011 This Is Your Practice Management System on AI

We got a fun one today. Over the weekend I put together kind of my ultimate roadmap guide for how to make our practice management systems better with AI. This is all stuff that's doable today. The tech is there, it's just a matter of building it. Let's talk about it, let's get excited. Things are only going to get better from here. This is I think, a great example of the it can't come soon enough type of application of AI. There's some, definitely some spooky stuff with AI and there are things to be nervous about, but this is one of those things that is just to me purely upside. And it is the various ways that really well validated AI tech that's already out there, that's secure, can make our practice management systems better. And when we're thinking about what the future of technology looks like specifically for accounting firms, everything kind of settles around the practice management system that's kind of the, the biggest player in the space. And when it comes to AI, much of making AI helpful comes down to who has the context to do that thing in a better way. So if you're looking at AI to automate more bookkeeping transactions, well, who has the most information about how to classify a bookkeeping transaction? Obviously you've got the banking information, but who's got access to the receipts, to the invoices? Those people will have more context with which to make a correct classification. And that question of context applies to. I didn't get that. Could you try again 2023 and Siri still does that. And that question of context applies to virtually every AI application. And this is another conversation, but we've since like with, with the cloud and all of that, we've kind of moved to more best of breed, integrate different apps as kind of the way we build quote unquote stacks of software. And I love that. But unfortunately I think AI is going to bring us more the opposite direction where all in one will probably become a little more in vogue because the more we have that context in a single place, the more meaningful things you can do with it. Now, specifically around practice management systems, I did a real meaty write up on LinkedIn about this that is targeted more towards product people. So if you're a developer, if you're on a product team in the space, probably go out and read the LinkedIn write up I did on this. This is going to be meant. Are you kidding me? Do you mind? This is going to be more targeted for accountants, less focused on the technical how to achieve These things and more on the practical, like, hey gang, let's get excited. But also, hey gang, let's go and share this with our practice management systems to like kind of show them the path towards leveraging AI. Now I've had the amazing opportunity the last few months to talk with tons of developers in our space, the people that lead product, the CEOs behind many of the practice management systems that we used. And I can say without a doubt, because I've had these conversations, there are people and companies who will come out ahead on this stuff and there are companies who won't. There are some with very mature understandings of how this works and how this will impact what we do and others that don't. But I think the scenario where we all lose is when Microsoft Copilot comes in and steals all their lunches, right? So Copilot's gonna be incredible. I've talked about it in past shows its ability to pull everything together to get into work on a Monday and say to your voice assistant, summarize to me the most urgent emails and create some to do tasks to handle those. Or I'm meeting with Jim at noon today, summarize all the most recent stuff we've done with him. And that's gonna pull across documents, across Microsoft Teams, chat across emails. Anyone in your team has, has had with that client. The idea of pulling all that stuff together into one summary. That is the value of having context all in one place. And if our space, the tech founders in our space don't lean into AI, we run the risk of ultimately AI development being, or I guess the people in our space being outpaced by Microsoft and Google basically because there's huge value in doing it right. And if we don't move from where we are to an AI enabled version of it, honestly we're all just gonna go to Microsoft 365. So part of this and part of my enthusiasm behind sharing this is to kind of like show this really exciting path that I think we should be investing in. Okay, so cruising through, I basically put my finger on 17 specific things that will fundamentally improve upon our practice management system. So first, a few applications for search. AI enables search in a really, really exciting way. And so imagine today the analogy I use is if you're a Mac user, Raycast or Albert or a very Albert Alfred or like a very basic version of Finder, the ability to pull up like a quick switcher, like maybe kind of like your Slack and in a single box be able to find anything. So imagine having that in your firm to search across organizational emails, across both internal and external team chat, across docs like down to what is inside the documents, across projects, across transcripts from meetings. I called it kind of one search to rule them all. That's really, really powerful. The other phenomenal thing about AI enabled search is it is not just text match search. So it's not like control F in a PDF. There's a different way of searching. It's called vector embedding search without getting too into the weeds that basically is searching for semantically similar concepts. So it's not just does the text match exactly, it's what are the highest scoring things