It's 19:30 on a Tuesday in a 12-lawyer firm somewhere in Manchester or Leeds. The last fee earner left an hour ago. The partner running the firm has a pile of paper file notes on his desk, two dozen unposted time entries, and a reminder to review the draft engagement letter for tomorrow's new-client meeting. The PMS has an "AI assistant" icon in the top right. He's tried it twice. Both times it offered to "summarise this matter" in a way that sounded like a LinkedIn post.
On the wall outside, a glossy insert from a legal tech conference invites him to a session on "AI transformation for mid-market practices." The session is in London. He is not going.
This is the firm I build for. Two to twenty partners, UK-based, running on Clio or Actionstep or LEAP or DPS, losing fee-earner hours every week to admin the software was supposed to eat. This guide is what I'd want that partner to read before he buys another "AI for law firms" pitch.
The short answer
AI is genuinely useful in UK law firms in 2026, but most of the value is in narrow workflows that the big legal-AI vendors aren't selling: file note generation, time capture, matter intake, precedent drafting, disbursement matching. The work sits outside the practice management system, not inside it. For a small UK firm (two to twenty partners), expect to save 5-12 hours per fee earner per week once built properly, and to pay between £30-£80 per fee earner per month for PMS-bundled AI or £12,500 upwards for a custom automation layer that targets one or two high-volume workflows end to end.
Why AI is stalled in most small UK law firms
AI was supposed to arrive in legal at enterprise scale first. That mostly happened, and the magic-circle firms have teams of twenty specialists running on Harvey and Legora. The effect on a 14-lawyer firm in a regional market is almost zero. Three reasons.
The cultural gap is wider than the technical one.
The technical problems in legal AI are now genuinely tractable. Claude and GPT handle the majority of document-level tasks well. Retrieval against a firm's own knowledge base is a solved engineering problem. The blocker is that partners who trained in the 1990s and 2000s built their practice on specific ways of drafting, reviewing, and recording work that don't yield to general-purpose AI without careful tuning. Nobody inside the firm wants to be the one who signs off on a workflow change that later generates an SRA complaint. So the default is to do nothing and wait for the PMS vendor to ship something safe.
The PMS vendors are drip-feeding AI features carefully. Clio Duo, Actionstep AI, LEAP AI: the feature sets are growing but deliberately narrow. The vendor does not want to be blamed for a hallucinated clause or a file note that misquotes a client. The result is features that are genuinely useful for note-taking and summarisation but do not touch the deeper admin burden. The most valuable automations in a small firm (matter intake, client onboarding, precedent drafting, automatic disbursement matching) are not on the PMS roadmap because the PMS cannot automate across tools it doesn't control.
The SRA and PI insurance create a justified caution. Solicitors have strict duties around confidentiality, competence, and supervision of output. A partner reading the Law Society's "Generative AI: the essentials" comes away with a sensible but long list of things to think about. PI insurers are starting to add AI-specific questions to renewal forms. Any plan that moves confidential material into a US consumer AI service is dead on arrival. Any plan that doesn't have clear human review gates is dead on arrival. Most off-the-shelf AI tools for law firms optimise for something other than these constraints, and partners can smell it.
The firms that get this right stop waiting for the PMS roadmap and build a narrow automation layer alongside their existing stack, with SRA constraints built into the architecture from day one.
What's actually worth automating in a 2-20 partner UK firm
Every legal tech listicle will tell you to automate everything. That's vendor copy. In practice some automations compound and some waste build hours. Here's what I've seen actually work in small UK firms, in rough order of ROI per workflow.
1. Client intake and matter opening
A new client enquiry arrives by email or phone. The current process: fee earner takes details, sends an engagement letter drafted from a Word template, runs AML checks through a separate service, emails the client for ID documents, chases, eventually opens a matter in the PMS. A decent workflow collapses this into: structured intake form (web or email) → automated AML submission to your provider → engagement letter drafted from template with matter-specific variables filled by an LLM → matter opened in PMS with client and counterparty records populated → ID document collection link sent to client.
Time saved: 45-90 minutes per new matter. Multiply by the 8-20 new matters a small firm opens per week.
Build complexity: medium to high. The AML integration is the hardest part. The LLM work (drafting the engagement letter paragraph specific to the matter) is the easiest.
2. File note generation from calls
A fee earner finishes a 20-minute client call. They currently spend 8-15 minutes typing a file note, paraphrasing what was said, adding context, saving into the PMS. A workflow that takes the call recording (from a voice memo app, an Otter-style transcription tool, or a softphone with recording), transcribes it, summarises the legally relevant points into the firm's house style, and writes the note as a structured activity in the PMS replaces that admin with a background job. The fee earner reviews and approves rather than writes from scratch.
Time saved: 6-10 minutes per call. Many fee earners do 6-12 client calls a day.
