The Legal AI Number Nobody Owns

There is no agreed method for measuring legal AI ROI. Waiting for one is the surest way to never report a number, and the business won't wait with you.

At a roundtable Execo hosted in September, one in-house legal leader described a position many teams will recognize. The tool has just gone live, the company has set its AI targets, and every function head now has AI goals in their performance review. What nobody has settled is how to count legal’s share.

The tool's usage report covers only part of it. The rest has to be collected by hand, and hand collection takes discipline, which is exactly where measurement tends to stall. What holds is a routine: a chosen level, three measures with the cost beside them, a dated baseline and one person who owns the numbers. 

Own legal’s number, not the company’s 

"How do you measure AI ROI?" has no single answer, because it gets asked at several levels at once. The company may set group-wide ambitions for what AI should deliver. Each function measures its contribution in its own way. Underneath that sit the tool's usage report and benchmarks on individual tasks.

Legal can’t answer for all of that, and shouldn’t try. Report at the level of legal’s own business case, the one number legal can stand behind. The company-wide figure belongs to whoever set it. The usage report is an input, not the answer.

Start by asking whoever set the company's AI target what was committed, and at what level. The answer will tell you which of the measures below will carry the most weight. If there is no company target, legal's own business case stands on its own, and that's a perfectly good place to start.

Three returns, and what they cost 

A business case that holds up rests on three measures: capacity, hires avoided, and outside spend. Finance will file each one under its own label, so use theirs from the start.

Capacity is time freed on a task: hours before against hours after, multiplied by a year's volume. Volume comes from the intake system, the matter list or a CLM report. If nobody tracks volume, count one quarter, annualize it and say that's what you did. Start with tasks that drop from hours to minutes. They make the case fastest.

Finance calls this productivity. It’s real, and it counts for more once the team has put the time to use, so add a short line on what the freed hours went to. Before the first report, agree with finance what a freed hour is worth. Use the team’s internal loaded cost, meaning salary plus overhead. Outside-counsel rates would overstate it, and finance will know.

Report turnaround alongside capacity. Capacity counts hours of work, while turnaround counts elapsed days. A contract can take two hours of work and still take ten days to turn. Turnaround is usually the change the business feels first. Use it when the question is speed.

Hires avoided applies when legal had a hire request, open or already in plan. The hire the team no longer needs is a figure finance can price. It only counts against a real request. If there wasn't one, drop the line and let capacity carry the report.

Outside spend can fall where the tool takes over volume work, such as first-pass review of routine agreements, and firms stay on the work that needs their judgment. Ask finance for spend by matter type. Where firms bill by task, the comparison gets sharper.

This is the hard saving, money that no longer leaves the budget. It only earns that label once you’ve accounted for what else moves spend: deal volume, rates and panel arrangements. Name them in the report. Then check it against capacity. Spend can fall because the tool took the work, or because the hours quietly moved onto your own team, and the capacity line is where that shows up.

Then the cost. List all of it: the license, implementation, the team’s time on rollout and training, ongoing fees and the quarterly collection itself.

Three rules keep the whole thing credible. Lead the report with the hard line. Count every hour once: an hour that shows up under capacity and again under hires avoided will be found and discounted, and it takes every other line down with it. And estimate conservatively, then say how. A smaller figure the business accepts beats a bigger one it doesn’t.

The baseline, and when to set it

Every measure above needs a "before": what the same work took, in hours, days and money, when it was done the old way. The best time to record it is before go-live. Once the old process is gone, so is your comparison. Put a date on it. We made the case for measuring before the old process disappears in From Pilot to Production.

Pick the matter types the tool will touch. For a sample of each, record how long the work took, how long it took to turn and what outside counsel charged. If your team doesn't record time, build the "before" from past matters. Ask the people who handled a handful of comparable ones how long the work took, and write the answers down with a date.

Already live? Rebuild the baseline from records first: matter open and close dates in the intake system or CLM, email and document timestamps, and outside-counsel invoices on comparable matters. Use recall only to fill the gaps, and take the lower estimate whenever two numbers compete. It makes the gain smaller and harder to argue with.

Say in the report that the baseline was rebuilt. It's weaker than a recorded one, and saying so up front is what makes the rest credible. If both the records and the people who did the work are gone, use the first quarter after go-live as your baseline and report change from there.

One owner, every quarter

In a survey by the Association of Corporate Counsel, released in November 2025, 83% of 284 chief legal officers, general counsel and legal ops professionals worldwide said they track outside-counsel spend. Only 12% track technology ROI. That figure covers legal technology in general, not AI specifically, but the gap is hard to miss.

The reason is simple. Outside spend sits in invoices with a finance owner behind it. AI value usually has no owner, and the person who feels the benefit, the lawyer with their hours back, is rarely the person who reports it.

So name one person per team to submit the hand-collected figures each quarter. Whoever already prepares the legal budget report is the natural choice. In the report, mark which lines come from the tool's usage report and which are gathered by hand.

Before go-live, agree three things with your finance partner: what counts as time freed, which outside-counsel spend you'll compare against, and the date of the first review. One legal leader we spoke with plans a first measurement about three months after launch, and that's a sensible default.

And start crude. Report AI value next to outside-counsel spend on the same schedule, so it sits beside a number finance already owns. The first comparison can be as simple as whether outside spend held steady or fell after go-live. Note its limits, then break it down by type of work once reporting is a habit.

A rough number, and someone who owns it

You don't need everything in place on day one, and nobody expects you to. Start with the rough version and let it get sharper each quarter: by type of work, by team, by what the freed time went to. Each report makes the next conversation with the business a little easier.

There’s still no agreed method, and there may never be. But a rough number the business recognizes as its own beats a perfect one nobody reports. And if the survey above is any guide, simply reporting one puts you ahead of most legal teams.