Industry · Professional Services

Law firm automation in South Africa, with the checking built in

FICA onboarding that records who verified what, AI research that will not let an unchecked citation reach a filing, and document review a practitioner signs off. Built around the LPC Code of Conduct, the FIC Act and POPIA, for law firms, accounting and audit practices and consultancies.

7 of 9
authorities cited in a KwaZulu-Natal notice of appeal that the court could not find
Source: Mavundla v MEC, KZN High Court (8 Jan 2025)
High
the FIC's rating of the inherent money-laundering risk in the legal practitioner sector
Source: FIC sector risk assessment: legal practitioners
89%
of audit firms in IRBA's 2025 survey are using generative AI in audits or integrating it
Source: IRBA Staff Audit Practice Alert 11 (Sep 2025)

Why AI in a law firm is a supervision problem first

In September 2024 a Pietermaritzburg judge asked the court's two law researchers to find the nine authorities cited in a supplementary notice of appeal. They found two, and one of those was cited wrongly. The candidate legal practitioner who drafted it denied using AI. The judgment, Mavundla v MEC, handed down on 8 January 2025, says the real source of the authorities remained unknown. It still went to the Legal Practice Council, and the firm was ordered to pay the costs of two extra appearances itself.

For a managing partner, the useful lesson is not about any one tool. The court relied on rule 18.3 of the LPC Code of Conduct, which requires a practitioner to exercise proper control and supervision over staff. It agreed that this includes checking anything staff take from generative AI. Six months later, in Northbound Processing, the fictitious citations came from a paid legal research tool that claimed to be trained only on South African law. That matter went to the LPC too.

So law firm automation has to put the check into the workflow, not into a policy memo. Research tools should return only authorities they can link to a primary source and flag the rest. A filing should be blocked while any citation is unverified. FICA onboarding should record who verified each document and when. When the LPC, the FIC or a client asks what happened, the audit trail should answer it.

Accounting and audit firms face the same principle, and their regulator has set it out. IRBA's September 2025 practice alert says auditors who use generative AI must stay accountable for the audit and warns against over-reliance on the tools. Our FICA document-verification platform for a neo bank and our ERP-to-warehouse integration for a financial services finance team are the closest examples of both sides: automated checks, and data a professional can trust.

How it works

How a new client moves through automated FICA onboarding

This is the workflow we would build first in most law firms. Software collects, reads and checks the documents against your risk management and compliance programme. Your programme decides what is low risk. A person decides everything else.

  1. STEP 01

    Matter opened, documents requested

    A new-matter form starts a secure request for ID, proof of address and, for companies and trusts, the documents that show who owns and controls them. Reminders go out automatically, so nobody has to chase the client by email.

  2. STEP 02

    Documents read and cross-checked

    Document intelligence extracts names, ID numbers, registration numbers and addresses, then checks that they agree across documents. Fields it cannot read with confidence go to a person. It does not guess.

  3. STEP 03

    Ownership and control mapped

    For legal persons, trusts and partnerships, the ownership and control structure is built out until it reaches the natural persons the FIC Act requires you to identify as beneficial owners.

  4. STEP 04

    Risk-rated against your RMCP

    The client, the matter type (conveyancing, trust work, entity formation) and the jurisdiction are scored against the risk factors in your risk management and compliance programme, using the rules your compliance officer sets.

Automated

Low risk, complete file: cleared to open

If every check passes and the risk score is inside the limits your compliance officer sets, the file is marked complete and the matter can be opened. The record shows which checks ran and which rule cleared it.

Human decides

Higher risk or gaps: routed to compliance

Missing documents, mismatches, complex ownership or higher-risk matters go to the compliance officer with a one-page summary and the reason for referral. Any decision to report something to the FIC is always made by a person.

Every verification evidenced

What was collected, what was checked, which rule applied and who signed off are all retrievable by query, so an FIC inspection or an LPC enquiry gets a report instead of a search through email.

Guides

Law firm automation, explained: how to do it and why it matters

These are the questions partners, compliance officers and practice managers ask us, answered plainly. Each one starts with a short answer you can take into a partners' meeting. The steps or reasons follow.

01 · How to

How do you automate FICA client onboarding at a South African law firm?

Short answer

Automate collection, reading and cross-checking, and keep the risk decision with your compliance officer. The FIC Act requires client identification, beneficial-ownership checks for legal persons and trusts, and a risk management and compliance programme. Software can gather and check the evidence. It should record who verified each step, and higher-risk files should go to a person.

  1. 1

    Start from your RMCP. Automation enforces the programme you have and cannot replace one you lack.

  2. 2

    Send one secure document request when a matter opens, with automatic reminders, instead of chasing documents by email.

  3. 3

    Extract and cross-check names, ID and registration numbers, and send low-confidence fields to a person.

  4. 4

    For companies, trusts and partnerships, map ownership and control down to the natural persons who must be identified as beneficial owners.

  5. 5

    Risk-rate the client and matter using the rules your compliance officer sets, and route higher-risk files to them with the reason.