Build complexity: low to medium. Transcription is commoditised. The quality work is in the summarisation prompt and the PMS write-back.
3. Time capture from calendar and email
The chronic admin pain in every small firm. Time entries submitted at week's end from memory, guessed at, under-captured. A workflow that watches your calendar and your sent email, clusters activities by matter, proposes draft time entries in six-minute increments, and waits for fee-earner approval recovers 10-30% of previously-unbilled time for most firms I've talked to.
Time saved: 30-60 minutes per fee earner per week, plus materially higher billable capture.
Build complexity: medium. Calendar and email APIs are straightforward. The matter-matching is the interesting problem.
4. Precedent and first-draft document generation
A new NDA, a straightforward employment contract variation, a standard engagement letter. The partner or senior fee earner pulls a previous version of the document, edits it for the new fact pattern, reviews, sends. A workflow that reads the structured inputs (client, counterparty, key terms, jurisdiction), selects the right precedent, fills in the variables with LLM reasoning where needed, and returns a review-ready first draft saves the repetitive 40-70% of that work.
Time saved: 20-45 minutes per document. A transactional fee earner may draft five to ten of these a week.
Build complexity: low to medium. The precedent library needs structuring once. Everything after is variation on a template.
5. Disbursement matching
Every matter accrues disbursements: search fees, court fees, expert reports, travel. They arrive as emails, PDFs, supplier invoices. Ops or a fee earner manually matches each to a matter and posts to the PMS ledger. A workflow that reads the incoming supplier email, extracts the matter reference, and creates the disbursement entry cuts that work to exception review only.
Time saved: 2-5 hours per week of ops time depending on firm throughput.
Build complexity: low. Email parsing plus PMS write-back.
What I wouldn't automate first
Drafting court documents. The risk of hallucination or procedural error is real and the consequences are career-shaping. LLMs are not good enough yet to be trusted unsupervised for court filings, and human review erases most of the time savings.
Legal research for advice outputs. Claude and GPT are better in 2026 than they were in 2024 but still confidently wrong on UK case law more often than is acceptable. If you must use AI for research, use a specialist vendor (Lexis+ AI, Vincent AI, or equivalent) with proper citation grounding, not a general-purpose LLM.
AML for high-risk clients. The SRA is actively enforcing in this area. AML work should be mediated by a compliance professional using an approved tool. Automate the admin around it (document collection, reminder chasing) but not the risk assessment itself.
Client-facing email without a human gate. The risk of tone or content drift is higher than the time savings.
The typical stack in 2026
A working AI and automation stack in a small UK law firm has four layers.
Layer 1: the practice management system. Clio, Actionstep, LEAP, DPS, Quill, Linetime, or similar. The system of record for matters, contacts, documents, time, and billing. Clio and Actionstep both expose usable REST APIs. LEAP and DPS are more closed and automation usually has to interop through email, CSV, or scheduled exports.
Layer 2: the workflow orchestration layer. This is where the automation logic lives. Options: Zapier or Make at the low end, n8n self-hosted for more control, or custom code on Trigger.dev or a Python service for the cases where neither Zapier nor n8n handles the fan-out properly. For most small firms the orchestration is the piece that determines how well the whole system works, and it's the layer vendors don't talk about.
Layer 3: the data and compliance services. Companies House API (free, indispensable), Land Registry, OGP, your AML provider's API (Smartsearch, Veriphy, Credas), and whichever search providers you use. A firm doing commercial property work will lean heavily on Land Registry and Companies House. A firm doing private client will lean on Smartsearch and HMRC.
Layer 4: the LLM decision layer. Claude (Anthropic) or GPT (OpenAI) called over API from the orchestration layer. One or two LLM calls per workflow is typical. Use API tier, not consumer apps. Confirm your data is not used for training (it isn't, by default, on Anthropic and OpenAI API tiers, but document it in your DPIA).
The interesting question is what you use at layer 2. This is where agency archetype matters.
Who builds this for you (and who pretends to)
I wrote a longer version of this framing in my AI automation agency buyer's guide. The short version, applied to legal specifically:
- The Zapier or n8n reseller will quote you £800-£5,000 per workflow built in their Zapier account. Usable for the simplest automations (intake form to PMS, email-trigger reminders) but breaks on anything involving AML integration, document drafting quality control, or the fan-out logic of one matter across several fee-earner pipelines.
- The AI-labelled consultancy will sell you a £12,000-£30,000 "AI in legal" strategy engagement. The deck is usually competent and cites the Law Society guidance. The implementation is either out of scope or subcontracted.
- The engineer-led custom shop will quote £12,500-£30,000 to build a narrow automation layer that reads and writes to your PMS over its API, runs on your own AWS or Render account, and is documented well enough that a future engineer can maintain it. You own the code and the process.
Laires Labs sits in the third category. The build floor is £12,500 because smaller builds don't cover scoping, SRA-aware design, build, handoff, and the two weeks of post-launch support a real legal automation needs. Smaller jobs are better served by the reseller tier or by PMS-bundled AI features.
