in mind. First one, the admin team or the management team, you know, myself, the wife and everything else. We get to see all the really sensitive numbers in the background, right. And we get to play around with it and we get to figure it out. And it updates beautifully from our practice management solution updates beautifully from our other bits we've got. And we've got that tool at our fingertips, but crucially for us, we also have a team member facing version. So they can see it. So, you know, we're not doing this in a vacuum anymore capacity planning used to be me in a dark and room one day, just going, I think I've got this right. Yeah, it seems about right to me, right. Like now there's ownership and people, people can see what their capacity budget is, you know, and they can argue it with us or whatever they need to do and they can, you know, and we can build on that and we can take their feedback and go, well, actually, probably we've probably missed this step. And we're saying that about people going to the bank, whatever, maybe we've got the equivalent that we've just not taken every into consideration our bank. And now it's something that we've got to go and do. So, yeah, there's certain software, certain scenarios, certain situations, certain solutions where no one's going to be able to build it for you, unfortunately. And that's exactly where this comes in. And the final thing I tell my point for, you know, because actually we've rambled on for almost an hour now, which is, you know, typical really isn't for a quick podcast. But what I would say is that if you, you know, if you're interested in this, you want to give it a go, start with the problem that you have that's a smallish problem. And start with the biggest problem, because if you start with your biggest problem, and you can't fix it, it'll be bloody awful, right? The problem, the two problems that I started with where I really hate doing these flat files and I really need a place for us to store all the links to external sites that we use all the time in a sensible fashion. And that with, that was the stub of this, that this is what grew into this. When I started doing integrations, the first one was, I want to wait a post a year's worth of prepayments in one go without me having to post 12 separate journals, help me. Right, and it did it, those are problems where if I failed to do that and I had to keep posting prepayments the way we were doing it for no big deal, right, not a real problem, just a nice to have save me, you know, an hour every quarter or whatever, but not a disaster for everyone wrong. If you try and build something that's going to revolutionize your practice and you get it wrong, you're going to just feel sad. So don't sit yourself up for failure, start with a simple problem and get more and more complex as you go. So share that story, right, because between me and you, like what's that message and stuff, we've both been like, oh, ah, you know, I mean, like you just need that conversation spark and everything else that goes with it. So yeah, hopefully that's been useful to everyone because we are vastly running our time. Let us know like going forward, do you want it to be more practical, do you want us to show you the bits that we brought into play and all that sort of stuff, let's know, Marcus says, this has been great guys enjoyed listening to this. Well, thank you very much, Marcus, I mean, he's definitely one who's going to be seeing a lot of enhancements in his new role, right, because that product is changing on a regular basis. I have, it's just to improve something. I tell you what we need, we need, what we need Marcus to tell us one day is how do you keep on top of all these updates, because I now kind of feel sorry for people like I keep releasing, oh, I've just tweaked this or tweaked that on our working papers or whatever it's going to be right. And I'm like, well, I need to tell everyone that. So I'll team them and send it out to him. And then it's like, oh, you know what, like there's got to be better way to do it. So what have I built now a release schedule directly in the, in the, in the work papers. And I'm like, oh, you know what? I feel sorry for anyone who actually does develop commercial software at this sort of pace, because they've, we've all seen that I'm trying to do this in this software, but when I searched the user guide, it sent me to a menu that no longer exists. And I just feel I feel terrible for people that have to keep on top of that, because that must be they, they probably got another tool they build with code to keep them on top of that, right. Yeah, you need a whole team of just customer success that keeps on top of it. And yeah, now I get it. And he says, thank you very much. Can I see a context, we certainly well looking forward to a context of make sure you come to the booth, make sure you come and get yourself some of the new one, I won't show who it is, but we have some brand new top Trump's cards. Yeah, that's always exciting brand new first edition, they've come for full disclosure, then I won't tell you who this one is, producer Lizzie's final job, or for one of her final things that she did before she left, I think she got my message a little bit. So this silver shiny that we have here, I think we got 50 of this one person, so we might need to get rid of that. Yeah, oh, great, great point there Marcus, I have found, because you know, the new Claude can now record and does a kind of stop creation for you, et cetera, et cetera, et cetera, and goes through there. One of my favorite use cases is to let it go, let it, let it listen to what you're doing and you tell it what it's not doing right in the product of your building in there. And then just say any idea to fix this and it will then generate a prompt and off to the races, it goes. Thank you very much for watching, thank you very much for commenting, thank you very much for everything else Tim, what have you got coming up anything you want to plug, plug, plug away. I've nothing to plug, I've done this instead of going to the West show, you've been to the West show now. North J next week, so that's OK, so I'm going to go home, feed the cats and collapse into a heap on the floor, ready to do the same whole entire day again tomorrow, how about you? It can't have been a year then since we came over clearly it is and my own sparkling, isn't it? That does not make any sense at all. Wow, we're on full account text north mode, we're looking forward to it as is the lady bookkeeper as well. So yeah, sure be good, we're getting stuff in the background sorted. producer Kelly is absolutely flying in fill in the no slots. We've already got a full schedule of pits up podcasts. So we're looking at it, including the hundreds. I can't tell you that is yet because we're debating internally if we made a number, but yes, the hundredth is there. The problem is, one of the options could be one of my mortal enemies and I don't feel like that's right. I'll leave you all in the comment section to figure out who that is. Be and you'll be okay. Exactly. Well, thank you so much for watching. Thank you for interacting. Thank you very and for that. I will see you in the next episode. It's a bye for me. And me. Bye bye.

