AI for law firms
Most conversations about AI for law firms are about research and case law, where the tools are strong and the professional risk is loudest. Ask a managing partner where the money goes and the answer is different: enquiries that were never followed up, engagement letters that took four days to go out, first drafts assembled from precedent by someone billing at a partner's rate, and bills raised late because nobody had time to write the narrative.
Four leaks, none of them exotic
- Intake dies at the first hurdle. An enquiry arrives at 18:00, nobody replies until Tuesday, and the client has instructed someone else. Conversion in professional services is mostly a speed problem wearing a quality costume.
- Conflicts and onboarding take days. The checks are necessary. The four days between the client saying yes and the file opening are not, and they are where the enthusiasm dies.
- First drafts start from a folder of old matters. Someone senior finds the closest precedent and edits it. The finding is not legal work. The editing is.
- Time recorded from memory on Friday. Every firm knows what that costs and every firm does it. Reconstructed time is under-recorded time, reliably, by a margin that would embarrass most billing partners.
The lines that do not move
- No legal advice reaches a client without a qualified person having read it. Not as policy, as plumbing: the path from machine to client passes through a fee earner.
- No citation gets used without verification. Fabricated authorities have ended careers, and the fix is checking every reference against a real source, mechanically, every time.
- Client confidentiality decides the architecture: EU hosting, no training on your matters, and each automation reaching only the files it needs.
- Conflicts stay a human decision. The sweep surfaces candidates; a person clears them.
Where the capacity actually comes from
- Intake that answers immediately, captures the matter in the enquirer's own words, runs a first conflicts sweep and books the call — at 18:00 on a Friday as readily as at 10:00 on a Tuesday.
- Onboarding assembled from what intake captured: engagement letter, identity checks, matter opened, all drafted for a fee earner to check rather than typed.
- First drafts from your own precedent bank, with the reasoning visible: which clause came from where, and what was changed for this matter.
- Time capture from what actually happened — calls, documents, emails — as a draft narrative for the fee earner to correct on the day rather than reconstruct on Friday.
Not yet, if
Three conditions, and the first is the common one. If your precedents live in individual fee earners' folders rather than in a shared bank somebody keeps current, that is the first project and it is worth doing whether or not anything is automated afterwards. If you take fewer than a handful of new enquiries a week, the intake gain has little to scale against. And if nobody in the firm will own the drafts a machine produces, do not start: a system that produces work nobody checks produces liability, not capacity.
The same shapes, already in production
- Text factory — long-form documents produced end to end by machine and published for real — with the human check-and-approve step kept in the loop on purpose, which is exactly the shape a firm needs.
- Returns & defects handler — rules applied consistently the moment something arrives, with one batch to sign off each morning. Intake and conflicts want that shape: immediate handling, deliberate approval.
What partners ask
Is AI for law firms safe to use on client matters?
It is when the architecture is right: hosting in the EU, a provider that does not train on your data, narrow access per automation, and a qualified person between any output and any client. Those are decisions made at the start, not features added later.
What about hallucinated case law?
Every citation is checked against a real source before it can appear in a draft, mechanically. A drafting tool that cannot show you where a reference came from should not be inside a law firm.
Will it draft advice?
It assembles a first draft from your own precedent and the matter facts, and shows what came from where. It does not decide what the client should do, and nothing reaches a client without a fee earner having read it.
Which piece pays back first?
Usually intake, because the loss is measurable — count the enquiries that got no reply within a day last month. Time capture is often second, because under-recording is a bigger number than most firms want to look at.
We are four people. Is this only for large firms?
A small firm feels intake speed more sharply, not less: there is nobody to cover the enquiry that arrives while everyone is in court. The build is smaller in proportion.
Bring last month's enquiry log
How many came in, how many got a reply the same day, and how many became matters. If the gap between the first two numbers is small, I will tell you the intake build is not your problem and where to look instead. That hour costs nothing.
Book that hour →