Real costs: SaaS vs custom build over 2 years
Same 8-fee-earner UK firm, same desired automation (the first three workflows on the list above: intake, file notes, time capture). Here's what each path costs across two years.
SaaS-plus-reseller path
- Clio Grow and Clio Manage (8 fee earners): £90/seat/month × 8 × 24 = £17,280
- Clio Duo AI add-on: £30/seat/month × 8 × 24 = £5,760
- AML integration (Smartsearch): £150/month × 24 = £3,600
- Zapier Professional + task top-ups: £200/month × 24 = £4,800
- Call transcription tool (Otter Business or equivalent): £100/month × 24 = £2,400
- LLM API costs (moderate): £150/month × 24 = £3,600
- Setup by a Zapier legal-tech reseller: £8,000 one-off
- Annual tweaks: £3,000/year × 2 = £6,000
2-year total: £51,440. Of which £37,440 is recurring seat, subscription, and API fees, £8,000 is build, £6,000 is tweaks. Net incremental automation layer cost (above what you'd already be paying for the PMS): roughly £14,000.
Engineer-led custom build path
- Clio Grow and Clio Manage (8 fee earners): £90/seat/month × 8 × 24 = £17,280 (same, the PMS stays)
- AML integration (API direct to Smartsearch): £150/month × 24 = £3,600 (same)
- Render or AWS hosting for the automation layer: £40/month × 24 = £960
- LLM API costs: £200/month × 24 = £4,800
- Custom build by engineer-led shop: £18,000 one-off
- Retained maintenance (ad-hoc, estimated): £3,000-£6,000 across 2 years
2-year total: £47,640-£50,640. Savings vs SaaS path: around £1,000-£4,000 on an 8-fee-earner firm over 2 years.
The numbers are closer than in the recruitment equivalent, because the PMS seat fees dominate both paths. The custom path's real advantage shows later, as the firm grows. Seat fees scale linearly with headcount. A custom automation layer's marginal cost of supporting a tenth fee earner is roughly zero. A 16-fee-earner firm paying for 8 extra Clio Duo seats, 8 extra Otter Business seats, and double the Zapier task volume starts seeing five-figure annual gaps.
The other argument for custom is control. PMS-bundled AI does what the vendor decides. A custom layer does what you decide, in the order you decide. For firms where a specific workflow is central (a firm that drafts 40 engagement letters a week, say, or a probate specialist opening 30 matters a month), that control compounds.
SRA, PI, and the compliance questions you'll actually face
Every automation in a law firm has to survive three questions. A good build answers all three in writing before anyone signs off.
Where does the data live and who can see it? Client-confidential information can't sit on a US consumer AI service. On the API tier, Anthropic and OpenAI default to not training on your data, but the agreements differ slightly. Your DPIA needs to reflect the actual data flow: where the prompt goes, where the response is stored, how long it's retained, and what happens on deletion. UK or EU regions where possible.
Who is accountable for the output? SRA supervision rules require that a qualified solicitor signs off on anything reaching a client. An automation can draft, summarise, and file, but it cannot send work product to a client unsupervised. Build the human review gate in. The workflow that skips the gate is the workflow that ends up in a complaint.
What would your PI insurer ask? At 2026 renewal most PI insurers are adding AI-specific questions: are you using generative AI on client matters, what categories of work, what supervision, what data handling. A firm that can answer those questions clearly from day one tends to get better renewal terms than a firm that can't.
Book a call
Twenty minutes, no pitch. Tell me your PMS, your firm size, and the two admin burdens your fee earners complain about most. I'll come back with a scoped build plan: what I'd build, how I'd phase it, what it would cost, and which of the workflows above I'd park for your specific firm. You keep the plan whether you hire me or not.
Don't book if you're under three fee earners (the maths doesn't work), if you want a "legal AI transformation" strategy deck, or if your managing partner isn't bought in. The cultural gap is the biggest blocker and the one I can't fix.
Book a 20-minute call or read the full AI automation agency buyer's guide if you're still deciding between archetypes.
Thinking about a system like this?
20-minute call, no slides. We'll map it against your stack and I'll tell you if it's worth building.
Book a callQuestions buyers actually ask
Any workflow in a law firm where a step that used to be manual (drafting a first-pass document, summarising a call, extracting clauses from a contract, generating a file note) is handled by a large language model like Claude or GPT. In 2026 most practical AI in UK firms sits alongside the practice management system rather than inside it, because PMS-native AI features are still narrow and conservatively rolled out.
Robin Laires
Ten years software engineering, former tech lead at Jellyfish — one of the UK's largest independent digital agencies. Now I build custom AI systems that replace manual business processes: ads ops, sales intelligence, intake routing, research pipelines. One engineer, installed into your stack.