424 5 Ways To Use Your Meeting Transcripts

AI meeting recorders. They're. They're all the rage right now. Accounting firms are picking them up in record numbers today. We're talking about five ways to use them, five things to do with them, because it's easy to buy into the value of them. But oftentimes afterwards, people are like, where do I put them? What do I do with them? Now that's what we're talking about today. Come on in. Jason's AI meeting recorder just joined the meeting. Can we. Can we first acknowledge the practical annoyance that is other people's meeting recorders? The fact that they don't join at the exact same time as their human companion? I find it quite irritating because you will get there and you'll be sitting in a room full of meeting assistants by yourself. If you want, you can talk with them, you can tell them things. It'll show up in the transcript. In fact, who was it? I think it was Chad Davis that, like, had Chat GPT write a bunch of accounting jokes or something like that, and he just started rattling them off to the meeting recorder until the person actually joined the. Also, if we can further get off the rails, hey Gen tool called hey Gen, if you've heard of that, hey gen.com h e y g E N Their party trick is they will make visual AI avatars that look super, super realistic. They're very good. Over 12 months ago now, I made one of myself and my wife couldn't tell the difference. Justine Bateman got super mad, retweeted some stuff that I'd said about it, and I got a whole bunch of Hollywood folks shouting at me, but, hey, Jen will now let you create a visual avatar of you or anything else, as long as it doesn't think it's a celebrity and you can send to go join a meeting, a meet, like join a zoom meeting. You can even give it, like, its own knowledge base. They launched a streaming version of it. So you can put an avatar of yourself on a website and folks can, like, vocally interact with it and it will pull from the knowledge base. So similar to how we have Jason Bot here at the podcast that is trained on over 400 episodes of the podcast. All the content and I can ask it questions and it gives me very realistic, true to what I would say responses, because, crying out loud, it's got 420 episodes of a podcast in his brain. The same way that all of that stuff is backing that chat, like text assistant, you put that stuff behind a visual avatar. So imagine sending that thing to a meeting. Pretty weird. But imagine then when they hop into a meeting and another AI avatar is already there. Who talks first? How do they start? Nice weather we're having. What do they, what do they say? Where did that conversation go? Weird times, strange, strange times. We are entering, my friends. Okay. I had the pleasure of doing an AI accounting conference in Melbourne and Auckland last week. Super, super cool, a lot of fun. And it was a bunch of just firm owners rocking up, talking about what they're doing and what they're not doing. And the most common through line, there are two of them. It was ChatGPT and it was meeting transcript tools, meeting recorders. Just real short on the value of meeting recorders. If, if maybe you're, you're new to the podcast here, it, it is just so valuable to have perfect recall. Like, that's what these tools enable. You end up with a transcript. And the transcripts are really freaking good now, like, very accurate. Imagine if you had had a transcript for every meeting, client meeting you've had in the last 10 years, how valuable that would be. How many times it comes up where it's like, oh, what about, did they say this thing or that thing? And you could just go back and search. Another thing that's happening that's really valuable is it's not just like dumb control F. Find an exact word search. You can use large language models to search more contextually so that the words don't have to be exactly, exactly. You know, the text string doesn't have to match perfectly. You can ask questions more conceptually about, say, their daughter and whether they were going to claim their daughter on the odd years or the even years or something like that. Maybe in the transcript it's the daughter's name that's referenced, but you don't remember the daughter's name. Situations like that, like, it will figure that out for you. And so it makes that information much more discoverable. And the hard part right now, which is why we're having this conversation, is admittedly what to do with those transcripts after you have them in the most useful ways to make them available for the rest of your team and all that. Why I put such urgency on this is the ways that we can work with our transcripts will continue to get better and more powerful. But those meetings, once they happen, you can never get them back. And that's why I've really been pushing people since last year. Capture that stuff, because it's gold. It's so important. I remember. Oh, man, we were in. Where were we? We're in Pasadena this was August who 19 months ago. 18 months. I don't know, long time ago. And we did like a whole dedicated episode to like let me just convince you to start rolling meeting transcript tools. And honestly my favorite recommendation, the tool that I recommended back then is the same one right now, Fireflies. It's as good as some other third party ones. I would still say it's a little bit better. Others have gotten closer. I personally like third party tools like Fireflies AI but because they will work across every different type of meeting platform. As opposed to Microsoft Teams Premium only works in teams. Zooms only works in Zoom. With a third party tool like Fireflies it can auto join any like virtual conferencing event. Like if it sees a link to a virtual conferencing platform, it can automatically join anyone that it sees on my calendar. I can set it to only do that on ones where I'm the host or where a colleague of mine is the host. But then it's not specific to a platform. Also they have a great mobile app so that in person meetings, all I gotta do is put my phone out just like an old timey tape recorder like you would use in a college class or something like that and it records that. So that regardless of where