The three AI projects every professional-services firm should do first

Professional-services firms — law, accounting, consulting — are sitting on the ideal conditions for AI: enormous volumes of text, repetitive document work, and expensive people doing some of it. And yet most of them either do nothing or leap at the flashiest possible project and stall.

Here's the order we'd actually recommend, drawn from the engagements we've had. Three to do, in sequence. Two to leave alone for now.

Do first: document search that actually understands the question

Every firm has a graveyard of past work — contracts, filings, memos, prior advice — that contains most of the answers to today's questions. The problem isn't that the knowledge is missing. It's that finding it depends on the one partner who remembers which matter it was in.

A retrieval assistant over your own document archive turns "ask the partner who's been here 20 years" into "ask the system, get the answer with a citation to the source document." It's the highest-value, lowest-risk place to start because it's grounded — the AI quotes your own files rather than inventing anything — and the value is obvious from day one.

Why first: highest return, lowest risk, and it builds your team's trust in AI on a task where the answer is checkable.

Do second: first-draft generation for your standard documents

Every firm produces the same document shapes over and over — engagement letters, standard clauses, recurring report sections, routine correspondence. The expert work is in the judgment and the edge cases, not the boilerplate that surrounds it.

An assistant that produces a solid first draft from your own templates and precedents doesn't replace the professional — it deletes the blank page. The expert still reviews, edits, and owns the output. They just start at 70% instead of 0%.

Why second: big time savings on work your people don't enjoy anyway, but it needs the trust and the grounding you built in project one.

Do third: intake and triage

By the time a new matter or client reaches the right person, it's often passed through three people who each re-read the same email thread. AI-assisted intake — summarizing the incoming request, extracting the key facts, flagging what's missing, routing it to the right team — compresses that.

Why third: it touches more of your process and more of your people, so it's worth doing only once the first two have earned organizational confidence. Lead with this and you'll spend your goodwill on change management instead of results.

Don't (yet): the autonomous "AI associate"

The pitch is seductive: an AI that takes a matter and runs it end to end. The reality in 2026 is that this is where projects go to die. The judgment, accountability, and edge-case handling that define professional work are exactly what current systems are worst at, and the liability of getting it wrong is exactly what your firm cannot absorb. Build the three above, get years of grounded, supervised use behind you, and revisit. Not now.

Don't (yet): a public-facing chatbot that gives advice

A bot on your website that answers prospective clients' legal or financial questions is a regulatory and reputational hazard wearing a friendly avatar. The moment it gives something that looks like advice and gets it wrong, the problem is yours. A grounded internal tool is an asset. An ungrounded public oracle is a liability. The line between them is whether a professional reviews the output before anyone relies on it.

The pattern

Notice what the three "do" projects have in common: they're grounded (the AI works from your real documents, not its imagination), they keep a human in the loop, and they start internal where mistakes are cheap. That's not a coincidence — it's the recipe for AI that adds value without adding risk.

If you want this turned into a costed plan for your specific firm — which projects, what they'd cost, what they'd save — that's exactly what an AI Roadmap delivers. Or book a free AI Review and we'll sketch the priorities with you in half an hour.