Peter Geovanes: AI Costs, Prompt Engineering & Legal Pricing
Peter Geovanes, Founder of Juris Tech Advisors, talks to Tim Kenney at ILTACON 2026 about AI maturity in law firms, knowledge management, the token costs reshaping AI budgets, and why better prompts may matter more than bigger models.
“We were budgeting $15,000 a month and now we're well over $60,000.”
Key takeaways
• Firms arrive at very different levels of AI maturity: some need help choosing tools, finding use cases and training attorneys, while others are working through what AI means for pricing, engagement letters and client conversations.
• The common blockers he sees are too many tools at once (one firm had five), pilots that run for more than two years, and security and infosec not being in order.
• Adoption spreads through early champions, cross-pollination between practice groups, one-on-one and practice-specific training, and work on live matters. He won't take a client without direct access to the board or managing partner.
• AI could become a 'super KM tool', bringing together the best practices of partners who don't collaborate into a single playbook for the first time.
• Consumption pricing is the big risk: one firm he advises went from a $15,000 monthly AI budget to well over $60,000 in month one, with some attorneys running around $1,000 a day in tokens and no way to charge it back.
• Better prompting is part of the answer. His CRAFTED approach argues a well-structured prompt can beat a best-guess prompt on the most powerful model, and good prompts are model-agnostic and can be shared in a prompt library.
• He expects law firm AI spend to rise two to four times on consumption pricing, and says firms need a pricing conversation with clients: an AI-enhanced hourly rate, or token costs passed through at cost.
Explore more
Legal AI · Document & Knowledge Management · Law Firm Operations, Pricing & Billing · Innovation & Change Management · Legal Tech Go-to-Market & Growth · Security, Risk & Compliance · ILTACON 2026
Related conversations
Rob Martorana & Pamela Fehring (Theo Ai)
Tim Follett & Neal Chatrath (StructureFlow)
Full transcript
[00:14] People in Legal: Peter Geovanes, welcome back to People in Legal. Tim,
[00:18] Peter: nice to meet you. I did this a couple years ago. Happy to be here with you.
[00:21] People in Legal: Yeah, unfortunately, I didn't get to do that, but they were still talking about what a great interview it was. So I said, hey, give me a try.
[00:28] Peter: I appreciate that. Thanks for having me back.
[00:29] People in Legal: So you're the founder of Legal Tech Advisors, but you've got a long storied history.
[00:33] Peter: So Juris Tech Advisor.
[00:34] People in Legal: I'm sorry, Legal Tech. Juris Tech. Juris Tech Advisor. Long storied history in legal. So give us a little of your background for those who didn't see that. Yeah,
[00:42] Peter: real quickly, I started my career many years ago as a naval officer, did eight years of active duty went back to school, got my MBA, did 20 years in consulting, had a midlife crisis, went to law school. And for the last 10 years, I've been in law, typically at large US law firms. AmLaw 100 is kind of the term of art. But really, just by dumb luck, with my background as a JD, MBA, and a data scientist, I was one of the first kind of in this field. And then four years ago at this very conference, just by dumb luck again, I was seated next to two of the co-founders of a company called Casetext. And after ILTA ended, about maybe a month later, they called me and said, we've pivoted the company. There's this little startup that you haven't heard of called OpenAI. And we're doing some work with them. And they ended up building CoCounsel, which Thomson Reuters bought. And then I was a chief AI officer for another AmLaw 100 firm. And right now I've been kind of working as a consultant advising law firms around the globe on their AI journey.
[01:42] People in Legal: What's AI I've never heard? Just joking. It's all everybody's talking about. So you're at the right place at the right time. Tell us about your business.
[01:49] Peter: Business. So like I said, it's really been advising law firms and they're all coming in at different levels of maturity. Some of them need help in what tools should we select? What are the use cases? Help me train my attorneys. Others are looking more kind of strategically. What does this mean for the future of the firm? What does this mean for pricing? What does this mean for our engagement letters? Who's going to educate our partners so that we can have that adult conversation with our clients about how we're using AI in a responsible ethical way? It's the whole gamut.
[02:15] People in Legal: And it's got to be kind of the Wild West when you go into every different firm because every firm is different on what there's a lot of rogue usage of AI out there. Oh my God. There's a lot of structured usage. There's frustration because they want to use it, but maybe the data or the knowledge isn't compartmentalized in a way that compartmentalized in a way that they can actually use it.
