The AI-to-P&L Note · No. 2

POPIA section 71: what boards must ask about automated decisions

As South African financial services firms put AI into credit, fraud, claims and collections decisions, section 71 of the Protection of Personal Information Act stops being a compliance footnote. It becomes a question about how the business is run, and boards should be asking it.

Watch the 40-second version

What section 71 says

Section 71 of POPIA (Act 4 of 2013) deals with automated decision-making. In summary:

  • A person may not be subject to a decision that has legal consequences for them, or affects them to a substantial degree, which is based solely on the automated processing of personal information intended to profile aspects of that person, such as their performance at work, creditworthiness, reliability, location, health, personal preferences or conduct.
  • The prohibition does not apply where the decision is taken in connection with concluding or performing a contract and either the person’s request has been met or appropriate measures protect their legitimate interests; or where the decision is governed by a law or code of conduct that specifies such measures.
  • Those appropriate measures must give the person an opportunity to make representations about the decision, and require the responsible party to give them sufficient information about the underlying logic of the automated processing.

Read that against a modern financial services firm and the reach is obvious. Credit scoring, affordability checks, fraud flags that freeze accounts, claims triage, pricing and collections strategies can all involve automated profiling with substantial effects on customers. Staff are covered too: performance at work is named in the section.

The legal question is narrow. The operational question it raises is not: do you actually know which of your decisions are made by machines?

Why this is a board matter

AI changes the economics of automation. Decisions that once needed a person can now be made at volume by a model, often one supplied by a vendor and updated outside your release cycle. The word “solely” matters: a human reviewer who cannot see the reasons for a recommendation, or who approves by default, may not be the safeguard management believes it is. Boards should test whether human oversight is real.

The exposure is not only legal. Unexplained automated decisions generate complaints, conduct concerns and reputational damage. The safeguards section 71 requires, representations and an explanation of logic, are also good customer practice and good model-risk practice.

Eight questions boards should ask management

  1. Inventory. Do we have a complete list of automated and AI-assisted decisions that affect customers or staff, including those inside vendor systems?
  2. Solely automated. Which of those decisions are made without meaningful human involvement, and how did we decide what counts as meaningful?
  3. Effect. Which have legal consequences or substantially affect people, for example declines, limits, account restrictions, claim outcomes or pricing?
  4. Basis. For each, which section 71 exception do we rely on, a contract or a law or code of conduct, and is that documented?
  5. Representations. How does a customer challenge an automated decision, how quickly is it reviewed, and by whom?
  6. Explanation. Can we explain the underlying logic of each model in plain language that a customer and a regulator would accept?
  7. Vendors. Do our contracts give us enough information about third-party models to meet these obligations, and the right to audit them?
  8. Accountability. Who owns model risk for each decision, how is performance and drift monitored, and how does it reach this board?
A practical first step

Ask for the inventory first. Every other question on this list depends on knowing what is already automated, and the inventory is the evidence the rest of the governance work is built on.

Governance that enables value

None of this argues against automation. Firms that can show the regulator, and their own customers, how automated decisions are made and challenged will be able to automate more, with more confidence. Governance designed alongside the business case is how AI value gets permission to scale.

This note is general commentary, not legal advice. Take legal advice on your specific obligations under POPIA.

Take the AI Governance Check

Sources

  1. Protection of Personal Information Act 4 of 2013 (South Africa), section 71: Automated decision making. Summary above is a paraphrase; refer to the Act for the authoritative text.
  2. Information Regulator (South Africa): inforegulator.org.za.
Next step

Apply this to your organisation.

The Operating Model Diagnostic gives your executive team a prioritised, governed view of where AI can move the P&L. Pricing on request.

Request an Operating Model Diagnostic