AI for UK law firms, built the way the SRA now expects.
Augustova builds AI agents and automation for small and mid-sized UK law firms: client intake and triage, AML onboarding, conflict screening, matter status, document chasing and conveyancing requisition checks. Each is designed around the SRA’s August 2026 warning notice: a solicitor stays accountable, client data never enters a public tool, and every output is logged. Fixed price, £4,000 to £15,000 for a first agent.
What the regulator has actually said, and what it means for a build
On 17 August 2026 the Solicitors Regulation Authority issued a warning notice on the misuse of AI. Its two concerns are the ones every partner already worries about: fabricated citations reaching a court, and confidential client information typed into tools that keep it. Its position is blunt. "AI has no separate legal personality; solicitors and regulated individuals who use AI in the course of delivering legal services remain accountable for their work and outputs." Reliance on an AI output, it adds, "would not be a suitable defence". The Law Society Gazette reported that the SRA had received 42 reports of AI-related wrongdoing in the year to July 2026.
The courts got there first. In Ayinde and Al-Haroun, handed down in June 2025, the Divisional Court found that in one case eighteen of forty-five citations put before it did not exist, referred two solicitors to the SRA, and set the duty in plain words: those who use AI for legal research "have a professional duty therefore to check the accuracy of such research by reference to authoritative sources". The Upper Tribunal went further in November 2025: uploading confidential documents into an open tool such as ChatGPT "is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege". The judiciary’s own guidance says anything typed into a public chatbot "should be seen as being published to all the world".
None of this is a reason not to build. In May 2025 the same regulator authorised the first purely AI-based law firm, on the condition that named solicitors remain accountable for every output and that hallucination monitoring, confidentiality and conflict controls are in place. That is the design brief. The table sets each rule against what it requires of a system, and every agent on this page is built to it.
| The rule | Where it comes from | What the build must do |
|---|---|---|
| A solicitor remains accountable for every output | SRA warning notice, Aug 2026 | No agent acts on a client matter without a named fee earner approving; the approval is logged with who, what and when |
| Check AI research against authoritative sources | Ayinde [2025] EWHC 1383, para 7 | Agents that draft or summarise cite the source document they drew from, never a model’s memory; anything that looks like a citation is flagged for verification |
| Confidential data in a public tool waives privilege | Upper Tribunal [2026] UKUT 00081, para 60 | Client data runs only through accounts the firm owns, under a contract that forbids training on it, in a UK region; no free consumer tools anywhere in the workflow |
| Do not feed confidential information into tools you do not control | Law Society, May 2025 | The firm holds the provider accounts and the keys; we hold access under a written processing agreement and lose it at handover |
| Managers must keep effective supervision systems | SRA warning notice, Aug 2026 | Every agent has an audit log a supervisor can read, an error rate measured on the firm’s own files, and a person who owns it |
| Document inputs and outputs; keep a review mechanism | Law Society checklist, May 2025 | Inputs, outputs, model version and reviewer are recorded per matter, exportable for a file review or an insurer |
| eIDVA is not a substitute for a client and matter risk assessment | SRA AML report 2024-25 | Electronic verification is an input to an onboarding agent, never its decision; the risk assessment is a solicitor’s, with the evidence assembled for them |
| Consumers expect accuracy, informed consent, human oversight and redress | Legal Services Board, Jun 2026 | Client-facing agents say they are automated, offer a person, and record consent |
How much AI UK firms use now, and how little of it is embedded
The headline numbers say adoption is nearly universal; the detail says it is shallow. LexisNexis found 61% of UK lawyers using generative AI for work by August 2025, up from 46% in January, but only 11% using it heavily, and 67% describing their firm’s AI culture as slow or non-existent. Clio’s survey of 500 UK solicitors found 96% of firms integrating AI in some form and 53% with no AI policy, or unaware of one. In other words, most firms have staff using tools, few firms have a system they control, and fewer still have written down what is allowed.
What the tools are used for is telling. Document drafting leads at 36%, contract review 29%, general non-legal tools 24%, e-discovery 20% and legal research 17%. The tasks that actually consume a small firm’s non-chargeable hours, onboarding, chasing, status updates and compliance file work, barely feature, because a chat window cannot do them. They need an agent connected to the practice management system, which is the work this page describes.