[02:36] Peter: It's all the above and I think you've hit the nail on the head. And I have this, I'm going to say, a poor reputation is I'm honest to a fault. So usually I have to preface to the board, don't shoot the messenger. Here's what I see. I believe from what I've seen. You've got five different AI tools here. That's not the right approach, right? You've kind of got to figure out what our lane is going to be, put it all in. I had another law firm that had been doing a pilot for over two years. It's like it's kind of long for a pilot. I think the other firms are going to pass you up. And then another firm came to me, and the issue was a combination of not having their ducks in a row from a security standpoint, from an Infosec standpoint, right? So, to your point, it's data, it's Infosec. There's so much that's encompassing, and I would really say we're at this inflection point, or maybe even past it now.
[03:22] People in Legal: Yeah, and different, I'd have to think it's difficult for you because so many different firms are in different places, and even segments within the firm, right? So, you might walk into a firm and hey, you know, our litigation people, they've got it, they've got the data they need, they've got the security need, but you know, these other practice areas, they don't have anything.
[03:40] Peter: And I think that's one of the challenges with these tools. Extremely powerful. We're seeing advances every three to six months with the next level of model, which is great, greater capabilities, but there's really never been an AI for dummies book, right? Or what are the use cases, or how do I get started? Or where are these efficiency gains or competitive advantage that you promised me?
[03:58] People in Legal: Right, and I think so. The trough of disillusionment has to be coming soon, at least for part of these firms, right? And you've got, so then you get the shadow shadow IT gone where one of those guys now, I'm gonna do it myself. So, how do you control that? When you go into a firm, how do you harness all the different ideas and the problems that you run into? That's gonna be very difficult. Yeah,
[04:20] Peter: I think part of it is identifying those early champions. So, it could be an M&A attorney, it could be a young litigation associate, and allowing them basically the stage to share not only what their idea was, but why they did it. I'm also seeing a lot of cross-pollination. So, even though it was a litigation associate that's kind of doing the lunch and learning and showing what they did. Somebody on the other side on transactional side, you know, that's interesting. If I took that concept and applied it to my practice, I wonder if. So that aha moment happens all the time. But setting up that structure of kind of the initial, I call it the one-on-one training, and then getting very practice specific, and then also working with them on specific matters and being able to show with them what is the art of the possible, that's where I think the wheels start turning, and then sharing those and making that kind of operationalized across the firm. But it isn't easy, it's really difficult.
[05:07] People in Legal: And then you get a long road ahead. So let's say, oh, boom, wow, if we just captured this knowledge in a way that we could share it, oh, well, you know, we've got security issues, we've got data integration issues, we have the big one I want to talk about now, okay, let's say we start using AI. How many different agents do you have out there? What are the different things that you got going? And then we got to talk about the token consumption, right?
[05:32] Peter: We do have to talk about that, but I want to hit one other point that you kind of briefed over, which is really important. We call it KM or knowledge management. So I'm seeing all levels of maturity across firms. And one of the use cases that at least has me intrigued is, I was speaking with a partner at a UK firm. And he said, and we were talking about playbooks. And I kind of shared with him saying it's really important that we identify what those gold standard or exemplar documents or contracts that we want to use. And the playbooks we basically will distill the best information from that across the entire practice. And this partner said, well, I got to be honest with you. Our culture is we don't really collaborate. So there might be five partners in this M&A practice, but we don't talk to each other, we don't work on each other's matters. And a light bulb went off thinking, I wonder if this is kind of the super KM tool too, where I can, in aggregate, take Tim's best practices or Grace's best practices and bring them together in such a way for the first time and build a playbook that is representational of all their best ideas. That might be something we can approach.
[06:33] People in Legal: I think that's kind of the. The word I'm looking for is, I think that's the golden ticket everybody's looking for, right? So we really don't have KM, or we've got it in segments. How do we take all that tacit knowledge and make it explicit knowledge and operationalize it,
[06:51] Peter: perhaps, right?
[06:52] People in Legal: And something that's secure and consumable. That's not easy. So when you go in and see these roadblocks. How do you untangle it? Because it just seems like there's so many different directions to go in. How do you get consensus at these firms to move in a specific direction?