I'm having these meetings and conversations, I could pipe them all back into a single place. And if you really want the JSON on firm's gold star, you'll use a VoIP system, a phone system, voiceover IP that integrates with your meeting recorder. Then you have literally transcripts of all that stuff. So biggest blocker for people. I don't like my conversations being recorded. My stance on that, if my client doesn't want what they're telling me to be recorded, I don't have that conversation to begin with. I, I don't like sometimes when you're the firm owner, you want to pick and choose and be like, I want this to be an off the record conversation. I want this to be an on the record conversation. You won't extend that same thing to the rest of your team. So I want the rule in my firm to be you don't want to have that conversation, then I don't want to have that conversation. It's just a better way to, to approach things. Those transcripts can actually be a great cya. I am personally not interested in off the record conversations. It's not going to go there. All right, when you've got the transcripts, five ways to use them and I'll spoil the punchline here, not Great, none are perfect. But these are the best options that we have right now. Just know this is the worst it's ever going to be in terms of interoperability, like the tools that we use of being able to do useful things with these transcripts. So number one, it sucks, but it's better than nothing. Straight up control F searching bub. Like having the text of the meeting is still better than nothing. Like being able to search that stuff in old timey ways. Actually I did a main channel YouTube video. We'll link that down in the show notes that was all about meeting recorders. This was maybe a month or two ago. And for example, if you use zoom and you only want to enable the zoom meeting recordings, there's a way to just get the text files for all those transcripts. Even that is better than nothing. Even if you just got to do the control F search to find exact words, that's a whole lot better than having nothing. But the way I use it the most number two is natively in the recording tool because Fireflies, the tools that I tool I use, has a built in large language model so that you can basically chat with the transcript. And that's where talking about things more conceptually becomes possible. Where you don't have to get the words exact, right? You can ask it, hey, at any point in time did rental properties come up in here? Or at any point in time do they talk about the April month end close and it'll tell you and it'll call out the points in the transcript and it'll give you an answer and maybe that answer is enough, but it'll also point to you exactly to where the act transcript is so you can read the real thing for yourself if you want. That's the value that one of the most valuable things large language models are doing for us now is just giving us a better version of search, like being able to pull that stuff from big documents instantly in a way that honestly in the last two years has gotten much, much more accurate. In the beginning, like early days large language models, you could do this stuff. It just wasn't very good. It just wasn't very accurate. It was actually really expensive. I built a tax research platform In January of 2023, three months after ChatGPT came out that was doing retrieval augmented generation, when that was really new, really novel early days lang chain, if that means anything to you. And my thinking was like, holy cow, these language models are unbelievably good at pulling answers from documentation. What a fundamentally better way to do Tax research. And on the original model that ChatGPT ran on GPT3.5, we could do it and it would generate these super convincing answers, but they were just almost always wrong. It was like, I can't, like I've got a reputation. I could not put out a product like this where it will, no, in a novel way, very quickly pull from all these sources and then give you an answer that's still wrong. And so we kind of shelved it. Then GPT4 came out, the next model that was, was it February or March? A big improvement, Very big step up. Almost always right. Maybe not quite to the level that I would like, but it cost a, in that configuration, $2 and 20 cents per query. $2. That is over two Freedom Bucks for a single query. And I'm like, by, I don't know how to make any, any sort of business model around this. And so we kind of scrapped it at that point. And since then we've got, now we've got like other dedicated research tools like your tax GPT, your, your Blue Jays doing some cool stuff, but language models now, like, even if you're just rolling chat GPT chuck a document in there, it does this stuff really, really well. Asking it questions about the documents looking for. This episode is sponsored in part by Cloud Accountant Staffing, who helps you find super talented accountants in the Philippines. Boy, is it, is 2025 going to be the year of the Philippines. Like, it's not even just an accounting thing. I have talked into like so many people in other industries lately where they're like, yeah, no, we've started picking up people in the Philippines. Just the whole trend of like folks in the Philippines working for Western companies that like, has become so normalized now. I was talking with a listener of the podcast the other day, completely unprompted. They're like, yeah, no, I, this was a good idea. I changed this and that and I kind of finally got off my butt and started hiring people in the Philippines and I hired like three people with Cloud Accountant staffing just last month. Holy smoke. First of all, chill out, guy. Second of all, what took you so long? Third of all, what's taking you so long? Okay, it's the same thing I did. I made one hire and I'm like, okay, this is actually working really well. I'm going to make some more hires. This is not me just shilling wares for people that do placements. Actually, I guess that's technically exactly what this is, but it's also exactly what I did. It worked really well. A lot of listeners who have started doing this, I've heard super, super positive things from they're