Most firms do not lose billable hours at the point of capture — they lose them between capture and the invoice, in write-downs, unbilled disbursements and work performed outside an agreed scope. AI helps by making the leak visible entry by entry, but the amount you can bill is capped by a professional standard, not by what the system recorded.
Key takeaways
Four leaks account for most of the loss in a small professional firm, and only one of them is a capture problem. The first is unrecorded work — short calls, quick reviews, the email that answered a question. The second is recorded work that is written down at billing review because the narrative is too thin to defend. The third is disbursements incurred and never rebilled. The fourth is work done outside the agreed scope, which is not a billing failure at all but a scoping failure.
Firms that go looking for the leak almost always start with the first one, because it is the one software vendors sell against. In practice the second and fourth are usually larger, and neither is fixed by capturing more.
So the first useful exercise is arithmetic, not technology: for one month, compare hours recorded, hours invoiced and cash collected on the same set of matters. The two gaps in that chain tell you which problem you actually have.
Unrecorded activity is the easiest win. Practice-management tools already generate entries from calendar events, tasks, notes, documents and email — Clio describes adding time entries directly from those sources and correcting them afterwards. What a model adds is triage: pointing at the twenty-eight minutes of activity on a matter that produced no time entry at all, and drafting the entry for review.
Thin narratives are the second. An entry that says "attend to file" gets written down; an entry that names the document, the act and the reason survives. Rewriting weak narratives into specific ones, from the underlying activity, is exactly the kind of work a language model does well — and it is reviewable in seconds by the person who did the work.
Unbilled disbursements are the third and the most overlooked. Filing fees, searches, couriers and travel that were paid by the firm and never made it onto a bill are pure loss. Matching a payment record to an open matter is a data problem, and it is the sort of reconciliation covered in more depth in AI for expense reports at a firm.
There is a hard limit on this exercise. In Ontario, rule 3.6-1 of the Rules of Professional Conduct provides that a lawyer must not charge or accept any amount for a fee or disbursement unless it is fair and reasonable and has been disclosed in a timely fashion. The commentary lists what makes a fee fair, including the time and effort required and spent, the difficulty and importance of the matter, the results obtained and any estimate or range of fees the lawyer gave.
Read that list next to a tool that promises to increase captured hours. The estimate you gave the client is one of the fairness factors. If better capture takes a matter well past the range you quoted, the answer is a conversation, not a larger invoice.
British Columbia’s Code of Professional Conduct states the same duty in rule 3.6-1, and in Quebec the Barreau du Québec points out that no directive dictates a pricing model — the choice is yours, subject to the Code of ethics of advocates. Accounting practices sit under their provincial CPA body rather than a law society, but the professional-judgment principle is the same: the engagement terms bound the bill.
The numbers here are illustrative, not survey data. Suppose a five-lawyer firm records 6.1 chargeable hours per fee earner per day and bills 4.7. On review, 0.5 is written down for thin narratives, 0.4 is written off as duplicated or inefficient work, 0.3 is out-of-scope work being absorbed rather than charged, and 0.2 is disbursements never rebilled.
Only the first and last are worth pointing software at, and together they are 0.7 hours a day per fee earner. The 0.4 write-off is a training and delegation question. The 0.3 is an engagement-letter question — the scope was too loose, which is why work outside it feels unbillable.
This is the reason the arithmetic comes first. A firm that installs capture software to solve a scoping problem will conclude, correctly, that the software did not work.
Pick one practice area. Turn on activity-based entry proposals for it, leave every other setting alone, and instrument three numbers weekly: hours recorded, hours invoiced, and the write-down rate at billing review. If recorded goes up and the write-down rate goes up with it, you are generating entries nobody can defend.
Then fix the narrative standard before you scale. Agree what a billable entry must name — the document or the person, the act, and the reason — and have the model draft to that standard rather than to a generic one.
If the firm’s real problem is administrative load rather than leakage, the better starting point is turning admin hours back into billable hours.
Will this increase revenue?
It can increase billed hours where genuine work was going unrecorded. It cannot increase what a client will fairly pay for a matter, and treating it as a revenue lever rather than an accuracy lever is how firms get into trouble.
Is automatically rounding entries acceptable?
Rounding rules are common and are configurable in most billing systems, but the fee that results still has to be fair and reasonable and consistent with what you told the client.
What about fixed-fee work?
Keep recording time anyway. On fixed fees the record is how you learn whether the price was right, which is a pricing input rather than a billing one.
A 30-minute call is enough to tell you whether AI pays for itself here.