[07:14] Peter: I think consensus is a very optimistic word. I think what we're looking for and what I've seen is really a key factor to success. And I will not take a client unless I have direct communications with either the board and/or the managing partner. So that executive level commitment and real understanding of where we are and benchmarking them perhaps against their peer firms. Coming up with a plan, they're bought into it, and then the ability to execute gets much easier. It's always difficult to as a consultant coming in, and partners have different levels of tolerance or maybe even patience, and having the support of the managing partner really helps in that execution. Yeah,
[07:52] People in Legal: well, KM's come a long way. I did KM consulting around 2000, and it was largely an eye roll, right? Oh, okay, right, whatever. But now the...
[08:01] Peter: In the age of AI, it's so important.
[08:02] People in Legal: The value is seen, right? If only we could leverage that information. Oh, we don't have that information.
[08:07] Peter: We can't find it. That's the other issue. I have five million documents in my DMS. I can't find it.
[08:13] People in Legal: And ironically, it comes back to the same battle that search people have been doing for years and years and years. Garbage in, garbage out. I don't care what kind of filters you have in there. If you're searching garbage and you get garbage back, the same with AI.
[08:25] Peter: Right.
[08:25] People in Legal: But let's talk about the other thing. I want to really. Well,
[08:27] Peter: you teed me up on tokenization. And this is something near and dear to my heart as we're facing it right now. So one of the law firms that I've been dealing with was an early adopter of Claude for Legal. Unfortunately, their contract changed, which was basically a per seat license, which is relatively reasonable and gave you a pretty good number of tokens and now we're on consumption only. And a couple early signs have me absolutely terrified to the point that I've had to brief the board. We have seen some attorneys, as soon as the new model comes out, they think, I'm a snowflake, my practice is unique, I need the latest model, which is also the most expensive. We're seeing $1,000 a day in token costs. Wow. With right now, no way to charge that back to a client is basically some cost. Four to five X, our monthly, our previous monthly, is what we're seeing right now. And this is month one into this new experiment. So we are actively looking for opportunities to retrain those attorneys from perhaps Claude and get them into a tool that we have that is more license-based and includes unlimited usage because the economics are completely upside down.
[09:35] People in Legal: Right, and it's not the old days where you can charge back postage and things like that. Clients are not going to pay for your tokens.
[09:43] Peter: But you know, I wonder about that, and I know there's all kinds of different thoughts on from the ABA and stuff.
[09:49] People in Legal: They'll pay for value, right? Well,
[09:50] Peter: my question is: if I'm a client, would I rather spend five hours of attorney time at $1,500 an hour? or $1,000 in token costs, and now we're all whole.
[10:00] People in Legal: Right.
[10:01] Peter: So I don't know if we're there yet, but I think that conversation is going to have to happen.
[10:05] People in Legal: And it just has to be monitored so closely, right? What is the news? Well,
[10:09] Peter: the good news is, I mean, I don't know if a lot of these AI tools allow the attorney to basically say the matter, the project, what they're working on. So we would have a way to know token consumption on a matter or basis or attorney basis. But I do think something's going to have to be done.
[10:24] People in Legal: So what about teaching everybody better prompting, better prompt engineering? Has that ever cost your money?
[10:31] Peter: Wow, that's a novel idea. Just by coincidence, and I'm sure that's what it is, I did. I was an early adopter on these AI tools and I saw that was one of the areas that was really a challenge for attorneys and really everyone. We didn't go to school in an era that we were being taught how to talk to these tools. And I kept hearing the same comment from attorneys, like, you know what? We're going to pick on the IT team. The IT team rolled out product X, Y, or Z. We don't have adoption because nobody taught us how to talk to this thing. So after seeing this over and over again, the light bulb finally went on for me and it was almost dumb luck. I was using a technique called meta prompting, which is basically using the power of the large language model to help you craft a prompt. And I kept using that word craft and it finally dawned on me, oh my God, I started my career in the military, everything's an acronym. I think this works. And the word CRAFTED in my book that I wrote is an acronym. And what I do is I teach legal professionals all the different levers and lots of different examples of how to use a prompt. And there was a fascinating study about four weeks ago that said a well-structured prompt will get you better output than just using kind of your best guess of natural language and the most powerful model. And going back to our token consumption, I mean, I think there's a direct correlation there of better prompting, better outputs. I don't need Fable 5 or Opus 4.8. I can maybe do this with Sonnet at one, one hundredth of the cost. So that's where I think we're going.