still human beings, it's still managing people. So it still sucks in many ways. But cloud account staffing is going to make it as easy as possible for you. You can explore that. Check out the link to learn more about Cast down in the show. Okay gang, this podcast is sponsored in part by Team up, the folks that help you find super talented accountants in the Philippines and hire them directly. Teamup just launched a. What do they call it? A salary guide. A salary guide for hiring accounts in the Philippines. But it's so much more than a salary guide. You know all the embarrassing, ignorant questions we're afraid to ask about hiring people in the Philippines? How much does it cost to get this type of. Can I even get, I don't know, a tax pro, an accounting manager, an executive assistant? Is that a thing? This guide and it's going to be linked down in the show notes. It'll show you not just how to like get each of those different types of people, but what they'll cost. How to build out a job description all of the questions that you have been too embarrassed to ask to find folks at Team up, they put them into a guide that you can swipe get that one down at the link in the show notes. So number three, third place you can use this is keeping them in your chat GPT Claude, another really good AI tool. If you use that, pretty much everything I'm going to say here applies to Claude Also. Now I used to recommend people use custom GPTs. I no longer recommend that. There was a time when they were the best way to talk about about specific documentation. That's no longer the case because custom GPTs don't use the latest reasoning models from OpenAI. So reasoning models, they do what they sound like. They think about an answer before they give you a response. So they'll take, you know, 10 to 30 seconds to think about that answer. Turns out that vastly reduces the likelihood of hallucinations. Then it gives you the answer. When I'm doing research, if I gotta wait 30 seconds for a non hallucination answer like I'm willing to do that, right? Unfortunately on custom GPTs right now you can't use the best model for this which is unfortunately named O3 mini hi H I G H. You can only use the 4O model which has over double the likelihood of hallucinating. Very good model. Awesome for creative stuff and 12 months ago was mind blowingly good or nine months ago when it came out, but now there's a better one and you can't use it in custom GPTs. So if I got a meeting transcript, one way that I can organize them inside of chat GPT is, is within what it calls projects. So inside of cus, inside of chat GPT you have this concept called projects, where you can upload any number of documents to a project and then any conversations you have, like your conversations nest under projects. So then when I have a conversation inside of a project, it will reference just the documents I gave it in that project. Interestingly, you can also put custom instructions inside a project. So I want to have a pirate project. I can put a custom instruction that says always talk like a pirate or talk a big Latin or give me concise answers, or in the case of research, search quote the exact excerpt that answers my question word for word from the source documentation. For example, if I don't want the AI making an answer, and if I just want it to tell me what the source docs say, I could build that in as a custom instruction inside of a project. Now with projects you get, you get basically those two things. Custom instructions and being able to upload documents to projects. But you can also just put conversations into projects. So in many ways they become kind of like folders inside of your chat GPT. And from the beginning of time, ChatGPT has just been this like diarrhea of conversations on the left hand panel, right? Got a little better when you could search them, which is really nice, being able to find old conversations. Now, projects are almost like a folder structure to help you organize those conversations. If I wanted to pull my meeting transcripts in there, I could have a project, you know, one project per client, and pull the transcripts in. The rub is that's a lot of projects if you have very many clients. And if you're on the team plan, which if you're an accountant and using ChatGPT, you should be on the team plan for the data protections it offers. If you're on the team plan, you can't share those projects with team members. Super annoying. Super, super annoying. That's actually maybe one argument for custom GPTs, because you can share custom GPTs with your colleagues, so you can create a custom GPT for every single one of your clients. And those are just private to your organization. Share all that with with your colleagues. Problem is going to use the older model, so there is more of a risk of hallucination. It's still, it's exceedingly rare at this point. Hallucinations are not what they used to be. I told you I wasn't going to have any perfect answers for you here. But that is one thing you can do if you don't have very many clients or if you only have a few clients you want to reference the transcripts for, you could always pull them into a project in ChatGPT. Fourth way you can use it is just in an ephemeral chatgpt conversation. So this is just starting a new chat with ChatGPT, dragging the transcript in, asking it a question. My recommendation for situations like this is you click the little toggle to make this a temporary chat. And when it's a temporary chat, as soon as you leave it, it self destructs, it goes away. And this ought to be how more things in our tech lives work is once we do them, they just go away. Particularly accountants can be hoarders and we hang on to anything and everything to kind of COVID our own hineys. But the reality is the more data that's out there, the more stuff there is to potentially leak. ChatGPT very secure ways to use it. But regardless of any platform that I use, I don't want that platform to have any more data than it needs to have. And with ChatGPT you can enable temporary chats and that can actually be the default so that anytime you start a new chat by default, it will self destruct. Self destruct at the end. I highly