How we helped a business law firm save $200K+ a year on AI tools and cut partner review from 90 to 20 minutes.

Their AI investment still ended at the same review queue.

The firm's leaders wanted to see what their AI investment was buying. They also wanted more of their partners' knowledge available to the lawyers preparing deal documents. Both goals depended on whether the work reaching review was actually getting better.

But associates still needed partners and senior associates to bring drafts back to the firm's negotiating positions. An AI-assisted draft could contain plenty of text and still leave that same correction work for a partner. Producing a document hadn't removed the bottleneck.

The knowledge already existed in past mark-ups, closed deals, and partner judgment. It wasn't yet a shared standard that lawyers and their tools could consistently use, which is why we began by measuring the review work before deciding what to build.

Most first drafts still needed a partner to correct standard positions.

We asked to study one working week, but the firm first needed to understand who could see client work. We explained how the study would report work types without exposing deal content to us. It ran on their own systems and included lawyers who used AI and lawyers who didn't.

Only ~15 to 20% of first drafts passed without partner overrides on standard firm positions. Median partner review took 90 minutes per primary deal document. Associates and partners were going back and forth on positions the firm needed drafts to reflect from the start.

We put those findings into a memo and workflow maps. A practice leader used the readout to choose a route through approved firm positions. And so our job became clear: make those positions available while a draft was being prepared, before a partner had to correct it.

We made partner judgment usable in the first draft.

A collection of old deal files couldn't tell a lawyer which position the firm wanted to use next. So we captured the negotiating positions in past mark-ups and closed deals, wrote them down, and had their partners review and approve them.

That approval was essential. A past deal showed what the firm had done in that deal, but their partners still had to decide which positions should serve as standards. We turned that judgment into an approved reference their lawyers and tools could use across drafts.

We gave lawyers and their tools the approved positions before review, so partners spent less time bringing drafts back to the firm's standard.

The check showed where a draft differed and which past deal supported the position.

With the standards approved, we configured their existing tools first. Where those tools fell short, we built a drafting and review check that compared a draft with the approved positions and pointed back to the past deal behind each one.

That source gave the reviewer something concrete to examine alongside each suggestion. Their documents stayed in the firm's document system, and their security team applied existing access rules before retrieval, so the check worked within the permissions they already used.

The system could bring an approved position to a lawyer's attention, but a partner still had to decide what the current deal required. We built support for that judgment into the workflow while keeping the legal decision with their team.

The check had to work without copying the completed test deal.

We tested the drafting check against closed deals their partners already knew. But giving the system the completed test deal would let it draw from the answer, so we restricted each test to earlier deals as sources.

That meant the check had to work with information the deal team could have had at the time. Their partners reviewed the suggestions and accepted, corrected, or rejected them.

We'd agreed on first-draft acceptance and partner review time as the measures because those were the problems we'd set out to improve. The test kept attention on whether the check helped produce a useful draft and reduce correction work.

More drafts met the standard, so partners spent less time correcting them.

We put the system into daily drafting and review. Their firm directed lawyers to the approved route and retired the former one, while we tracked rework, repeated questions, and the types of deal work brought to the system.

In the follow-up study, 78% of first drafts passed without partner overrides on standard positions. Before our work, only ~15% to 20% passed that check. More drafts reached review with the firm's agreed positions already in place, so partners had fewer routine corrections to make. They still reviewed the work and made the changes each deal required.

Median partner review fell from 90 to 20 minutes per primary deal document. Both studies covered one working week in the same group, with stable staffing and deal mix. The follow-up week carried more concurrent matters.

Their leaders could now judge the investment against a measured change in the work: more drafts meeting firm positions and less partner time spent bringing them there. They used that evidence in internal decisions about how to support the work.

The approved workflow also replaced paid AI tools. Retiring those tools removed $200K to $300K a year in recurring AI expense, alongside the measured reduction in partner review.

We taught their team to carry the method into another practice.

The firm's standards would need to cover new deal types, so instructions for using the check weren't enough. We trained their knowledge team and practice champions to run work studies and add approved positions.

Their team went on to apply the method in a second practice without us. Our training gave them a repeatable way to capture approved judgment, use it in drafting, and measure what changed. They could extend the capability we'd built as the firm's needs changed.

Inside the drafting check and its evaluation.

Read how we handled approved positions, document access, historical testing, and measurement.

Their partners' knowledge now shaped the draft before review.

They came to us to make their AI investment and shared knowledge more useful, but producing drafts still left partners correcting the same standard positions. We measured that burden, helped establish approved standards, and built the check that brought them into drafting.

As a result, more first drafts met their standard and partners spent less time correcting them. We also gave their team the skills to take the method into a second practice, and they did. Their investment produced a measurable improvement in deal work and a capability they could keep extending.