AI for Professional
Services Firms
Senior people are spending half the week on work nobody can bill. The regulator just wrote AI use into the rulebook. We build the workflows that take the admin off the partnership without breaking the duties you sit under.
190 hrs
PER FEE-EARNER, PER YEAR (THOMSON REUTERS)
21.5%
OF BILLABLE HOURS GO UNRECORDED (TRIBES.AI)
9 Mar 2026
RICS AI RULES TURN MANDATORY
Stop bolting ChatGPT onto a partnership.
A consumer chatbot in a browser tab is what 61% of UK organisations already let staff do, per the CIPD's Autumn 2025 Labour Market Outlook. It's also what got a barrister and two solicitors referred to their regulators in Ayinde v Haringey last year.
The other option used to be Harvey or CoCounsel. Two thousand dollars a seat, twenty-five-seat minimum, a sales call before you see the product. Fine for a Magic Circle firm. Not for a fifteen-partner practice in Leeds.
There's a middle path now. That's what we build.
WHAT MOST FIRMS HAVE
- A ChatGPT tab on each partner's laptop
- Client data going into tools nobody signed an agreement with
- No audit trail, no log, no policy
- Knowledge still trapped in three partners' heads
- A Harvey quote that priced you out
WHAT YOU NEED
- Workflows wired into your matter system, not bookmarks
- Models that don't train on your client data, contractually
- Every output sourced, logged, reviewable
- Your firm's precedent and house style baked in
- Priced for an SME firm, not a global one
Five workflows worth building first.
Same pattern whether you're a law firm, an accountancy practice, a consultancy, a surveyor or a recruiter. The ones we see pay for themselves inside a quarter.
Intake and conflicts
New enquiry to engagement letter, with conflicts and KYC checks done first, scoping language drafted, and the matter opened in the practice system. Days of partner time, gone.
First-pass drafting
Advice notes, engagement letters, contracts, reports, valuations. Built off your precedents, in your house style, sourced to the underlying documents. A senior reviews, not writes.
Document review
Disclosure, due diligence, lease bundles, year-end packs, tender responses. A junior used to spend a week. The workflow does it overnight, with every flag traceable back to the page it came from.
Time capture and WIP
A workflow that reads emails, calendar entries and document edits, drafts the timesheet, and asks one question instead of forty. Tribes.ai puts the leakage at 21.5%. This is where it lives.
The firm's brain
The knowledge in your three senior partners' heads, captured as searchable, citable memory. Succession is the part nobody plans for until it's too late. The brain stays even when they don't.
Where we come in.
No twelve-month transformation programme, no McKinsey deck. A short audit, the first workflow live, and a plan to do the rest at your pace.
We've built AI workflows for recruiters, financial advisers, surveyors, accountants and law firms. The mechanics are the same every time. The risk register is the bit that changes.
BOOK A PRACTICE AUDITPractice audit
We sit with two or three of your partners and one of your seniors. Map the day. Pick three workflows worth automating, ranked by hours back and risk. Scope and price up front, fixed-fee, no slides.
Capture the firm's knowledge
Precedents, advice files, model letters, internal manuals, the bits in partners' heads. Pulled into a private knowledge layer your workflows can read. Hosted on infrastructure you can point your regulator at.
Launch the first workflow
Sensors, model, tools, audit log, human review gate. Live, wired into the systems you already use. Your team starts using it straight away. You see what AI can actually do inside a real practice.
Add the rest, sensibly
Next workflow, then the one after. Each one lands with the regulator-shaped paperwork done: risk register, client notice where it's needed, supervision policy, model card. Built once, reused everywhere.
It's already a rule, not a suggestion.
Three things that landed in the last twelve months. If your firm sits under any of these regulators, the policy work isn't optional any more.
Ayinde v Haringey.
Divisional Court referred a barrister to the BSB and two solicitors to the SRA after generative-AI-fabricated case citations reached the bench. Verifying every AI output is now an express professional duty, and the Doughty Street tracker has been logging fresh UK incidents at pace ever since.
AI in surveying, mandatory.
"Responsible use of artificial intelligence in surveying practice" turns from guidance into a regulated requirement on 9 March 2026. Every regulated firm needs an AI risk register, written client notice when AI is used on their matter, and an opt-out route.
First agentic-AI audit rules.
The FRC followed up its June 2025 paper with the first audit-regulator guidance globally on generative and agentic AI in audit. If your accountancy firm does statutory audit work, the documentation bar just moved.
Sources: SRA Risk Outlook on AI in the legal market (2024, refreshed 2025); Bar Council Generative AI Guidance for the Bar (updated November 2025); RICS Responsible Use of AI in Surveying Practice (September 2025); FRC AI in Audit guidance (June 2025) and Generative & Agentic AI in Audit guidance (March 2026); Thomson Reuters Future of Professionals Report 2025; ICAEW Evolution of Mid-Tier Accountancy Firms 2026; MCA Member Survey 2025; CIPD Autumn 2025 Labour Market Outlook; Tribes.ai survey of UK professional services managers; AI Opportunities Action Plan (Matt Clifford, January 2025).
The ones we get asked first.
If AI does the first draft, what stops our juniors learning?
The same thing that's always stopped them: not getting feedback. A workflow that drafts and a senior who reviews properly is a better training setup than a junior who writes and a senior who's too busy to mark it up. The ICAEW 2025 research has the under-25s using AI weekly already. They're learning with it, not instead of it.
Our client data can't leave the building. How does that work?
It doesn't have to. We work with enterprise-tier models from Anthropic, OpenAI and Google under contracts that ban training on your inputs, plus UK-hosted options where the data residency rules require it. Every prompt and output is logged in your tenancy. If your regulator asks who saw what, you can answer.
We looked at Harvey / CoCounsel. Why you instead?
Harvey is priced for global firms: $1,200 to $2,000 per seat, per month, with a twenty-five-seat floor. CoCounsel is friendlier, but still a horizontal product. Either way you're paying for features built for somebody else and adapting your practice around them. We build for the workflow you actually have, against the precedents you actually use, on infrastructure you own.
If AI drafts in minutes, doesn't our hourly billing model break?
It's already breaking. Clio's 2026 UK & Ireland Legal Insights Report has fixed and flat fees at 53% of matters, hourly at 32%. AI moves the question from "how long did this take" to "what was the work worth". Our job is to help you reprice cleanly and not give the saving away in undercut quotes.
What if the AI hallucinates a case or a clause?
It will, sometimes. That's why we don't let it answer from memory. Every output is grounded in documents we've fed it (your precedents, the actual statute, the actual contract) and cited back to the source line. The human review gate is non-optional. Ayinde is what happens when there's no gate.
How long, and how much?
A short practice audit, fixed fee. First workflow live in days from go, scoped and priced before we start. Most clients are paying us less than the Harvey quote they got, and we've built them something that fits.
Who owns what you build?
You do. Source code, prompts, knowledge base, audit logs. Hosted in your cloud account or ours, your call. No lock-in, no per-seat licence per fee-earner, no separate bill if you hire ten more graduates next September.
Run the practice. We'll run the build.
Thirty minutes on a call. Tell us what the partners hate doing and where the seniors are about to retire. You'll get a clear answer on the three workflows worth building first and what the regulator will want to see when it's running.