[12:03] People in Legal: Like any other powerful tool, it has to be used efficiently, right? It just, yeah.
[12:08] Peter: And I think too, a well, I'm going to use my word again, a well-crafted prompt also allows you to save it into a prompt library and share it and start to again build this institutional IP and not have to reinvent the wheel each time.
[12:20] People in Legal: So as these models keep getting rolled out, and we know they are from everywhere, right? How do you keep up with the proper prompt for the proper model?
[12:28] Peter: Well that's a really great thing is the prompts are kind of model agnostic. So whether I'm using, I'm just going to name some of my favorite tools, whether it's a Harvey or Legora or CoCounsel or Claude or Lexis or whoever it may be, the prompts are universal. As long as we upload the information, and if it is, again, I'm going to use my acronym, CRAFTED properly. it will work from one system to another. And a lot of it too is at the back end, they're all using the same engines. Whether it's from OpenAI or from Google or Anthropic, they're all using those same large language models or foundational models to power these.
[13:02] People in Legal: That's fantastic. So looking forward, I don't know, I kind of think we're at an inflection point right now.
[13:08] Peter: Not another one.
[13:08] People in Legal: I don't know if I can make another one. Where are we going to be a year from now? So is the usage going to increase? Are we going to hit a trough of disillusionment? What's going to happen?
[13:22] Peter: So I think there's going to be, especially on the consumption-based model, which seems like everyone is moving toward. There is going to be a rise of 2x, 3x, 4x in legal law firm AI expenses. Wow. So at some point, there's gonna be that calling. Clients have now shifted from we're hesitant for you to use AI to say, I want you all in, I want you using AI on my matters. I do think we're gonna have to have that adult discussion and say, listen, I can either come up with a new billable hour, which is an AI-enhanced lawyer, so I'll use simple numbers. I bill that's $500 an hour, but if I'm using AI, it's $750. And that's the win-win for both of us. Or I'm going to track token consumption on certain matters and attorneys that are working on your matters and see if you're open to charging. I won't mark them up, but if my attorney spent $1,200 on this, you know that saved them three hours, four hours, five hours of work. But I do think that's what's going to happen. I think we've got to have that pricing discussion. And I think the ABA has got to jump into this as well.
[14:25] People in Legal: Very, okay. They certainly do this.
[14:27] Peter: They love this. Yeah,
[14:28] People in Legal: they certainly do. But anytime anything goes their committee, it's going to take a while. So firms are really going to have to figure this out for themselves, aren't they? Based on clients.
[14:36] Peter: Unless we build some type of consortium or start working together as an industry, I think that the pricing part is going to be an issue as we are getting deeper into consumption-based pricing. So cool. So like I said, we were budgeting $15,000 a month and now we're well over $60,000.
[14:52] People in Legal: And so Peter, you can sell to large law firms. If I'm not in a large law firm, your book will help small, medium-sized firms too, right?
[15:00] Peter: My sweet spot really has been over the last year and a half, two years, are those 20 to maybe 250. And actually it's been around the globe. I've got a couple Australian firms at Hong Kong, US, obviously, and UK. I've also been asked to do kind of bespoke training. So I wrote the book, but I also had a couple firms ask me to come out and teach their attorneys. And paralegals and legal assistants as well. Also, I also think we keep forgetting the business side of law. There are so many great use cases for AI, for the marketing, business development team, HR, talent management, right? And I love those because they have immediate ROI and it's really easy to see the impact.
[15:40] People in Legal: Great point. Great point. So firms can go out to Amazon today and get your book.
[15:45] Peter: It is on Amazon. It's CRAFTED. the CRAFTED approach to legal prompt engineering and it's uh it was i can say with certainty i have the screenshot i was number one on amazon i'm sure it was for three seconds but i had the screenshot and i'm just hoping at ILTA somebody will say number one bestselling author well but just number one bestselling author Geovanes for a sliver of time thank you so much thanks for coming back on Peter thank you i appreciate it wonderful talking with you best of luck you're in the right place right time thanks