Law firms running their casework AI and their client-discovery AI as two separate projects end the year with two half-built systems and nothing worth showing a managing partner. The matter side gets a tool that drafts and summarizes. The marketing side gets a website rewrite aimed at AI answer engines.
Neither team talks to the other. The budgets don't line up. And the questions a prospective client is now asking a chatbot never make it back to the intake workflow. Firms making real progress figured out early that this is one shift wearing two costumes, and staffed it accordingly. A useful look at how attorneys are rebuilding casework and client acquisition at once shows that the two projects share a spine.
The pressure is arriving from both directions at once. Inside the firm, associates are already using generative tools whether the partnership has blessed them or not. Outside the firm, the person with a legal problem is asking an AI model before they ever type a firm name into a search bar. Handle those as separate initiatives and the firm ends up answering the same strategic question twice, poorly, with two different vendors.
Phase One: The Two Fronts Open Independently
Most firms are still living inside the first phase, and it looks like parallel improvisation. A litigation group pilots a research tool. A transactional group tries a contract-review model. Marketing hires someone to figure out why the firm no longer appears in AI answers when a prospect asks who to call about a wrongful termination.
Practice-side adoption has moved fast. A recent survey found the share of legal organizations actively integrating generative AI nearly doubled between 2024 and 2025, and a meaningful slice of law firm users are in these tools multiple times a week.
None of that adoption is coordinated. The associate using an AI drafting assistant on a memo has no idea what the marketing team is publishing to make the firm citable in a chatbot answer. The marketing team has no idea which practice areas the associates are getting faster at, which is exactly the information that should shape where the firm spends to attract new matters.
Phase Two: The Cracks Show Up in Governance and Intake
The cracks appear a quarter or two in. Confidentiality questions surface. Someone asks who approved the tool the summer associates are pasting client facts into.
The website starts appearing in AI-generated answers, but for the wrong practice area, and nobody owns the fix. This is where firms discover that the governance work for the casework side and the content work for the discovery side are the same work, done by different people, with no shared vocabulary.
The numbers also get uncomfortable around here. A prospective client asking a model for help with a landlord dispute is getting a synthesized answer plus, sometimes, a nudge toward hiring a lawyer. That nudge is a referral channel the firm never signed up for and cannot see in its analytics. Meanwhile, the intake team is still counting form fills from a search funnel that is shrinking.
Phase Three: A Single Program Replaces the Two Pilots
Firms that get past the cracked-pilot stage do one specific thing. They stop running AI as two workstreams and stand up a single program with one owner, one budget, and one roadmap. Gowling WLG's 12-week pilot is a useful reference point. The firm rolled out generative AI across practice groups and business services in parallel, with a lawyer-in-the-loop protocol and structured iteration, rather than letting each department run its own experiment.
A unified program forces decisions separate pilots let firms avoid. It puts the practice groups and the business-development team in the same room to decide which matter types the firm wants to attract more of, then aligns both the internal tooling and the external content to that answer.
Phase Four: What the Program Actually Contains
A real program has a handful of parts the fragmented version never builds. None of them are unusual. They're the pieces that fall through when two teams each assume the other is handling them.
Phase Five: What the Program Buys You
The payoff isn't that the firm becomes an AI firm. The payoff is that the firm stops paying twice for overlapping work and starts compounding the two efforts against each other. A well-run intake process feeds the marketing team real language from real prospects, which sharpens the content the models cite, which sends better-qualified prospects to intake. The loop closes.
Firms that keep the two efforts separate will still see gains. An associate saves an hour here. A landing page ranks in an AI summary there. But the results will read like line items rather than a business. Managing partners who want the second thing, not the first, should be building one program this year instead of approving two more pilots.
Timely Updates 
Headlines Team