that are most similar to to this. And it surfaces those things and there's so many things that are happening within an accounting firm. And maybe it's one thing if you're running your own firm, but even running your own firm if we're capturing really valuable sources of context like meeting transcripts, the ability to surface a meeting transcript from 6 months ago or 24 months ago when the client just mentioned that one little thing that's really, really valuable, the discoverability of information is about to get way better. Second example I shared is really building on that. I called it extended search. So for example, you search what W2s did John Smith have in 2021 or you know, an accounting. A similar accounting example is, you know, pull up the Columbia bank statement from March 2023 ending in this or for client XYZ that becomes honestly like pretty trivial for AI once all of that stuff is quote unquote embedded and then stored in a way where it's stored by the practice management system in a way where it's going to be discoverable. And so extended search is doing a little more around the classifications of specific documents. So your practice management system can know this is a W2 or can know that this is a bank statement not necessarily just based on the naming of the file, but based on what makes up that file. So right now with just vanilla GPT, you can try this in chat GPT. If you take the text of a document and you kind of use GPT to get a general idea of like what the classification of that document is like, what is it does a really good job especially if you just say, you know, the, like the, the way you roll out a extended search feature like this is probably just starting with the five or ten most common things that people search for. And so functionally for the developer, that looks like throwing the text of the file at GPT and saying is it any of these files. And you know, a W2 is an easy example because a W2 has kind of a prescribed set of text that you would expect to have in a W2. And so GPT does a really good job at classifying those documents. But the value to the end user is then when you're searching for it, to be able to pull up these specific documents or even pages within specific documents that answer your question. Because a better version of search is not here's the documents I think you want, it is here's an answer to the question and then here's the relevant info if you want to look at it. So using that W2 question, what W2s did John Smith have in 2021? It can go out in your file system, find all of the 2021 W2s and return the employer on those along with copies of them if you want to look at them. Or in the bank statement question, it can go out and fetch that bank statement along with any other relevant information that you may want to see when you ask a question like that. So search can go beyond just finding the thing to now actually doing a little more of answering the question. And then the third example of search is external search. So since context ultimately wins the day when it comes to AI, whoever has the most context will be able to do the most meaningful things. If you have an app that authenticates into your client's QuickBooks files, for example. So more practice management systems are doing this to support with uncategorized transactions, you've got this connection out to the client's QuickBooks file. Well, imagine having that one search to rule them all. That went down to the client ledger level. The idea of being able to search across all of your client QuickBooks files from one place or do like roll up summaries of information all in one place. Many of these practice management systems already have are already authenticated into those QuickBooks files. And so pulling more of that information in to be discoverable, embedded to be able to be searched along with all of your other internal documentation. Really exciting. I'm turning my watch off completely now. I don't know what's happening. Okay, a few other like non obvious sources of context, tax returns, being able to answer questions based on amounts that are in specific boxes of a tax return. GPT generally can't do this just looking at, you know, a Form 1040 for example. But by layering in a bit more context about what those boxes mean, that search functionality can be super extended to actually answer helpful questions about Clients. That also gets really powerful when you start talking about how you reply to client emails. Meeting transcripts. I think most of us probably still don't capture our meeting transcripts, but the upside of doing so now looks completely different when the information in those transcripts is much more discoverable. Engagement letters, like just being able to pull that stuff into search. When you're thinking about how you communicate with clients, having that context alongside replying to an email about scope, that becomes really helpful if it can surface that information for you. But, but then even I had the firm's accounting file where you can see payment and invoicing history for a client. That could be helpful context even as you're talking with clients. Okay, next section, client communication. Fourth thing I had here is generative replies. Arguably the holy grail of AI applications is a really intelligent suggested email reply engine. And so that's going to do a few different things I listed out, like things that it ought to know when it's suggesting a reply to that email. So that suggested reply ought to take into account that a colleague emailed the client yesterday that you're generally snarky with this client. I can see that from your email history, that they have a project that's