Your lawyers don't need a bigger model. They need a better prompt.
Peter Geovanes, Founder of Juris Tech Advisors, in conversation with Tim Kenney at ILTACON 2026
One law firm Peter Geovanes advises budgeted $15,000 a month for AI. One month into a new consumption-based contract, it is spending well over $60,000. Back on People in Legal for a second time, Peter told Tim Kenney why that number has him briefing boards, why the fix may be teaching lawyers to write better prompts rather than buying bigger models, and why the pricing conversation with clients can't wait.
Naval officer, consultant, lawyer, AI officer
Peter's path into legal AI is not the usual one. He spent eight years on active duty as a naval officer, went back to school for an MBA, spent 20 years in consulting, then, in his words, "had a midlife crisis, went to law school." For the past decade he has worked in law, mostly at AmLaw 100 firms, where being a JD, MBA and data scientist put him in the field early.
Four years ago at ILTACON he was seated next to two co-founders of Casetext. A month later they called to say they had pivoted and were working with "this little startup that you haven't heard of called OpenAI." That work became CoCounsel, which Thomson Reuters went on to acquire. Peter later served as chief AI officer at another AmLaw 100 firm and now advises law firms around the world through Juris Tech Advisors.
Every firm is at a different stage
Firms come to him at very different levels of maturity. Some need help choosing tools, finding use cases and training attorneys. Others are asking strategic questions: what AI means for the future of the firm, for pricing and engagement letters, and who will educate partners so they can have "that adult conversation" with clients about using AI responsibly.
By his own description he is "honest to a fault," and often opens board briefings with "don't shoot the messenger." The patterns he sees:
Too many tools. One firm had five different AI tools. "That's not the right approach." Pick a lane and commit.
Pilots that never end. One firm had been piloting for more than two years. "I think the other firms are going to pass you up."
Foundations not in place. Firms that don't have their "ducks in a row" on security and infosec stall before the tools get a fair go.
Part of the problem, he says, is that models leap forward every three to six months, "but there's really never been an AI for dummies book."
Champions, cross-pollination and buy-in from the top
To move past shadow AI and scattered experiments, Peter starts by finding early champions, such as an M&A attorney or a young litigation associate, and giving them a stage to show what they built and why. The ideas often jump practice groups: a transactional lawyer sits in on a litigation lunch-and-learn and sees how it applies to their own work.
From there it's one-on-one training, then practice-specific sessions, then working alongside lawyers on live matters to show "the art of the possible," and finally operationalising what works across the firm. "It isn't easy, it's really difficult."
On getting consensus, he is blunt: "I think consensus is a very optimistic word." He won't take on a client without direct access to the board or managing partner. Executive commitment, benchmarking against peer firms and an agreed plan are what make execution possible.
AI as the 'super KM tool'
Knowledge management kept coming up. Firms tell Peter they have five million documents in the DMS and can't find anything. A partner at a UK firm admitted his M&A practice simply didn't collaborate: there might be five partners in the group, but they don't talk to each other or work on each other's matters.
Peter's usual advice is to identify gold-standard documents and distil them into practice playbooks. That conversation sparked a bigger idea: use AI to bring every partner's best practices together, for the first time, in a playbook that is "representational of all their best ideas." Tim, who did KM consulting around 2000 when it drew "largely an eye roll," called that the golden ticket: turning tacit knowledge into explicit knowledge that is secure and consumable. As Tim noted, the old search rule still applies: garbage in, garbage out.
The token bill nobody budgeted for
This is the issue Peter calls "near and dear to my heart." One firm he works with was an early adopter of a leading AI platform on a per-seat licence that came with a generous token allowance. The contract moved to consumption-only pricing, and the early signs have him "absolutely terrified," to the point that he has briefed the board.
Some attorneys decide their practice is unique and want the newest, most expensive model the moment it ships. The firm is seeing around $1,000 a day in token costs, with no way to charge them back to clients, and monthly spend running four to five times what it was. In month one.
"We were budgeting $15,000 a month and now we're well over $60,000."
The firm is now actively looking to move those attorneys onto a licence-based tool it already has, with unlimited usage, "because the economics are completely upside down."
Better prompts, cheaper models
Tim's suggestion of teaching better prompting is exactly where Peter has focused. He kept hearing lawyers blame IT for poor adoption because "nobody taught us how to talk to this thing."
While using meta prompting, where the model helps you write the prompt, he kept using the word "craft." His military background did the rest: CRAFTED became an acronym, and then a book on legal prompt engineering, now on Amazon.
He points to a recent study which found that a well-structured prompt gets better output than a best-guess, natural-language prompt, even one run on the most powerful model. The cost link is direct. With better prompting, a firm may not need the top-tier model at all and might do the work on a smaller one "at one, one hundredth of the cost."
There are two further benefits. Good prompts can go into a shared library and become firm IP. And they are largely model-agnostic, working across Harvey, Legora, CoCounsel, Claude or Lexis, because most tools run on the same handful of foundation models.
"A well-crafted prompt also allows you to save it into a prompt library and share it and start to build this institutional IP and not have to reinvent the wheel each time."
The pricing conversation is coming
Peter's prediction for the next 12 months: as the market shifts to consumption pricing, law firm AI expenses will rise two, three or four times. At the same time clients have flipped from hesitant to "I want you all in, I want you using AI on my matters."
He sees two ways to have that adult discussion with clients:
An AI-enhanced billable rate. Using simple numbers, a lawyer who bills $500 an hour might bill $750 when working with AI, a win for both sides.
Token costs passed through at cost. Track consumption by matter and attorney and charge it without markup. If $1,200 in tokens saved three to five hours of work, the client comes out ahead.
His framing for clients: would you rather pay for five hours of attorney time at $1,500 an hour, or $1,000 in token costs? He also wants the ABA to weigh in, and suggests that without an industry consortium, pricing will stay a problem as consumption models spread.
Not just for big law, and not just for lawyers
Peter's sweet spot over the past two years has been firms in the 20 to 250 range, across the US, UK, Australia and Hong Kong, including bespoke training for attorneys, paralegals and legal assistants. He also reminds firms not to forget the business side of law. Marketing, business development, HR and talent use cases have "immediate ROI and it's really easy to see the impact."
What law firm leaders should take from this
Read your AI contracts again. A move from per-seat to consumption pricing can multiply spend in a month.
Fewer tools, shorter pilots. Pick a lane and commit before competitors pass you.
Train people to prompt before you upgrade the model. Better prompts can mean cheaper models.
Build a prompt library and playbooks. Turn individual know-how into firm IP.
Get the managing partner in the room. Executive sponsorship is what makes execution possible.
Start the pricing conversation with clients now. Decide whether AI shows up in your rates, your disbursements or both.
Watch the full conversation
This episode is part of People in Legal's ILTACON 2026 coverage, alongside conversations with leaders from ILTA, iManage, NetDocuments, BigHand, Cicero and more.
Quotes lightly edited for clarity.
People in Legal is where legal leaders and legal technology meet. Want your leaders in front of our audience at the next event? Partner with us or email team@peopleinlegal.com.
Comments