Why South African Boards Need AI Grounded in South African Rules
South African boards need AI that connects relevant local rules to the organisation’s facts, the period under review and the decision being made. A useful analysis identifies its sources, applies the appropriate framework, exposes missing evidence and shows how it reached its figures. Recognising the names of South African laws and codes is only the beginning.
Consider a board reviewing two proposals for improving the company’s B-BBEE position. Both sound sensible. Both have a budget. Both arrive with confident explanations of the expected benefit.
Before choosing, the board needs to know whether the proposals were assessed under the applicable code, whether the expenditure and evidence qualify, and whether the projected result accounts for the relevant conditions.
Those details can change the recommendation. They belong at the start of the analysis.
What does “South African context” actually mean?
Local context is often treated as a finishing touch: use rands, mention POPIA, add a reference to King and adopt a familiar business tone.
For consequential work, it goes much deeper. The context determines which questions to ask, which evidence to collect, how to calculate and what the result means.
Three examples illustrate the point:
B-BBEE planning: The applicable generic or sector code, entity circumstances, measurement period and evidence behind the inputs.
Governance review: The relevant governance framework, its version and its application to the organisation and reporting period.
Tender preparation: The particular tender’s eligibility requirements, evaluation method, submission instructions and required evidence.
In each case, adding a paragraph about South Africa after completing a generic analysis is too late. The local requirements have already determined what a sound analysis should contain.
Why knowing about B-BBEE is insufficient
B-BBEE provides a useful example because the starting point matters. The Department of Trade, Industry and Competition publishes both the amended general codes and sector codes, including codes for ICT, tourism and financial services. Its guidance addresses when a measured entity must be assessed under an applicable sector code. Sources: the dtic’s codes library (https://www.thedtic.gov.za/financial-and-non-financial-support/b-bbee/b-bbee-codes-b-bbee-acts-strategies-policies/) and sector-code guidance (https://www.thedtic.gov.za/financial-and-non-financial-support/b-bbee/b-bbee-charters/).
An AI system can explain empowerment policy fluently and still begin a planning exercise on an untested assumption about which rules apply.
A useful process establishes that basis explicitly. It identifies the entity and measurement period, confirms the applicable code and separates documented inputs from estimates. Only then should a calculation support a recommendation.
This is why a source link, although valuable, is insufficient on its own. The analysis needs to connect the source to the specific rule being applied and the facts that make it applicable.
A worked planning example: the missing supplier evidence
Imagine a fictional company comparing two options: increasing qualifying skills expenditure or changing part of its supplier mix. Its finance team provides a spending schedule, but some supporting supplier evidence is missing.
A defensible planning process can still make progress. It should make the missing information visible and distinguish confirmed results from conditional scenarios.
A procurement schedule with incomplete supporting evidence: Which entries are evidenced, which require confirmation and how those gaps affect the result.
A proposed skills budget: Which assumptions about eligibility and recognition need checking before estimating a benefit.
A target B-BBEE level: The calculation basis, constraints and conditions behind any projected improvement.
A preferred investment option: The cost, expected effect and outstanding evidence needed before committing.
The board can then see what it is deciding. It might approve a provisional budget, ask the team to close the evidence gaps or request a revised comparison. Each is more useful than treating an incomplete calculation as a settled outcome.
This example describes the discipline a planning workflow should support. It is not a calculated scorecard or a claim about a client’s achieved result.
Rules need a version and a date
Local knowledge changes. An answer can refer to a real framework and still use the wrong version for the work under review.
The transition from King IV to King V is a current example. The Institute of Directors in South Africa states that King V was released on 31 October 2025, supersedes King IV and is effective for financial years beginning on or after 1 January 2026. Source: IoDSA (https://www.iodsa.co.za/page/king-v).
A governance analysis should therefore establish the relevant period and framework before applying a checklist. Simply mentioning “King IV/V” does not explain the basis of the review.
This principle extends beyond governance codes. When rules change, someone has to assess the change, decide how it affects the workflow, update the relevant content and check that the resulting analysis still behaves as intended.
That is an ongoing responsibility. A product should make its update status and limitations visible rather than asking customers to infer currency from the confidence of its answers.
How we apply this at ExecutiveNavigants
ExecutiveNavigants uses versioned rule packs in its B-BBEE planning tool. A rule pack is a named set of rules represented as data that the calculation software can apply. The output identifies the pack used, making the calculation basis inspectable.
The distinction between sources is also visible. Our current implementation distinguishes packs encoded from primary gazettes from research-seeded packs that still require further source verification. The latter carry a warning. Even a gazette-sourced pack supports planning; the output is not a formal verification certificate.
The tool also checks the delivery of the calculation. After building the Excel workbook, it reopens the saved file and compares specified calculated values with the engine’s results. That helps detect a mismatch between what the software calculated and what the user received.
These checks serve different purposes. Source verification concerns the rules. Calculation checks concern their implementation. Evidence review concerns the inputs. A complete planning exercise needs attention to all three.
It is also important to be precise about scope: the existence of B-BBEE rule packs does not establish that every South African regulation has been encoded or that every update is automatically incorporated.
Local knowledge includes the documents in front of you
National rules provide only part of the context. The organisation’s own evidence and the specific transaction matter too.
For a tender response, the workflow should work from the actual tender documents, track their requirements and surface ambiguities that need clarification. A generic account of procurement practice cannot establish what this buyer has requested.
For a board review, the analysis needs to distinguish a statement in the pack from a conclusion it has inferred. Missing support for a recommendation should become a question for management.
The choice of material also matters. POPIA governs the processing of personal information, so teams should establish what they are authorised to supply and assess the service’s data handling before using sensitive material. A product’s South African positioning does not, by itself, establish compliance. Source: Protection of Personal Information Act (https://www.gov.za/documents/protection-personal-information-act).
What should a South African board ask an AI provider?
Ask the provider to demonstrate one relevant workflow using representative material your organisation is authorised to share.
Then ask which rules and versions it used, how it determined applicability, where the material facts came from and what happened when evidence was missing. Inspect how the figures were calculated and which parts of the delivered output were checked. Establish who maintains the domain content and how users learn about limitations or changes.
The answers should be visible in the process and the output. They should help the board understand both the work completed and the judgement it still owns.
South African executives can benefit enormously from global AI capability. Turning that capability into dependable local work requires deliberate choices about method, sources, calculations and review. That is the work we are building into ExecutiveNavigants (https://executivenavigants.co.za/).