AI for solicitors and small UK law firms: where the fee-earner hour goes

Most UK law firms with 5 to 50 fee earners have a partner-hour problem more than a billable-hour problem. The work is there. The clients are there. What is missing is partner and senior associate time, because too much of the day goes to document review, inbox triage, conflict checking, and writing the same matter-status update three times to three different clients before the chargeable work starts.

This guide is for the firm in that band. We work with UK professional services firms across legal, accountancy, broking, and surveying, and the same pattern recurs in most of the law firm conversations we have. For many firms the right fix is not another paralegal hire that the margin will not stretch to. It is an automation layer on top of the practice management system the firm already uses, taking the prep work off the partner and putting the partner back on the work that bills. Some firms do still need the hire as well; the two are not mutually exclusive.

This piece walks through where the time actually goes, the four AI workflows that recover the most of it, the SRA boundary every build has to respect, and a five-question diagnostic.

For the wider sector view, our pillar guide on AI for accountants covers the parallel problem in accountancy practices, and our AI for finance brokers post covers the broker side of regulated client work.

Where the partner hour actually goes

When we sit down with a partner at a firm in this band and walk back through their last week, the same illustrative pattern recurs. A long working week of senior time, and a meaningful share of it spent before any chargeable work begins. The rest goes to:

• Reading inbound documents on a new matter (contracts, leases, witness statements, expert reports, client correspondence) to work out what is actually being asked.
• Triaging an inbox that lands a steady stream of new mail every day, most of it routine but a meaningful slice that needs a real reply with care.
• Walking new clients through the intake process: KYC, conflict checks, engagement letter, terms.
• Writing the matter-status updates the firm's clients expect.

None of this is non-essential work. All of it has to happen. But very little of it requires the partner specifically. It requires somebody who knows the matter and can write clearly. AI does not replace that judgement, but it can take the first draft of every part of it off the partner's desk so the partner spends the senior hour on the senior work.

The four workflows that recover the most fee-earner time

Same four workflows recurring across the firms we work with.

Workflow one: document review and summarisation. A long-context model reads inbound documents on a matter, extracts the key facts (parties, dates, monetary amounts, deadlines, governing law, key clauses), and produces a structured summary the fee-earner uses as orientation rather than as analysis. In our experience, the orientation reduces the time the fee-earner spends working out what the document is asking before they engage with the substance. The fee-earner still reads the document for anything that matters legally.

Workflow two: email triage and reply drafting. Inbound mail is classified by matter and by urgency, routed to the right fee-earner, and tagged with a draft reply where one is plausible. The fee-earner reads the draft, edits it, and sends. In our experience, editing a sensible draft is materially faster than writing the same reply from cold; across a partner's day, the time saved compounds.

Workflow three: client intake and engagement-letter assembly. A new-enquiry workflow captures the prospective client's details, runs a first-pass match against the existing matter and contact list (a structured screen, not a final conflicts check), drafts the engagement letter from the firm's templates, and books the intake call. The relevant partner runs the actual conflicts analysis (related parties, beneficial owners, prior advice, sector context) and signs off the engagement. The AI does the assembly and the structured screen; the partner does the conflicts call.

Workflow four: matter-status reporting. A weekly or fortnightly status update is drafted from the matter file (recent activity, upcoming deadlines, outstanding tasks, current cost-to-date) and sent to the fee-earner for review and dispatch. Clients get a consistent rhythm of communication; the fee-earner spends less senior time on what is essentially structured assembly.

One practical note on messaging compliance: routine matter correspondence on an active engagement is generally treated as service communication rather than direct marketing, but the analysis depends on what is in the message, who the recipient is, and which channel it goes through. Anything that crosses into promotion (a newsletter, a sector update, a client-event invitation) is subject to PECR (the Privacy and Electronic Communications Regulations); for electronic mail to individual subscribers the default rule under PECR is prior consent, with the soft opt-in available in some return-customer scenarios, and UK GDPR sits alongside. Rules differ for corporate subscribers. We separate the two streams so the matter mail and the promotional layer are not muddled, but the underlying confidentiality, security, accuracy, and UK GDPR obligations on matter mail still need to be designed in.

Where AI must stop: the SRA boundary

The SRA does not, as of 2026, publish a single "AI for solicitors" rulebook. The relevant constraints sit across the SRA Principles, the Code of Conduct for Solicitors, RELs and RFLs, the Code of Conduct for Firms, and the supervision and competence duties on partners and COLPs. In practice, those add up to a clear boundary.