  6. 6

    Log every check, rule and sign-off so an inspection gets a report. (Our recommendation from delivery work, not a quote from the FIC.)

Source: LPC: Navigating FIC Act compliance in legal practice (Jul 2025)
04 · How to

How can a law firm use AI on client files without breaching POPIA or privilege?

Short answer

Treat the AI provider as an operator, and settle where the data goes before rollout. POPIA requires reasonable security safeguards and a written contract with any operator, and it restricts transfers abroad without adequate protection. The LPC Code separately requires practitioners to keep privilege and client confidentiality.

  1. 1

    Map which client data each tool will see, and leave out what the task does not need.

  2. 2

    Put a written operator agreement in place covering security measures and breach notification, as section 21 requires.

  3. 3

    Find out where the model runs and whether prompts or files are retained or used for training. If it runs abroad, check section 72.

  4. 4

    Use firm-approved tools only, so client files do not end up in personal accounts.

  5. 5

    Keep privileged material in environments the firm controls where the matter requires it. (Experience-based: this is how we scope it.)

Source: POPIA section 72: transfers of personal information outside the Republic
05 · Why

Why do South African audit firms need an AI governance policy now?

Short answer

Because adoption is already common and IRBA has said who is accountable. In IRBA's 2025 survey, 89% of participating firms had implemented generative AI in audits or were integrating it. IRBA's September 2025 alert says auditors must stay accountable for the audit and makes the engagement partner responsible for appropriate use of these tools.

  • IRBA's survey was of auditors of publicly traded entities, so the 89% describes that group, not every practice.

  • At least 56% of participating firms had an AI governance framework in place or in development.

  • ISQM 1 requires firms to respond to new risks, and IRBA points to its technological-resources requirements for GenAI.

  • IRBA warns that over-reliance weakens professional scepticism, and that putting client data into tools without the client's knowledge or consent may expose auditors to litigation risk.

  • IRBA has warned about AI-fabricated evidence and working papers since January 2023.

Source: IRBA Staff Audit Practice Alert 11 (Sep 2025)
Challenges We Solve

The hard problems Professional Services faces

Unchecked AI output

Staff are already using AI tools, whether or not there is a policy. South African courts have referred fictitious citations to the Legal Practice Council even where nobody could show how they were produced.

FICA onboarding by email

Collecting documents, mapping beneficial ownership and evidencing each check through inboxes and spreadsheets is slow, and it leaves a weak trail when the FIC inspects.

Document-heavy matters

Reading, extracting and filing information from contracts, bundles and financial statements by hand takes fee-earner time and introduces errors.

Confidential data in external tools

Client files are personal information and privileged material at once. Sending them to a model nobody has vetted creates POPIA, privilege and cross-border transfer questions that are hard to answer after the fact.

By line of business

Where automation fits, practice by practice

Professional services is not one market. A litigation practice, a conveyancing firm, an audit firm and a consultancy pay for automation in different places.

First win: Research and drafting checks

Litigation and advisory law firms

The risk the courts keep finding is unchecked authority. Citation verification before filing, structured summaries of bundles and contracts with page references, and matter intake with conflict checks all save fee-earner time without taking judgement away from the practitioner.

First win: FICA and beneficial ownership

Conveyancers and trust-account practices

The FIC names conveyancing and client accounts among the services that make legal practitioners vulnerable to money laundering. Onboarding that maps beneficial ownership and records every verification is where automation pays back first.

First win: Data and evidence

Accounting and audit firms

Client data pulled from Xero, Sage and other ledgers into one governed place and reconciled automatically. For audit firms, any generative AI in the engagement sits inside the ISQM 1 policies IRBA expects, with the engagement partner accountable.

First win: Knowledge and delivery admin

Consultancies and advisory firms

Proposals, research packs and status reporting are repetitive. An assistant grounded in your own past work, with sources shown and client data kept apart, frees consultants for the advice clients actually pay for.

Our Solutions

What we build for you

01

Verified research and document review

AI-assisted research and review that links every authority or extracted clause to its source, marks anything unverified, and blocks filing until a practitioner has confirmed it.

02

FICA onboarding that evidences itself

Document collection, extraction, beneficial-ownership mapping and risk rating against your RMCP, with every step recorded against a named person and higher-risk files routed to compliance.

03

Governed data and assistants

Practice, ledger and document data brought into one governed place, with assistants that answer from your own material, show their sources and keep client data where your POPIA obligations need it to be.

Regulatory context

What South Africa's courts and regulators have actually said

Each item below is a judgment, rule or regulator publication you can read yourself. They are dated so you can check them. Where something is our reading and not the document's words, we say so.

  1. Judgment 8 January 2025

    Mavundla v MEC, KwaZulu-Natal High Court

    Of nine authorities in a supplementary notice of appeal, the court's researchers found two. The candidate legal practitioner who drafted it denied using AI, and the court said the real source remained unknown. It found an apparent failure of supervision. When the judge tested ChatGPT, it confirmed that one of the non-existent cases existed. The court said relying on AI for legal research is "irresponsible and downright unprofessional" and sent the judgment to the Legal Practice Council.