near completion, that their engagement letter stipulated certain expectations of the client, the details they provided in a zoom meeting last week. All of those things can be taken into account when the practice management system suggests a response or suggests a few different responses that you can choose from. And this can be honestly really, really good because of everything that it knows. And as much as it pains us to say this, it's probably going to know more than you know if you run a team. Because a really hard aspect of running a team and supporting clients across multiple firm users is just that you don't have total transparency into all those conversations. I mean, how many times have you talked about going round and round if you do both accounting and tax on, okay, do they have a tax point of contact? Do they have an accounting point of contact, do they have a like customer relations point of contact? And does that consolidate those two into one or does that really just create a third type of contact, pulling all those conversations into a single source of context so that when you go to, you know, request something over email or reply and you have that other context, it's just gonna make you look way smarter. Now, important here is arguably the biggest sort of ripple effect of AI in the accounting software category will be the importance of how AI enables email communications. It's like arguably the most exciting improvement we're gonna get in the next 24 months is just how good those suggested replies can be. And I think it fundamentally changes what we're going to demand of our practice management systems. In the past, specifically, are you going to be sending emails out of your practice management system or not? There have been email first practice management systems where that is kind of like a help desk approach, where that's a really core part of their offering. There have been practice management systems where the email was secondary and really just pulled in kind of to be used as a CRM. Like, not that you're necessarily going to author a lot of emails out of the system, but it will connect to your email so that you can see your recent emails to and from that client, but still not really to like author emails out of. And then there's a whole other type of practice management system that steered clear of email altogether. Maybe they did messaging as an alternative or they just weren't going to get into the client communications part of it. And I think ultimately, like, I have a really hard time seeing any future alternative. Email has to live where the context lives. If you're going to write your emails in Outlook, but all of your context lives in your practice management system, you're not reaping any of the benefit of having that context around how you reply to emails. And so ultimately, I think in that situation, you're either going to keep emailing in Outlook and pull all of your contacts into Microsoft 365, or you're going to go out and find a practice management system that will do both well, that will do email and context well. I said in this write up, I said that the clock is ticking on contextless email because to me, the upside of suggested responses and suggested file attachments and all the other ways that AI can simplify managing client email are so powerful that I just don't think it's realistic going forward that you're going to have a separate system for email and a separate system for everything else. I think they have to live in the same place. So for practice management systems, honestly, I don't know a way around wading into email, whether you like it or not. Okay, another really helpful thing here, suggested documents. I've shared this example before. Somebody asking for a copy of an old tax return or asking for a copy of an old set of financial statements that you've already sent them, that's honestly really easy for AI to knock out. If you have those files in your file system, you get that email. It can create a Little boilerplate reply and suggest the relevant attachments to answer their question. Like, it can be as simple as that. That's not a big lift right now. Last on, like email communications, I think there's also an opportunity here for it to create on the fly generative attachments. So for example, you need a letter for a banker. A client emails you asking for something like that. Our organizational history may have 200 other such letters. Give me the option to select, you know, three that it surfaces as maybe having a kind of a similar purpose to what they're looking for. I check a few of those and it generates a starting point for me that draws on those past letters that we've already sent. That's not a big lift. Like GPT4 would actually do that really well. The value is it's going out and finding the most similar letters you've written in the past. And in the process of building a reply to that email, it's gonna give you a jumpstart on writing that letter for this client. How cool is that? You're not having to recreate the wheel from scratch every single time. All right, let's talk about files. This is a simple one, but really helpful. Renaming. A couple ways AI could help us in file renaming. First, it could simply go out and find the most similar documents and then give it a name. Kind of along those lines. So if this thing looks like a bunch of other W2s, it's probably a W2. Look at how those files were named, assuming they were named well, and give it a similar name, but one that's specific to this client in this document, or name it according to like an actual file naming convention. So this is something the practice management system could suggest for you, or it