What AI workflows are useful for a professional services firm?

Professional services firms share a structural constraint: the revenue depends on a small number of senior people whose time is the binding limit, and a great deal of what reaches them does not need to. Enquiries that will not convert, proposals rebuilt from scratch, questions that were answered last year for someone else. The workflows that matter are the ones that intercept those, and the ones that do not matter are the ones that make junior work faster without changing what reaches the top.

Enquiry scoping is the first. An enquiry arrives as a description of a situation and a partner needs a scoped question: what kind of matter, what size, what timescale, what the client has already tried, whether there is a conflict to check. Extracting those and asking the missing ones before the enquiry reaches a partner converts a conversation into a briefing, and it also produces the record that makes the enquiry countable.

Proposal drafting is the second and is the clearest saving. Most proposals are substantially similar to previous ones, and the work is retrieval rather than writing: find the engagements that resembled this, take the scope language, the approach and the team structure, and assemble a draft. This is a retrieval task over the firm's own material, the sources are named, and a partner edits rather than composes. The failure to avoid is the proposal that inherits a commitment from a previous engagement that does not apply to this one, which is a specific thing to check for rather than a general caution.

Knowledge capture is the third and the most valuable over time. Firms lose expertise when people leave and duplicate work when nobody knows an answer already exists. Making it cheap to capture — a recorded conversation turned into a note, a completed matter summarised into a reusable form — addresses both. The material has to be findable afterwards or the exercise is decorative, which means it belongs in whatever the firm actually searches rather than in a document store nobody opens.

Two constraints shape everything here. Client information is confidential, frequently by contract as well as by professional obligation, so which tools may see what is a question with an answer that should predate the tooling. And the firm's output is its professional judgement, which means anything that reaches a client carries a name, and the review that implies is not overhead but the product.

The workflow that most firms should not build first is the client-facing one. The internal saving is larger, the risk is lower, and the constraint being addressed is real. A client-facing assistant is visible and impressive and does not touch the thing that actually limits the firm.