M&A is the practice where legal AI does its heaviest lifting — due diligence and data extraction are the top-cited use case and the biggest measurable time saving. But it is also the practice where the work that matters is exactly what AI can't do: judging materiality, structuring the deal, and allocating risk in the warranties. And in South Africa a global tool misses the three gates every deal runs through — the Companies Act, merger control, and the Takeover Regulation Panel. This leaf is the deal read for the AI era: where the tool helps, where it must not, and the SA regulatory stack it doesn't know.
Corporate and M&A practice covers how companies are structured, governed, and — the deal work — bought, sold, and combined. The governing statute in South Africa is the Companies Act 71 of 2008. When a transaction materially alters a company, the Act treats it as a fundamental transaction with heightened protections for shareholders: the disposal of all or the greater part of a company's assets or undertaking (s.112), a statutory amalgamation or merger (s.113), and a scheme of arrangement (s.114). Each requires a special resolution (s.115), and dissenting shareholders get appraisal rights (s.164) — the right to be bought out at fair value.
Around the deal sit three regulators that a generic tool has never heard of: the Competition Commission and Tribunal (merger control), the Takeover Regulation Panel (affected transactions in regulated companies), and, for listed companies, the JSE Listings Requirements. Getting the private-law deal right and missing a regulatory gate is how a signed transaction fails to close.
The first real decision on any M&A matter is structure, and it's a judgement call AI does not make for you. The shape drives the tax, the consents, the liabilities that transfer, and which regulator you answer to.
| Structure | What transfers | Key feature |
|---|---|---|
| Share sale | The shares — the company and all its liabilities come with it | Simplest to transfer; buyer inherits history, so due diligence is everything. |
| Asset / business sale | Selected assets and liabilities, cherry-picked | Buyer avoids unknown liabilities; but every asset needs its own transfer formality and third-party consents. |
| Statutory merger (s.113) | Everything, by operation of law — the companies combine | The only structure where assets and liabilities pass automatically; no asset-by-asset transfer. |
| Scheme of arrangement (s.114) | A court/Panel-sanctioned arrangement between a company and its holders | The route for squeeze-outs and public take-privates; an "affected transaction" under the TRP. |
Due diligence is the buyer reading everything the target has — contracts, leases, IP, litigation, employment, financials — in a data room, and reducing it to an issues matrix: what's material, what's a risk, what needs a warranty or a price adjustment. It is document-heavy, repetitive, and time-boxed, which is exactly the shape AI is good at. Across the independent usage study (see the Legal AI leaf), document review / due diligence was the top in-house use case, and data extraction the task with the biggest measurable time saving — pulling defined fields (change-of-control clauses, assignment restrictions, termination rights) from hundreds of contracts at once.
AI extraction produces a first-pass issues list — fast, broad, and genuinely useful. What it does not produce is the judgement: whether a change-of-control clause is a dealbreaker or a footnote, whether a gap in the data room is an oversight or a red flag, and how to price and paper the risk. The extraction is the drudgery; the deal is in the interpretation. Treat the output as a verified starting point for a lawyer, never as the diligence report itself.
Under the Competition Act 89 of 1998, mergers are classified as small, intermediate, or large by financial thresholds (set by ministerial notice and revised from time to time — check the current figures). Small mergers need not be notified and can be implemented, unless the Competition Commission calls for notification within six months. Intermediate and large mergers must be notified and cannot be implemented until approved — the Commission decides intermediate mergers; the Competition Tribunal decides large ones. This is a hard gate on closing, with real timelines, and it is invisible to a foreign tool.
South African merger control is distinctive for its public-interest test, strengthened by the 2018 amendments. Alongside the competition analysis, the authorities weigh effects on employment, the spread of ownership (including B-BBEE), small businesses, and the ability of national industries to compete — and, increasingly, national security. A merger can be pro-competitive and still be conditioned or blocked on public-interest grounds. No generic legal-AI tool models this, and it is often the part of an SA deal that decides the outcome.
The Takeover Regulation Panel (TRP) — established under s.196 of the Companies Act — regulates "affected transactions" involving "regulated companies": all public companies, state-owned companies, and private companies whose memorandum applies the takeover rules or which meet a share-transfer threshold. Chapter 5 of the Act plus the Takeover Regulations govern mandatory offers, comparable and partial offers, squeeze-outs, and disclosure — and a regulated-company deal generally cannot proceed without TRP approval or an exemption. For JSE-listed companies, the Listings Requirements add a further layer of disclosure and shareholder-approval categories on top.
It's a common surprise that the TRP reaches private companies, not just listed ones — a private company can be a "regulated company" and its transaction an "affected transaction" needing Panel clearance. Assuming the takeover regime is a listed-market-only concern is exactly the kind of error a fast, confident, jurisdiction-blind AI answer will reinforce. This is a check to run early, on the structure, not late.
A clean AI-extracted issues list feels like the answer. It's a first pass a lawyer must verify and interpret. Signing the report the model produced, unread, transfers the risk to the client in a bet-the-company matter.
Diligence is as much about disclosure gaps as disclosed documents. A model reads what it's given and will confidently report "no issues found" on an incomplete room. Absence of a flag is not absence of a problem.
Intermediate and large mergers can be signed but not implemented until approved. Miss the notification, or misjudge the public-interest exposure, and a done deal sits in limbo — or gets conditioned in ways the price never accounted for.
Employee records, customer data, health and criminal information — a data room is a POPIA event. Hosting it, and running AI over it, is a section 72 question if the tool processes abroad. The access model is a legal decision before it's a convenience — see the Data privacy & POPIA leaf.
Diligence findings and legal analysis can be privileged. Running them through a third-party model, or sharing them into a joint data room, can jeopardise that privilege. Design the tooling so the privileged work product stays protected.
The sweet spot for AI-assisted diligence: a share or asset deal with a large but bounded data room, a tight timeline, and a buyer that needs the issues matrix fast. Extraction compresses the review; the lawyer's time moves to structure, warranties, and the merger-control call.
Corporate legal teams doing repeat deals and commercial contracting are the group the usage study found saving the most measurable time — data extraction from commercial contracts is easy to define and measure, and the volume makes the ROI visible where one-off matters don't.
Where AI helps least and the SA regulatory stack matters most: a listed or regulated-company target means TRP clearance, JSE Listings Requirements, and a public-interest-heavy merger filing. The tool drafts and summarises; the deal is won or lost in the regulatory strategy it can't do.
Beyond the Companies Act formalities, an SA deal runs through gates no US-trained model models: Competition Act merger control with its public-interest and B-BBEE lens; the Takeover Regulation Panel for regulated companies (private ones included); exchange control (SARB approval where funds or ownership cross the border); and, for listed targets, the JSE Listings Requirements. A confident foreign AI answer that omits these isn't wrong about company law — it's silent about the four things most likely to shape or stop the deal.
Every SA due diligence is also a POPIA exercise: the data room is full of employee, customer, health, and sometimes criminal-behaviour information — special personal information. Where that data is hosted, and whether an AI tool processes it in or out of the country, is a section 72 transfer decision that has to be made before the room opens. The residency answer for the deal is the same one the rest of the tree gives — keep it in country and the question disappears. This is the Data privacy & POPIA leaf, applied to M&A.
M&A is where the Legal branch's two live leaves meet: it's the practice where AI does the most work, and the practice with the most personal information in the room.
The Acts and the regulators — and note that merger thresholds and the takeover rules change by notice, so verify the current figures against the regulator before relying on them.