| Measure | Figure | Source, date |
|---|---|---|
| UK lawyers using generative AI for work | 61%, from 46% in January 2025; 6% have no plans | LexisNexis UK, Sep 2025 (700+ lawyers) |
| Using it heavily day to day | 11% overall; 7% at large and medium firms | LexisNexis UK, Sep 2025 |
| AI fully embedded in strategy and operations | 17%; 67% say the firm’s AI culture is slow or non-existent | LexisNexis UK, Sep 2025 |
| Firms integrating AI in some form | 96%; 62% of solicitors plan to expand use | Clio UK, Nov 2024 (500 solicitors) |
| Firms with no AI policy, or unaware of one | 53% | Clio UK, Nov 2024 |
| Tasks | Drafting 36%; contract review 29%; general tools 24%; e-discovery 20%; research 17% | Clio UK, Nov 2024 |
| Expect AI to significantly change the profession within five years | 87% of UK lawyers; in-house lawyers use AI at about twice the rate of firms | Thomson Reuters Institute, Sep 2025 (172 UK) |
| Would consider leaving a firm that does not invest in AI | 18% of private-practice lawyers; 26% at large firms | LexisNexis UK, Sep 2025 |
Where a small firm loses money: the enquiry, not the drafting
The best-evidenced leak in a UK law firm is the first contact. A mystery-shopping exercise across 100 personal injury and wills practices found that one in five wills enquiries by phone got no callback at all, and 16% of web enquiries received nothing. An earlier exercise found only 23% of firms made any follow-up call. Moneypenny’s overflow data put one in ten calls to a law firm as a brand new enquiry. Meanwhile the Legal Services Consumer Panel finds 44% of consumers now shopping around before instructing, comparing on price first (56%) and reputation second (48%).
Put those together and the arithmetic is uncomfortable: a firm that lets a fifth of enquiries go unanswered is handing them to whichever competitor replied. The second leak is the "where are we" call, which in conveyancing and family work interrupts fee earning to deliver information the firm already holds. The third is the compliance file: the SRA reviewed 5,873 files in 2024-25 and rated almost a third of firms not compliant, with a missing source-of-funds check the second most common failure. All three are process problems, and processes are what agents fix.
| Measure | Figure | Source, date |
|---|---|---|
| Wills enquiries by phone receiving no callback | 20%; 16% of web enquiries received nothing | First4Lawyers and insight6 mystery shop, via Legal Futures, Sep 2023 |
| Firms making a follow-up call after an enquiry | 23% of 40 firms | Concert Networks mystery shop, via Legal Futures, Jun 2017 |
| Calls to a law firm that are new enquiries | About 1 in 10 (10.4%) | Moneypenny overflow data, via Legal Futures, 2015 |
| Consumers shopping around before choosing | 44%; price the top comparison factor at 56%, reputation 48% | Legal Services Consumer Panel, Jul 2025 (3,750 users) |
| Clients’ preferred contact channels | Phone 45%; firm website 43%; chatbots 42%; email 38% | Clio UK, Nov 2024 |
| AML compliance ratings | Of 833 rated firms, 32.4% not compliant, 451 partially compliant | SRA AML report 2024-25, Oct 2025 |
| Most common AML failures reported | No client or matter risk assessment (162); no source of funds check (101); inadequate policies (99) | SRA AML report 2024-25 |
| Files lacking a source-of-funds check | 11% of 5,026 files requiring one; 18% showed inadequate scrutiny | SRA thematic review, Nov 2025 |
| Fully avoidable Land Registry requisitions | 4.6% average; 4.9% (78,894 of 1.6m) Oct 2025 to Mar 2026; 17% of firms at zero | HM Land Registry, Nov 2025; Today’s Conveyancer, Jun 2026 |
Six agents we build for firms, and the control each one carries
Each of these is a single agent connected to the practice management system the firm already runs, built in two to four weeks at a fixed price, with a person in the loop wherever a client, a court or the regulator could be affected. They are listed in the order most firms should build them, which is the order of the money leaking.
What it costs, against what the market charges
A first agent, one of the six above, integrated with your PMS and in production, is £4,000 to £15,000. A system of several agents sharing data, for example intake feeding onboarding feeding conflict screening, is £8,000 to £25,000. An AI adoption audit, if the firm does not yet know which leak is largest, is £2,500 to £6,000, credited in full against any build. Every price is fixed against a written scope with the exclusions listed, and the monthly running cost is stated before go-live. We do not sell day rates.