could be something of your own design. If there's like a rules system for how stuff gets named, AI can absolutely be helpful. There would probably be better at it than like human users who may sometimes do that and may sometimes not. More on documents summarization, give me a one or two line summary of what the document contains. Save me from having to open and close it. That's an easy one right now. Document similarity. Surface similar documents to me, that's gonna be helpful if I'm authoring a new document or even modifying a document. The banker letter is a great example, but it could also be like responding to a certain type of IRS notice. If I've got other letters in my system that are responses to a certain type of notice. Sweet. I'd love to see those other documents. If you can Surface those for me. That's gonna make my life easier. Now let's talk about requests. I love requests. It is one of the best aspects of practice management systems. Keeps humans from having to track down clients just to get pieces of paper, which is waste. Now, where can AI help here? Do we already have it? So nothing will make you look worse than asking for something the client already provided, right? Or they gave it to a different staff member or provided it in some other channel where maybe you weren't aware of it. Man, like AI is so suited to help with this because it can see that context anywhere, no matter how it came in, whether it was an email to your colleague, whether it came up as like a just kind of this sidebar in a zoom conversation, when you build that request, it can say, hey, like here's the most similar things you are, most similar bits of information you already have before you push out that request and look like a dummy, right? Another example is a request still valid. So on kind of the flip side, whenever you have new context from a client, maybe you just wrapped up a zoom meeting, maybe they mailed you something, apply that context to the existing requests to see if those requests are still valid so it stops sending auto reminders. And in a perfect world, when those tasks are completed, your project status automatically kicks back to some sort of active status so that the person that's working on it knows that it can be worked on. Now, now that could be something as obvious as maybe they uploaded a document and didn't label it in a helpful way. But it could be something non obvious, like maybe a client makes a statement in a zoom call today, and three months down the road, one of your colleagues is gonna put out a request for that exact same information that they just mentioned. It could keep them from doing that because now you have that context contained within the zoom call. Now a couple client side opportunities here when they are working within a client portal. So oftentimes clients will submit batch files that are not particularly helpful Here. AI can help us both with classifying what those files are, but also splitting those files in helpful ways. So I don't want like, generally I'm not gonna want 70 pages of stuff all lumped into a single document. GPT actually does this really well, finding the starts and ends of new documents and then like we talked about, classifying them in a helpful way, like renaming them. So like if they chuck in 20 pages of documents in less than 10 seconds, even less than 5 seconds, generally you're going to be able to get GPT to give you the logic, you need to suggest a split of those files, probably to the client. Maybe you want the client to confirm that that looks right. Similarly, when they make those uploads, in a perfect world, the practice management system then says like, hey, does this resolve this or that request? And the client can confirm that. So it doesn't take any guesswork on the other side. Are you catching my drift here? That there's a lot of stuff that can be done that hasn't been done yet. There's a lot of stuff that can be done. CRM stuff. Sentiment analysis. How grouchy are they or how happy are they? Historically, there haven't really been great ways to do this because the ways in which you communicate with clients are spread across so many places. But now you have all that context in one place, all of your staff communications, all of your meaning transcripts. This sentiment analysis can now actually be really useful. You can benchmark how like the sentiment of specific clients and their communications with all staff people, but also specific staff to client communications. And so some clients will just sound grouchy all the time. That's normal. But a really helpful thing here is if that client's really grouchy every time they talk with Jim, but not when they talk with Sally, maybe that's something that I want to know. Like maybe that's a training opportunity. Maybe that's something that we need to be aware of. Another just like cool thing to make you look smart is external updates. So fetching information about a client from their website, from their LinkedIn. You're gonna look like a real smarty pants when you go to reply to an email. And that suggested reply pulls in some announcement or some reference they made online a couple of days ago with like a link to that thing that they talked about. Like that's just gonna make you look just really smart in the eyes of the client. Okay, last agents, this is the thing that I am most excited about. And this is also like the long term way for practice management systems to differentiate themselves beyond what say Microsoft or Google would do. And that is agents that will actually do the work. And the way we've seen this in the past is more like bots that will do specific tasks. But this whole category is changing