A regional Detroit law firm
A 40-lawyer regional firm in greater Detroit, with a strong mid-market litigation and corporate-transactions practice. The paralegal team handles the bulk of document review, deposition preparation, and discovery indexing, work that scales linearly with case complexity, which had been climbing for two years. Firm leadership wanted to use LLMs to compress the linear scaling, but had no in-house AI capability and was reluctant to bring in an outside vendor with access to privileged client material.
The firm had a confidentiality problem more than a technology problem. Off-the-shelf consumer LLM products were not an option, client material could not leave the firm's controlled environment. Generic AI training would teach the team to use tools they could not actually deploy. Hiring an in-house AI engineer was not justifiable for a 40-person firm. Leadership needed a way to build durable in-house capability without exposing client data and without taking on a permanent specialist headcount.
We ran a three-day AI Workshop on premises with twelve paralegals and two associates. The first half-day covered the realistic capability envelope of current LLM tooling, what works reliably, what fails quietly, what to never trust. The remaining two and a half days were hands-on building, using the firm's own redacted document corpus and a private LLM deployment we had pre-configured. Each participant left with a working prototype tied to one specific recurring task in their day-to-day workflow. We deliberately wrote no code for the team during the workshop; everything the team built, the team owned.
Within sixty days, four of those workshop prototypes had been promoted into actual internal tools used by the firm, a deposition-prep summariser, a discovery indexer, a contract-clause finder, and a billing-narrative drafter. Two of those tools were extended by the team itself with no input from us. The firm declined our follow-up offer for a deeper development engagement, on the grounds that the team no longer needed external support to build what they wanted to build. That outcome is the offer working exactly as designed.
Published April 2, 2026. Sector and figures are generalised where necessary to keep the client comfortable being represented here. The structure of what we did and what changed is exactly as it ran.
Tell us the situation on a free first call. We will say which engagement family fits and what it would cost. If none of our offers fit, we will tell you that too.
Schedule a free consultation