Billable Hours When AI Does Part of the Work

OpenAI shipped a legal research configuration this week aimed at large firms, and vendors including Harvey and Legora are building products on it. Set aside whether it is good enough yet. The pricing question arrives before the capability question does, because any firm that bills by elapsed time has just been handed a tool that reduces its own invoices.

Which tasks compress

Not all of them, and the pattern is consistent across professions. The compressible work is research, first drafts, document review, summarising, and pulling structured information out of documents. Measured in hours historically, all of it, and all of it now faster.

What does not compress is the judgment about what the answer means for this client, the conversation where somebody has to be told something they will not like, the accountability for being wrong, and the relationship that got you the instruction in the first place.

Hourly billing charges the same rate for both categories. That was defensible when they took similar amounts of time. It is the thing coming apart now.

Four responses

Response	Works when	Fails when
Keep hourly, take the cut	You are capacity-constrained and can fill the time	Demand is flat, so the hours just disappear
Raise the hourly rate	Clients buy your judgment, not your time	Clients compare rate cards line by line
Fixed fee the standard work	The task repeats with narrow variance	You priced it on the good case
Subscription or retainer	Clients want access and predictability	Scope creeps and nobody renegotiates

Most small firms end up somewhere between rows three and four, holding hourly for the work that genuinely varies. That hybrid is easier to sell internally than a wholesale change, and it lets you move one service line at a time.

The trap in row three is pricing on the best case. A task that takes one hour most of the time and four hours when the documents are a mess will lose money at a fixed fee set from memory. Time it across at least ten real instances before you quote it, which is the same discipline as process mapping applied to your own service.

Raising the rate is harder than it sounds

The theory is clean: if you deliver the same outcome in a quarter of the time, the hourly rate should reflect the value rather than the clock. In practice clients compare rate cards, procurement departments benchmark, and a rate that looks out of line invites a conversation you did not want.

It works best where the client is buying a specific person's judgment and knows it. It works badly where your service is perceived as a commodity, which is exactly where AI is compressing hardest.

The disclosure question

Two separate things get muddled here. Whether you tell clients AI was involved, and whether you bill for hours nobody worked. The second is not a pricing strategy and there is nothing to debate about it.

On the first, check your professional body's current guidance rather than a blog post, since expectations differ by profession and are moving quickly. Commercially, disclosure tends to be the easier position. A client who finds out on their own treats it as something you concealed. A firm that explains where AI is used, and where a person reviews everything before it leaves, is describing a quality process rather than admitting to a shortcut.

The review step is what makes that sentence true, and it is not optional given where accuracy currently sits. OpenAI's own benchmark for its legal configuration reported 54% correctness on research questions, which we look at in the best legal research AI gets 54% right. Whatever you tell clients, somebody qualified is checking the output.

What to do this quarter

List your service lines and mark which have compressed. Be specific about the task rather than the matter. Research compressed; the advice call did not.

Time the compressed ones across ten real instances. Including the awkward ones. The variance matters more than the average when you are deciding whether a fixed fee is safe.

Move one service line to a fixed fee and leave the rest alone. Pick the one with the narrowest variance and the most repetition. Run it for a quarter and compare realised revenue per matter against the hourly equivalent.

Write down what you tell clients, once. A short paragraph in your engagement letter about where AI is used and who reviews the work. Having it written removes the improvised answer when somebody asks, and it is the same governance habit as a one-page AI governance framework.

The firms that handle this well will not be the ones with the best tools. They will be the ones who noticed early that their pricing model was quietly converting an efficiency gain into a discount, and changed it while they still had the choice.

Frequently Asked Questions

How does AI affect billable hours?

It compresses the tasks that were always measured in hours: research, first drafts, document review, summarising. A firm billing by elapsed time books less revenue for the same delivered outcome, so efficiency shows up as a smaller invoice rather than a better margin. That is not a reason to avoid the tools, since your competitors are quoting against the shorter timeline whether or not you adopt them.

Should professional services firms move away from hourly billing?

Not necessarily away from it entirely, and not overnight. The workable path for most small firms is a hybrid: fixed fees on the work that has become predictable, hourly on the work that genuinely varies. That takes the pressure off the compressed tasks without forcing you to reprice everything at once or to guess at scope on complex matters.

Do I have to tell clients I used AI?

Check your professional body’s current guidance, since expectations differ by profession and are changing quickly, and treat this as a prompt to look rather than as advice. Commercially, disclosure tends to be easier than concealment: a client who discovers it themselves reads it as something you hid, while a firm that explains where AI is used and where a person reviews everything is describing a quality process. Billing a client for hours nobody worked is a separate and more serious issue.

What is value-based pricing for professional services?

Charging for the outcome rather than the elapsed time, which means the fee reflects what the work is worth to the client rather than how long it took you. It requires enough repetition to estimate reliably, which is why it fits standardised work such as incorporations, standard agreements and routine filings better than complex litigation or an unusual advisory engagement.

How do I price work that used to take four hours and now takes one?

Start by finding out whether it reliably takes one hour or occasionally still takes four, because pricing on the good case is how firms lose money on fixed fees. Time the task across at least ten real instances. If the variance is narrow, a fixed fee set somewhere between the old and new totals usually leaves both sides better off. If it is wide, keep it hourly and compete on turnaround instead.