Choosing legal AI as a South African firm

Why the choice matters now

The early fear was that AI would shrink the legal profession. The argument now runs the other way. Rob Green, CEO of the GRM Group, put it like this in October 2026:

It turns out that AI does not replace lawyers. It replaces tasks. And when you make a valuable service cheaper and faster to deliver, you do not get less of it. You get more.

Economists call this the Jevons Paradox - the observation that efficiency improvements expand consumption rather than contract it. Legal AI reduces time spent on routine drafting and research by 30-60%, but rather than shrinking the pie, it expands the total volume of addressable legal work.

Firms are closing more deals, taking on matters they would previously have turned away on cost grounds, and serving clients who could never have afforded the hourly rate.

He also cites a survey in which half of the legal professionals asked report revenue gains of 6% to 20% from AI.

Green expects the biggest shift for smaller firms, and his illustrations are South African: a five-partner Sandton firm that, with the right AI tools, could produce the volume of a 20-person team, and a Cape Town SME practice running contract review, due diligence and regulatory research at a speed and quality that used to need big-firm infrastructure. Juniors, he argues, aren't disappearing; they move into higher-value work earlier. And on fixed-fee or matter-based work, time saved goes straight to the firm's margin.

Green expects AI to create legal work too, because generative AI won't be reliably accurate for years. That's the other reason the choice of tool matters: the right one should make it easy to check every answer, not ask you to trust it.

For most firms, the question is no longer whether to use AI, but which tool fits how your firm actually works.

General-purpose AI: ChatGPT, Copilot, Gemini

General AI assistants are fast, cheap and good at writing. The trouble for legal work is that they answer from general training data rather than from South African statutes and case law, so citations need to be checked one by one. South African courts have already dealt with matters in which lawyers relied on cases which don't exist.

Confidentiality is the second issue. Whether client information is retained or used for training depends on the plan and settings a firm chooses, and consumer accounts are a poor fit for privileged material, or for clients' personal information under POPIA.

Choose general AI if you need help with non-confidential writing, and you have a firm policy that keeps client information out of it.

Global legal AI platforms: Harvey, Legora

Harvey and Legora are platforms built for large law firms, and both have South African customers: Bowmans adopted Harvey in December 2024, and Webber Wentzel rolled out Legora in August 2025. They are used for document-heavy work such as due diligence, contract review and drafting, and both work inside Microsoft Word and connect to document management systems.

They serve many jurisdictions, so South African procedure - court days, the Uniform Rules, Gazette commencement dates - is not their primary focus.

Choose a global platform if you are a large firm with cross-border practice, an innovation team and an enterprise budget.

Legal publisher AI: Lexis+ AI and Juta

South Africa's established legal publishers are adding AI to the research databases many firms already use. LexisNexis launched Lexis+ AI in South Africa in June 2025, drawing on verified South African statutes, regulations and case law. Juta works with Thomson Reuters to offer Westlaw and Practical Law in Africa, and in 2024 expanded into practice tools by acquiring the Contractzone and Litigator businesses.

The strength here is authoritative editorial content and commentary. The trade-off is that publisher AI is primarily a research tool; drafting, deadlines and the rest of matter work generally happen elsewhere.

Choose publisher AI if your main need is AI-assisted research over content you already subscribe to.

South African legal AI tools

A number of local companies have built AI specifically for South African law, and they differ quite a bit in focus:

• Legal Larry is aimed at litigators, with research and pleading drafts grounded in an index of South African judgments and court rules, sold on monthly plans for individual litigators and small teams, and a 24-hour pass.
• Murphy's Law (by Kvell Dynamics) serves both legal professionals and the public, with drafting, document review and research, and has expanded to Namibia.
• Anvaya combines document review, research and matter management, with drafting inside Microsoft Word, and won the South African chapter of the Start-up World Cup.
• LexVivens offers research and drafting through its LexCore product, plus transcription and workflow tools.

Local tools share Squire's starting point - South African law first - and several are priced for individual practitioners.

Choose a local research or drafting tool if you mainly need fast, South African-grounded answers or drafts for individual practitioners, rather than a shared workspace for the whole firm's matters.

Where Squire fits

Squire is built for South African practice first, and for the work around research as well as research itself.

• The matter is the starting point. Upload a summons and Squire builds the matter; research, drafts, documents, correspondence and deadlines stay connected to it, so you never re-explain the background.
• Answers are checked, not just generated. Squire's Hallucination Firewall checks citations against indexed legislation and trusted sources such as SAFLII and the courts' own websites. Citations it can't verify are removed and flagged - so you can see what is supported, rather than being promised there are no errors.
• South African procedure is built in. Squire reads court orders, summonses and notices of motion and calculates deadlines in court days under Uniform Rule 4, recalculating dependent dates when an earlier step moves.
• Built for the firm, not just the individual. Shared matters, conflict checks across parties, contract management with approvals and renewal reminders, e-signature, and guest access for clients and outside counsel. Workflows do the routine follow-up - a review drafted when a document arrives, a renewal task a month before a contract expires - and pause for a person's approval where it matters.
• Your data stays yours. Firm data is stored in your jurisdiction, and documents and matter records are encrypted under a key unique to your firm. The people in a document are pseudonymised before it reaches an AI model. Today that analysis may run outside your jurisdiction, by enterprise AI providers contractually barred from training on your data; we are moving it onto servers we operate in your jurisdiction. Squire does not train AI models on customer data. On-premise deployment is available for Enterprise customers by arrangement.
• Pricing you can plan around. Fixed per-seat pricing in rand, with no token or usage charges. See pricing.

Questions to ask any legal AI vendor

Whichever tool you consider, ask these before signing:

1. Which South African sources are answers grounded in, and how current are they?
2. What happens when the AI can't find support for a claim - is it flagged, removed, or presented anyway?
3. Where is our data stored and processed, including by the vendor's AI providers?
4. Is our data ever used to train models - by the vendor or its sub-processors?
5. Will you sign a POPIA operator agreement with us?
6. Does it work around our matters and documents, or is each question a separate chat?
7. What does it cost per user per month, and are there usage caps or overage charges?
8. Can we trial it on a real (anonymised) matter before committing?