AI is making legal work faster. That does not mean it is making law firms more profitable.
For firms that still earn most of their revenue by selling time, the problem is basic arithmetic. If a task used to take five hours and now takes one, the invoice gets smaller unless the firm finds more work, changes the fee, or puts the saved time somewhere more valuable.
This is the efficiency paradox. The tool works. The economics do not.
Clio’s 2026 report on solo and small firms makes the gap unusually clear. AI use has reached 71% among solo practitioners and 75% among small firms, but fewer than one third of those firms report revenue growth from AI. Enterprise firms are close to 60%.
The difference is not access to a chatbot. It is what the firm does after the work gets faster.
The invisible discount
Hourly billing connects revenue to elapsed effort. AI separates them.
Imagine a recurring task that normally takes seven hours at $300 per hour. The expected invoice is $2,100. A lawyer uses AI for the first pass, spends one hour reviewing the output, and finishes the work in three hours total. Under an hourly agreement, the invoice is now $900.
The client received a $1,200 discount. The firm may have delivered the same quality faster, reduced turnaround time, and freed four hours of capacity, but none of that value appears on the invoice.
The example is illustrative, but the underlying pattern is already visible in the data. Clio found that 86% of solo firms and 78% of small firms have made no pricing changes after adopting AI. The technology changed the cost of delivery. The commercial model stayed still.
If the saved hour has no destination, it is not an asset. It is simply time that disappeared from the invoice.
This is why measuring AI only in hours saved is misleading. The important measure is what happens to those hours next.
Build a saved-hour ledger
Most firms can see software spend and billable time. Few can see what happens to time released by automation.
A saved-hour ledger does not need to be complicated. For each workflow, record:
- the matter type and task
- the old delivery time
- the new delivery time
- the time required for review and correction
- the fee charged before and after AI
- where the released capacity went
The last field matters most. Useful destinations include intake, business development, faster matter turnaround, client communication, higher-value legal work, and additional matters.
If the answer is “nowhere,” the firm has found a productivity benefit, not a business benefit.
This is also where planning needs to become concrete. A partner who saves four hours on drafting will not automatically spend four hours developing new business. If that time is expected to support growth, it needs an owner, a calendar slot, and a measurable outcome.
Tool adoption is not workflow adoption
Many firms describe themselves as AI adopters because lawyers have access to ChatGPT, Microsoft Copilot, or another general tool. That is tool adoption. It is not the same as redesigning a workflow.
Clio reports that 47% of solos and 48% of small firms use consumer-grade AI tools. Those tools can help with isolated tasks, but they usually sit outside the matter system. The lawyer copies context into a separate window, prompts the model, checks the answer, and moves the result back into the firm’s normal process.
That creates a second tax. The firm saves time inside the task, then gives some of it back through context switching, manual transfer, inconsistent prompts, and repeated review.
A real workflow has a defined starting point, approved data, a known output, a review step, and a destination. It might look like this:
- A new matter or document enters the approved system.
- The system applies a task-specific instruction and the correct matter context.
- The first pass is produced in a consistent format.
- A named lawyer reviews the work against a checklist.
- The approved output is saved back to the matter record with an audit trail.
The model is only one part of that system. The integration, permissions, review logic, and pricing model determine whether the result is useful.
Decide what the client is buying
The pricing conversation gets easier when the service is defined clearly.
For routine, repeatable work with stable scope, a fixed fee can let the client buy certainty while the firm retains some of the benefit from efficiency. For uncertain work, a capped fee, phased fee, or collar can share risk without pretending the scope is perfectly predictable. For genuinely open-ended work, hourly billing may still be appropriate.
The point is not to replace every hourly engagement. It is to stop using hourly billing by default for work whose delivery has already become predictable.
Start with one matter type. Choose work that is frequent enough to measure, similar enough to scope, and valuable enough that the client cares about speed and certainty. Then calculate:
- the expected delivery cost, including review
- the range of likely exceptions
- the fee that reflects the outcome and risk
- the trigger that moves work outside the agreed scope
Without those numbers, a flat fee is a guess. With them, it becomes an operating model.
Do not hide the AI
There is a tempting short-term response to the efficiency paradox: use AI internally, keep billing as before, and avoid the pricing conversation.
That is not a durable strategy. Under ABA Formal Opinion 512, lawyers billing by the hour must bill for the time actually spent. Client billing guidelines are also becoming more explicit about disclosure, data handling, review, and AI-related charges.
The better position is transparent and operationally defensible:
- explain where AI is used
- explain who reviews the output
- explain how client data is protected
- explain what the client is paying for
- tie the fee to a defined service and scope
That conversation is easier before an invoice is disputed.
A practical 30-day response
Firms do not need to redesign every fee arrangement at once. A focused first month is enough to expose the economics.
Week 1: pick one workflow
Choose a recurring task where AI is already being used. Document the current steps, systems, people, time, and fee.
Week 2: measure the real delivery cost
Include prompt preparation, data gathering, lawyer review, corrections, and administrative handoffs. Do not count only the model’s processing time.
Week 3: test a different fee
Build a fixed, capped, or phased fee with a clear scope boundary. Run it against recent matters to see where margin would have landed.
Week 4: assign the saved capacity
Decide in advance whether released time supports more matters, faster turnaround, client service, or business development. Put the decision into the workflow and calendar.
The operating model is the product
AI does not automatically create margin. It changes the relationship between effort, time, and output. The firm still has to decide how work is packaged, how it is reviewed, how it is priced, and where the released capacity goes.
The firms that benefit most will not be the ones with the largest collection of tools. They will be the ones that connect pricing, workflow, and data controls into one system.
Want to turn saved time into a better operating model?
Jinka helps law firms map AI-assisted workflows, connect them to matter systems, and build the controls needed to price and deliver the work with confidence.