How AI Is Actually Changing Law Firm Pricing: Fixed Fees, Capped Fees and Billable Tokens
AI isn't killing the billable hour, it's making certain work predictable enough to price differently. Here's a practical guide to the four fee models firms are actually using, and how to choose between them.
“AI will kill the billable hour” has been a confident prediction in legal commentary for several years now, and it keeps not quite happening, for a reason most of that commentary misses. The billable hour was never really about tracking time for its own sake. It’s a risk allocation mechanism: because a firm cannot reliably predict how complex a given matter will turn out to be before starting it, hourly billing shifts that uncertainty onto the client, who pays for however long the work actually takes. Any pricing model that wants to replace it has to solve the same underlying problem, not just move faster.
This is where AI’s actual effect on pricing is more interesting than the “it makes things faster” framing suggests. Speed alone doesn’t let you quote a fixed fee with confidence, because a task that usually takes ten hours but occasionally takes thirty is not a task you can safely price at a flat rate just because AI makes the ten-hour version faster. What actually unlocks a different pricing model is predictability: knowing, with reasonable confidence, what a given category of work will cost to deliver, consistently, before you start. That is the specific thing AI is beginning to provide for certain task types, and it’s why fee structures are shifting exactly where that predictability shows up, and nowhere else.
Where the Market Actually Is
The shift is already broader than most firms realise. Roughly 72% of firms in the US now offer some form of alternative fee arrangement, rising to around 90% among larger firms with fifty or more lawyers. The gap worth paying attention to is between offering an alternative fee arrangement as an option and actually re-engineering how a specific category of work is priced around the predictability AI provides. Most firms have done the former. Far fewer have done the latter with any rigour.
The Four Models Firms Are Actually Using
Fixed fees, for genuinely predictable work. Contract review against a known playbook, standard due diligence checklists, and routine compliance reviews are the clearest candidates, because AI-assisted review of this kind of work now has a track record consistent enough to quote against confidently. The discipline required here is resisting the temptation to fix-fee anything that merely feels routine; only quote fixed fees for categories of work where you have enough delivery data, AI-assisted or otherwise, to know the actual cost distribution, not just the average.
Capped fees, for phased matters where full predictability isn’t realistic but a ceiling is. Litigation is the obvious example: you cannot fix-fee an entire dispute, but you can cap fees for a defined phase, pleadings, disclosure, a specific application, giving the client budget certainty for that phase while the firm retains the upside if AI-assisted work comes in under the cap.
Portfolio pricing, for clients who generate a recurring volume of similar matters. Rather than pricing each matter individually, the firm prices the portfolio as a whole, betting that AI-driven consistency across similar matters evens out the variance that would make individual fixed fees risky. This suits in-house relationships with steady transactional or compliance volume better than one-off engagements.
Billable tokens, an emerging and still largely experimental concept where the unit of billing shifts toward the AI compute or processing involved in a matter rather than lawyer time alone. This is worth watching rather than adopting yet. It’s conceptually interesting because it acknowledges that AI-assisted delivery has its own cost structure distinct from lawyer hours, but the market has not converged on a standard way to implement it, and a firm leading with this model today is doing genuine experimentation, not adopting settled practice.
How to Actually Decide Which Model Fits Which Work
The decision framework that matters is not “does this matter involve AI.” It’s “how predictable does this specific task category become once AI is genuinely part of the workflow, based on actual delivery data, not assumption.” A task AI makes faster but not more predictable stays on the billable hour. A task AI makes genuinely predictable, where you can point to a consistent track record of cost and turnaround across similar past matters, is a candidate for fixed or capped pricing. Confusing “AI touches this work” with “this work is now predictable enough to price differently” is the mistake that produces fixed fees firms later regret.
One further practical point worth internalising: pricing innovation in this market tends to spread through competitive pitches for new work, not through renegotiating fee structures on existing relationships. If you want to pilot a new pricing model built around AI-driven predictability, the next competitive pitch is a considerably more natural place to introduce it than an attempt to restructure how you already bill a long-standing client.
Why This Matters More Than the Billing Rate Conversation Alone
AI investment on its own has not been shown to justify higher hourly rates, because the underlying capability is too evenly distributed across competing firms to function as a premium justification. Turning that same capability into a concrete, differently structured pricing offer, a fixed fee where a client used to get an hourly estimate, is a genuinely different value proposition, one a client can actually perceive and choose you for. That’s the more durable competitive position, and it’s a pricing strategy question your Productivity and Practice Management planning should be addressing directly, not something to leave to whichever partner happens to negotiate the next engagement letter.