recommend enabling that. It's, I mean it's to me, it's not unlike browser tabs where I don't want every browser tab I open to exist until the end of time. I use the arc browser arc. That's one nice thing that does is basically every 24 hours, it's 24. It kills a tab unless you've been back to it. So I don't have this like ever growing list of tabs to manage. I kind of want my chat GPT conversations to be the same way. So fourth way you can use it is just when you're looking for a very specific thing, grab a transcript or grab a stack of transcripts, pull them all into a conversation, ask the question, and then it goes away after that last one. Fifth suggestion, put them inside of a Microsoft or Google folder, whatever your kind of platform of choice is that can be chatted with. This is kind of an interesting one. I never really used Workspace or Microsoft stuff for file storage. I always had a different app for it. And so this is maybe most relevant to folks who do use it for like long term storage, but increasingly inside of like SharePoint, OneDrive and Google Workspace and all that you're seeing chat with this or that more commonly. But I believe now in Google Workspace at the folder level you can chat with the contents of the folder or look for specific things. So if you already manage your file storage here, like in, you know, in OneDrive, something like that, that may impact your folder structure a bit. Maybe like under a perm folder you have a meeting transcripts folder and all the meeting transcripts go in that one place. And at the folder level you can then do chats that see into all that different stuff. That's maybe the best way to do it right now to me, not a reason to like move those files to another system necessarily. Also, those chat tools are not as good and not as accurate at chat as chat GPT. Unfortunately the bummer here is and what what I want the solution to be and what it will probably be long term is practice management system search getting much better. Because your practice management system these days it probably ought to be your fil file system. Unless you're doing assurance work, unless you do an audit, a test. In most cases the right answer for small firms is for your practice management system to also be your file management system. Unfortunately, search and PMS right now is not very good. And this is something I've been pushing clear Back to my January 1, 2023 AI predictions. One of the big things was search across everything how to get much, much better. And right now this is one area where PMs are super lacking. I try to, I try to put the screws to software companies in the accounting vertical because everything else moving really fast and you don't want that stuff to pass them by. And I want the companies in the accounting vertical to win because their ceiling is higher in terms of how they can support us. Look at a tool like keeper where they're solving like Super Accountant specific workflows tool like Sorban that's doing tax intake really really well. We want tools inside of our vertical. If the tools we're using to prepare companies books and prepare tax returns are the same as any tools that consumers can use, that's a bad state for the industry. We do not want that. So we need to be cheerleading and honestly supporting and with our buying decisions supporting the companies inside of our vertical but also like pushing them to continue getting better and taking advantage of the opportunities that are out there. The same technology that I was using to do tax research in this MVP app In January of 2023, that same technology, retrieval, augmented generation. That's how you do search really well right now. And it's not implemented in a meaningful way in any of the PMs, despite being a thing that's been very accessible for over two years now. So that ought to be the answer long term is better search inside of your pm. We're not there yet, but basically the five ways I answer this question of what do I even do with them, these transcripts, once I have them. Number one, even if it's just a text file, that's better than nothing control F you can still find stuff in them. Number two, if you have a tool like Fireflies that will let you chat with the transcript, that's really valuable. That's what I do. Number three, pop them into chat GPT into a project. Number four, just throw them in chat GPT whenever you need them. Kind of a pain though, because then you got to go out and fetch them every time you're trying to find something. Or number five, put them in a OneDrive, a SharePoint or a Google Workspace folder that you can then chat with. Not a perfect answer, I know. But the biggest thing today is that we're capturing them because the ways that we can interact with them will continue to get better. And you will never regret having those transcripts, anything I missed here, or cool stuff you're doing with them. There's a lot of like other powerful ideas like content creation and ideation gets way easier when you're capturing your advising. Like these advising meetings are literally like the sawdust of a professional services firm. Like there's so much value baked into these meetings that could probably be harvested for other things. It's another thing that language models are really good at is you could take a couple advisory meetings and be like, hey, outline a couple blog posts or podcast episodes or something like that from these real life advisory meetings that I'm having in the course of my business and getting paid for those, those things are super, super valuable. And if, if all we ever do is we have those one on one meetings with clients and we give them a tremendous amount of value and that stuff never otherwise sees the light of day. You don't have those transcripts for, for your juniors to be able to query and be like, hey, has this conversation ever come up with a client? Like so many ways that we can use transcripts. Unfortunately right now, like the tech and where we put them, that's the blocker. But having them, you will never regret having them because there's so many different ways to use them. That's it for today, gang. Thanks for being here. I'll see you in the next one.