  2. Judgment 30 June 2025

    Northbound Processing v SA Diamond and Precious Metals Regulator, Gauteng High Court

    Heads of argument cited cases that do not exist. Counsel confirmed that the incorrect citations appeared to be AI hallucinations. They came from "Legal Genius", a subscription tool that claimed to be "exclusively trained on South African legal judgments and legislation". The court accepted there was no intent to mislead and still referred the practitioners to the LPC. Its view was that fictitious cases in written heads are as serious as in oral argument.

  3. Published 29 March 2019, in force

    LPC Code of Conduct: supervision and confidentiality

    Rule 18.3 requires attorneys to "exercise proper control and supervision over his or her staff and offices". The Mavundla court agreed that this includes verifying information staff take from generative AI. Rule 3.6 requires practitioners to maintain legal professional privilege and confidentiality. Both rules apply to any AI tool a firm adopts.

  4. Amended Schedule 1 in effect 19 December 2022

    FIC Act: legal practitioners are accountable institutions

    Item 1 of Schedule 1 covers attorneys practising for their own account, advocates with a fidelity fund certificate and law firms. They must register on goAML and keep a risk management and compliance programme (RMCP) that meets section 42. They must identify clients and beneficial owners, and they must submit a risk and compliance return under Directive 6. The first returns were due on 31 May 2023.

  5. 24 October 2025

    South Africa leaves the FATF grey list

    The FATF removed South Africa from its list of jurisdictions under increased monitoring. The country had been grey-listed in February 2023. The delisting did not amend the FIC Act, so firms' obligations stay the same. In our view, consistent onboarding and evidence are what keep the sector off that list.

  6. Issued September 2025 (non-authoritative guidance)

    IRBA Staff Audit Practice Alert 11: generative AI in audits

    IRBA's alert says auditors using generative AI tools "must maintain oversight and remain accountable for the audit process". It makes the engagement partner responsible for appropriate use of these tools under ISA 220 (Revised), and it cautions against over-reliance because that weakens professional scepticism. IRBA had warned about AI-fabricated evidence and working papers in January 2023 and in September 2024.

  7. In force

    POPIA: security, operators and cross-border transfers

    Section 19 requires reasonable technical and organisational measures to protect personal information. Section 21 requires a written contract with any operator that processes it for you. Section 72 restricts sending it to a third party in a foreign country without adequate protection or another listed ground. A hosted AI model is usually an operator, and often one abroad.

Where we start

Engagements we run in Professional Services

01

Research with citations checked before they leave

AI-assisted research that links every authority to its primary source and marks anything it cannot verify. A document cannot move to filing while a citation is unverified. The check is enforced by the workflow, not left to memory.

02

FICA onboarding that evidences itself

Client documents collected, read and checked against your risk management and compliance programme, with beneficial ownership mapped and every verification recorded against a named person.

03

Contract and bundle review

Clauses, dates, parties and obligations extracted from contracts and bundles into a structured summary, with the source page cited for each item and a practitioner signing off the exceptions.

04

Matter intake and conflict checks

New-matter details captured once and checked against existing clients and counterparties. Possible conflicts are sent to a partner instead of being missed in a spreadsheet.

05

Month-end and reporting for accounting firms

Client data pulled from Xero, Sage and other ledgers into one governed place and reconciled automatically. Only the mismatches go to an accountant.

06

Time capture that happens as you work

Draft time entries built from calendar, email and document activity, then confirmed by the fee earner, instead of reconstructed from memory at month-end.

Key terms

Professional Services terms, in plain language

Accountable institution
A business listed in Schedule 1 of the Financial Intelligence Centre Act that must meet its anti-money-laundering duties. Legal practitioners are item 1.
RMCP (risk management and compliance programme)
The documented policies, processes, systems and controls an accountable institution uses to meet the FIC Act, as section 42 requires. It covers client due diligence, record-keeping, reporting and training.
RCR (risk and compliance return)
The FIC's self-assessment questionnaire on money-laundering, terrorist-financing and proliferation-financing risks and compliance. Legal practitioners file it under Directive 6.
Beneficial owner
The natural person who ultimately owns or controls a client that is a company, trust or partnership. The FIC Act requires accountable institutions to identify them.
goAML
The FIC's registration and reporting platform. Accountable institutions register on it before they can file reports or a risk and compliance return.
Candidate legal practitioner
A trainee attorney or advocate doing practical vocational training, previously called an article clerk. The supervising practitioner remains responsible for their work.
Hallucinated citation
A case, quotation or reference produced by an AI tool that does not exist or does not say what is claimed, presented with the same confidence as a real one.
Operator (POPIA)
A person or provider that processes personal information on a firm's behalf under a contract, without coming under the firm's direct authority. Most hosted AI and cloud tools are operators.
FAQ

Professional Services questions

Ready to transform your Professional Services data strategy?

Let's discuss the specific challenges your business faces and design a solution that delivers real impact.