Like the trades case study on this site, this is a composite scenario built from patterns I see across legal practice clients, not one named firm — legal work involves confidential client matters, and the specifics here are illustrative rather than a real case file.
The practice: a five-partner high-street firm handling conveyancing, family law and probate, with a small team of paralegals and a part-time office manager. The partners were spending a disproportionate amount of billable-hour-capable time on client intake, status-chasing calls, and drafting the same categories of routine correspondence.
The Problem: Good Fee-Earners Doing Admin Work
New client enquiries came in by phone, email and the website contact form, and every one needed a first response, some basic fact-gathering, and a decision about which partner or paralegal should pick it up. That triage was falling to whoever had five minutes — often a partner, which meant qualified legal time was being spent deciding who should talk to a prospective client rather than doing the work only they could do.
Beyond intake, a significant chunk of correspondence followed predictable patterns: acknowledging receipt of documents, confirming next steps in a conveyancing chain, chasing a client for information the practice was waiting on. None of it required legal judgement to draft a first version, but all of it currently required someone qualified to sit down and write it.
What Was Actually Built
Client intake and triage. An AI-assisted intake process now handles the initial conversation for new enquiries — gathering the basic facts of the matter (what kind of legal issue, rough timeline, any conflicts to flag) through a structured web form and a follow-up call script for phone enquiries. It then routes the enquiry to the right partner or paralegal based on matter type and current caseload, with a same-day acknowledgement going out automatically. Nothing about the client's actual legal matter is assessed or advised on by the AI — its only job is gathering facts and routing them to a qualified person.
First-draft correspondence support. For the categories of routine correspondence that follow a consistent pattern — document receipt acknowledgements, conveyancing chain status updates, standard information requests — the AI produces a first draft pulled from the specific matter's details, which a fee-earner reviews, amends if needed, and sends. Every letter and email still goes out under a named solicitor's review; the AI never sends anything to a client unsupervised.
What Changed
The clearest change was where partner time went. Triage that used to consume unpredictable chunks of a partner's day became a five-minute review of an already-routed, already-summarised enquiry. Routine correspondence that used to take fifteen minutes to draft from scratch became a two-minute review of a draft that was already 90% right.
None of this changed how legal decisions got made, and that was the point. The practice didn't want — and shouldn't want — an AI system exercising legal judgement on a client's behalf. What it wanted was its qualified people spending qualified time on qualified work, and the admin layer around that work handled faster.
Where the Line Was Drawn, Deliberately
No AI system in this build gives legal advice, drafts substantive legal documents unsupervised, or communicates with a client without a solicitor's review. That's not overcaution — it's the only responsible way to use AI in a regulated profession where client trust and professional liability are on the line. The value here isn't "AI does legal work." It's "AI removes the administrative friction around legal work, so the people qualified to do that work spend more of their day actually doing it."
The Pattern for Your Own Practice
If you run or manage a legal practice, the honest starting question is: how much of a typical partner's week is genuinely legal judgement, and how much is administrative triage and routine drafting that a well-briefed junior could do if they had the time? AI won't touch the first category. It can meaningfully reduce the second.
If you'd like a straight, no-obligation look at where that line sits in your own practice, a free AI audit is the place to start.