For context, the specialist market prices in three ways. Microsoft-partner consultancies sell a discovery phase and then a retainer: FiveForward publishes a £1,500 fixed first step and an "embed" retainer from £750 a month with a three-month minimum. Legal IT providers quote ranges rather than prices: Quiss puts agent implementations at £10,000 to £100,000 depending on firm size. The legal-tech vendors, OneAdvanced’s compliance agents, Lexis+ AI, CoCounsel and Clio’s add-ons, publish no UK prices at all. Khiliad Legal, Nexian and BCN offer assessments and pilots without published figures.
| What | Augustova | Published elsewhere |
|---|---|---|
| Find the biggest leak first | AI adoption audit, £2,500 to £6,000, one to two weeks, credited against a build | FiveForward discovery £1,500 fixed, credited once follow-on spend passes £3,500; Nexian and BCN readiness assessments, price on request |
| One agent in production | £4,000 to £15,000 fixed, two to four weeks | Quiss indicative range £10,000 to £100,000; Khiliad four-week pilot build, price on request |
| Several agents sharing data | £8,000 to £25,000 fixed | No published equivalent |
| Ongoing support | A separate monthly line agreed at go-live; you can stop it and keep the system | FiveForward from £750 a month, three-month minimum |
| Product subscriptions | Not what we sell; we integrate with what you have | OneAdvanced, Lexis+ AI, CoCounsel, Clio Duo: no UK prices published |
Competitor figures read on their pages in September 2026; prices exclude VAT where stated.
How client confidentiality and privilege are protected in practice
Confidentiality is the question every partner asks first, and it deserves an engineering answer rather than a reassurance. This is the position on every legal build, written into the scope before any client data moves.
What to measure before and after
Non-chargeable hours per fee earner per week is the number partners care about and it is almost always estimated. Measure it for a fortnight before anything is built. Then time from enquiry to engagement letter, which is where instructions are lost quietly, and the share of enquiries that receive a same-day response. For onboarding, days from instruction to a completed risk assessment. For conveyancing, the firm’s own requisition rate, which HM Land Registry publishes by customer. For any document work, the error rate on a sample checked by hand; an automation with no measured error rate is a liability in a regulated practice, and we will not put one live.
Common questions
Can a law firm use AI agents without breaching SRA rules?
Yes, and the SRA has authorised an entirely AI-based firm on exactly these terms. The conditions are that a named solicitor remains accountable for every output, that confidential information only enters tools with contractual and technical safeguards against unauthorised access or training, and that supervision is effective. Every agent we build carries an approval step, a locked-down data path and an audit log for that reason.
Is it safe to put client documents into ChatGPT?
No. The Upper Tribunal held in November 2025 that uploading confidential documents into an open tool such as ChatGPT places them in the public domain, breaching confidentiality and waiving privilege. The Law Society’s guidance is not to put confidential data into any free online service. Our builds use business accounts the firm owns, under terms that exclude training, in a UK region.
What happens if the AI makes something up?
The court has said reliance on an AI output is not a defence, so the system is designed so that a fabrication cannot reach a client or a court unseen. Agents that draft or summarise cite the document they drew from; anything resembling a legal citation is flagged for verification; a fee earner approves before anything leaves the firm; and the error rate is measured on your own files before go-live.
Can an agent do our AML checks?
It can do the collecting, verifying and assembling, which is most of the time. It cannot make the client and matter risk assessment, because the SRA is explicit that electronic verification is not a substitute for it. The agent gathers identity, source-of-funds evidence and the risk factors and presents a draft assessment for a solicitor to complete and sign, with the evidence attached.
Which practice management systems do you integrate with?
Any system with an API or a reliable export: the common UK ones for small and mid-sized firms all qualify. Where a direct integration is not available we work from exports, which is still enough to automate chasing, status and compliance summaries. We confirm the integration path in the scope before quoting.
How much does an AI agent cost for a law firm?
A first agent, integrated with your PMS and in production, is £4,000 to £15,000 fixed. Several agents sharing data are £8,000 to £25,000. An audit to decide which to build first is £2,500 to £6,000 and is credited against the build. Running costs are stated per month before go-live. The market’s published alternatives run from a £1,500 discovery plus a retainer from £750 a month to indicative implementation ranges of £10,000 to £100,000.
Will this satisfy our professional indemnity insurer?
Insurers ask for the evidence, not the marketing: what data goes where, who can see it, how outputs are checked, and what is logged. All of that is produced by the build and exportable. Where an insurer or the regulator requires a certification rather than engineering evidence, we say so and do not imply otherwise.
Do you handle client phone calls for law firms?
Yes, as the intake agent: it answers at any hour, captures the matter and books the consultation into your diary, states that it is automated and offers a person at every turn. It never gives legal advice. For firms that only need the phone answered and booked, an off-the-shelf AI receptionist may be enough; our receptionist comparison sets out the options and prices.
Tell us which leak is largest.
A 30-minute call with a founder, not a salesperson. Bring the number of new enquiries a week, the practice management system you run and the compliance finding that worries you most. You will leave with a fixed price or a reason not to build yet.