First, a qualified solicitor remains accountable for the legal advice. AI can draft a clause, summarise a contract, or assemble an engagement letter. It cannot give legal advice in its own name. Anything that crosses into substantive advice or that reaches a client as the firm's position should be reviewed by a fee-earner before it goes out; routine administrative correspondence and internal-use drafts can be supervised on a risk-based basis the firm sets in its policy.

Second, client confidentiality and personal-data handling are the operational constraints on what can go into the model. Client documents and personal data should not be sent to consumer-tier AI tools whose standard terms allow the provider to train on inputs. We typically set up the AI workflow on a paid business or enterprise tier with contractual no-training terms, or on a UK-hosted alternative. The contractual tier is necessary but not sufficient: the firm still needs to think about lawful basis under UK GDPR, processor terms, security, transfer analysis, retention, and the internal controls that sit around the workflow. The data-protection position is reviewed at firm level before client data goes near the model.

Third, supervision and competence apply to AI work in the same way they apply to other delegated work. The duty sits in the SRA Code of Conduct for Firms (the supervision rule, currently rule 8.1, which expects effective supervision of work) and in the competence requirements on fee-earners. The COLP supports the firm-wide compliance position; in practice, the supervising partner on a matter is the natural owner of the workflows that touch that matter. The audit trail (model used, prompt, input documents, output, fee-earner who approved) is the artefact that supports the supervision position.

Fourth, on hallucination risk for case law and authority specifically: published incidents (the US lawyers who cited fabricated cases is the most-quoted example) generally happened in setups where the model was asked to recall authority from memory rather than retrieve it from a verified source. We design these builds so the workflow does not generate an authority citation from the model's memory; citations come from a verified source the firm trusts (Westlaw, LexisNexis, the firm's own knowledge bank, or the legislation database) or are flagged for the fee-earner to add manually. That separation is the load-bearing part of the design.

The tool stack we typically build on

The build sits on top of what the firm already runs. For most UK 5 to 50-fee-earner firms, that is:

• Practice management. Clio, Actionstep, LEAP, Tikit, PracticeEvolve, or a sector-specific PMS. Whichever the firm uses stays in place.
• Document storage and OCR. NetDocuments, iManage, SharePoint, or the document module of the PMS.
• Workflow layer. An automation platform that connects the existing systems and sequences the steps.
• Language model. A long-context model on a paid business or enterprise tier with contractual no-training terms, or a UK-hosted alternative if the firm has a residency requirement.
• Audit log. Every AI run logged: model, prompt, input file references, output, fee-earner who approved. This is the supervision trail under SRA Principle 7 and the Code of Conduct on competence.

For the head-to-head on workflow platforms specifically, our Make.com vs Zapier vs n8n guide covers the choice in detail.

A five-question diagnostic

If you run a UK law firm in the 5 to 50-fee-earner band and you want to know where AI fits, ask yourself these five questions.

One. What share of your partners' time this month went to document review, inbox triage, intake, and matter-status writing rather than to chargeable work? If you do not have a number, that is the first answer.

Two. Do you have a practice management system that is the single source of truth for matters, contacts, and documents, or is the matter file scattered across the PMS, email, and personal drives?

Three. What is your current position on confidentiality and data protection when AI tools are involved? Have you set out which tools are approved, on which tiers, and which client data can pass through them?

Four. Who in the firm would own the AI rollout: a partner, the COLP, the practice manager, or somebody else? Do they have a few hours a month to review the workflows and the audit trail?

Five. When was the last time the partners agreed in writing on what "good" looks like for client communication on a routine matter (cadence, tone, level of detail)? AI is most useful when there is a written standard to assemble against.

If two or more of those questions made you pause, the AI Opportunity Audit is a free 30-minute call where we map where AI fits in your specific firm. We will be honest about whether you are at the do-it-yourself stage or the consultant-build stage.

If the diagnostic raised a flag

If you cannot answer those five questions cleanly, the issue is usually one of three things: the firm has no clear picture of where the senior hour is actually going, the matter file is too scattered for the AI to read in one place, or the data-protection position has not been agreed at partner level so nobody knows which tools are sanctioned.

Each of those is a plausible first build. The question is which one is leaking the most senior time in your specific firm.

If you want a 30-minute conversation about where that system should start, book a free audit. We will tell you whether the first move is document review, inbox triage, intake, or fixing the data-protection position before any AI work begins.