Podcast Summary

Key Points:

  1. The podcast series will focus on practical AI applications in accounting and practice management, not theoretical or controversial debates.
  2. AI is used to automate repetitive tasks like meeting note summarization and data formatting, improving efficiency and reducing human error.
  3. The creators built custom internal tools using AI platforms like Claude and Lovable, solving specific pain points without commercializing or replacing existing software.

Summary:

The podcast introduces a new AI-focused series that emphasizes real-world, practical applications of artificial intelligence in accounting practices. Instead of promoting AI as a replacement for human professionals, the hosts advocate using it to automate mundane, time-consuming tasks—such as extracting and organizing meeting notes or cleaning financial data—so accountants can work more efficiently. A key example is an automated tool that scans meeting notes from a platform like Plaud, summarizes key points, and posts them directly into the practice management system, freeing time and improving client communication.

The hosts also share how they built a custom payroll onboarding workflow using AI and API integrations with secure, compliant systems, saving significant time and reducing manual errors. They stress that these tools are designed for internal use, not for sale, and are built to solve specific, personalized problems. The process involves iterative testing, clear prompts, and structured workflows using AI assistants like Claude and Lovable.

Importantly, they emphasize data security, compliance with GDPR and Guernsey laws, and the need for human oversight. The core message is that small, incremental AI projects—starting with simple, manageable problems—can lead to meaningful improvements in workflow and client relationships. The creators recommend starting with a single pain point, such as organizing external links or flattening financial files, before scaling up.

This approach enables practices to gain control over their operations, boost team efficiency, and build stronger client trust—all without requiring major investments or replacing existing software.

FAQs

The AI Watch series aims to provide practical, real-world insights into how AI is being used in accounting practices, focusing on efficiency, automation, and client relationship building rather than replacing human roles.

AI helps automate repetitive tasks like scanning documents, extracting meeting notes, and formatting financial data, saving time and reducing errors so accountants can focus on higher-value activities.

Yes, accountants can build simple internal tools using AI platforms like Claude and Lovable, even without coding, by leveraging prompts and workflows to create automated solutions tailored to their specific needs.

AI should enhance human work by automating boring tasks, not replace people. Accountants still need to review, verify, and make final decisions—AI acts as a 'third pair of eyes' to catch errors and improve accuracy.

An AI tool automatically extracts meeting notes from platforms like Plaud, organizes them, and posts summaries to a practice management system, saving time and ensuring team members have real-time access to client information.

By solving real client pain points—like inefficient note-taking or delayed access to information—AI demonstrates tangible value, builds trust, and strengthens client loyalty through meaningful, problem-solving interactions.

Chat with AI

Loading...

Pro features

Go deeper with this episode

Unlock creator-grade tools that turn any transcript into show notes and subtitle files.