really, really quickly. If you're interested, check out, it's called Auto GPT and it's kind of the best point of reference we have today for what these agents could look like. But basically it's a little bot that just prompts the heck out of itself until it Gets to something useful. In fact, I can pull up something that I just was in my Twitter feed today. So not related to what we do, but just an example of how AI agents work. So this guy, he says, I pretended to be a fake shoe company and gave Auto GPT a simple objective. Do market research for waterproof shoes. Get the top five competitors and give me a report of their pros and cons. So by itself, it goes out to Google, does a whole bunch of research and starts creating questions for itself to refine the result. It asks itself things like, what are the pros and cons of each shoe? What are the pros and cons of each top five waterproof shoe, Top five waterproof shoes for men. It's doing all these different searches, but then it's like self prompting is what they call it. It's basically the AI having a conversation with itself to distill the result down to something more helpful. He talks about examples where it went out and looked at the reviews for these shoes. It acknowledged the fact that many of these reviews might be fake and so it would go out and then validate the reviewer themselves to see if they looked legitimate. It even spawned, this is a cool thing about sort of self prompting stuff like this. It spawned an additional agent to carry out the task of analyzing the websites where he was looking at these things to like validate the legitimacy of the websites. All of this by itself. So the whole idea of these kind of automated agents is they will go out and kind of managing, manage these tasks themselves. And if you think about that in the context of a practice management system, where in the past we're used to bots that could do like predefined sets of tasks, this looks a little more like simply assigning a task from a human user to your AI agent and saying, go wild. Now that sounds pretty wild and pretty far fetched. It's actually closer than I, than I like, it's closer than it seems. In the past two weeks, the biggest repositories on GitHub, which are like the main kind of open source software platform, all three of them have been automated agents like this one by Microsoft and two others that are open source projects. And so progress on this stuff is happening really fast. So the place to start with that is probably a more like administrative agent, something that can go through and like manage file naming and do some of those more helpful sort of internal things. And anytime it's going to go take an action, generally it's going to report back to some sort of supervisor. Or maybe the person that assigned this task to them says I'm going to go do this, does that sound good? And then you say yeah, whatever, that sounds fine. But where this gets really cool is domain specific agents. So let's go back to the practice management system that's connected to all of our clients QuickBooks files. Now we're looking at getting an agent to go out and perform actual tasks in those files. And the way that we're doing this is through your task management system and assigning tasks to agents that could be reassigning them from a human user to the AI. That that could be task templates with recurring tasks that automatically go to that agent. Like I get how wild and ridiculous this sounds. We're closer to this than you think. And agents I think are the really, really cool long term opportunity for practice management systems to not only give us a platform to make our human users as productive as possible, but to then like augment our teams with these like quasi users that are actually AI agents that will go out and will actually do the work for us. How cool is that? That's really exciting. That was like Machine gun, whole bunch of featureless things. Like I get that, but that is just like how rich of a landscape this is right now. Just like Full of opportunity for doing cool exciting new AI stuff, especially contextual stuff. That is a problem that I have that maybe Microsoft Copilot may not solve for. So who are the winners going to be in this space? Ultimately I think like in much of the way that really thought leadership now is coming into a new AI paradigm in the same way that you had different thought leaders in the cloud accounting paradigm to the, you know, kind of the OG accounting paradigm. I think we're seeing a, a sort of new era of thought leadership emerge around AI in terms of products. The people who win and the people will lose. They're gonna look different in the age of AI as well. Now the interesting thing to me is with the practice management systems kind of being the home of context going forward, how does that reshape everything else around it? Does that eliminate some space for other things around it? Does that just create new opportunities for AI enabled stuff around it that didn't exist before? Stuff like agents, like having a plug and play agent that will integrate with my PM. This may be phenomenal at doing construction QuickBooks accounting or something like that. Right. That's exciting. So that's a rundown of everything I'm excited about as it pertains to practice management systems. I'm going to link the LinkedIn write up that I did. That's more technical. Please share that stuff with your practice management peeps. Like if we can help them see the path to something more exciting, that's only going to help us more ultimately, right? That's enough excitement for me today. We'll be back tomorrow. Probably a little less AI Sweaty. Thanks for coming by.