AI will not transform your business, but it can stop you losing points

No algorithm has ever created a job, built a supplier or paid a dividend to a Black shareholder. Transformation is done by people who decide to hire, to buy differently, to share ownership and to stick with a programme for long enough that it works. AI cannot do any of that for you. We build AI tools ourselves, and that is exactly why I would rather say so plainly.

What AI can do is narrower and, for anyone who has prepared a verification file, surprisingly valuable. It can stop you losing points you have already earned. That is a less exciting promise than "transformation, automated", but it is an honest one, and it is where most businesses are quietly leaking value.

Where points are really lost

Ask anyone who has prepared a verification file where the pain sits, and the answer is rarely strategy. It is evidence. The spend was made, the learners were trained, the supplier was supported. But the proof is scattered, incomplete or inconsistent.

If you have been through a verification, you will recognise the pattern. A key supplier's certificate expired three months before year-end and nobody noticed. An affidavit is unsigned. An invoice cannot be matched to the supplier who issued it. The payroll system, the training records and the B-BBEE spreadsheet each give a slightly different headcount, because the same data was captured three times by three people. A learner's file is missing one document, and the claim falls away.

None of these are failures of commitment. They are failures of administration, under time pressure, in a process that still runs largely on email attachments and spreadsheets. And they cost real points, sometimes enough to miss a sub-minimum and lose a level.

The supplier certificate is the clearest example of how the arithmetic works against you. Spend with a supplier who has no valid certificate or affidavit counts at 0%, but it stays in your total measured procurement spend. You are measured on the money and given no credit for it.

Where AI helps

This is work AI does well: reading large volumes of documents, matching records, noticing what is missing and doing it every day rather than in the month before the agency arrives. We see four uses that earn their place.

• Capture once. Payroll, procurement and training data joined into one governed source, so the headcount in your management control calculation is the same headcount in your skills claim.
• Read and check. Certificates, affidavits and invoices read as they arrive, with levels, ownership and expiry dates extracted and filed against the claim they support.
• Watch the score. The level projected through the year, element by element, with sub-minimums watched and an early warning when one drifts towards the line.
• Draft the report. Monthly progress, committee packs and the social data in an ESG report drafted from the same records, so people spend their time reviewing rather than retyping.

What these have in common is timing. AI does the reading and the checking continuously, so problems surface in May rather than at verification. A supplier certificate that expires mid-year becomes a reminder to ask for the new one, while there is still time to get it.

Where it must not decide

There is a line, and it should be drawn clearly before any system is switched on. AI drafts and flags. A person confirms every figure that goes to verification.

Let the machine do the reading. Keep the judgement with people who can be held accountable for it.

That is not caution for its own sake. B-BBEE claims are signed and defended by people. A verification agency will test the evidence, not the software, and "the system calculated it" is not an answer anyone can stand behind. Some calls also need judgement that no model should make on its own:

• Whether a person is recognised in a particular category, which depends on facts about real people that must be confirmed, not inferred.
• Whether a structure or relationship carries the indicators of fronting. AI can flag a pattern; deciding what it means is a governance matter.
• Whether a contribution qualifies under the Codes when the documents are ambiguous. That is a professional judgement, and it should be recorded as one.
• Anything that affects a person's job, bursary or place in a programme.

Designed this way, the person in the role does more of the work that needs them. Instead of spending weeks assembling evidence, your champion spends that time on the exceptions the system has raised, and on the decisions only a person should take.

Design POPIA in before anything is switched on

Transformation data is some of the most sensitive data a company holds. It records people's race, gender and disability status, alongside their pay, their training and sometimes their shareholding. Under POPIA, race and health information is special personal information, and disability information will usually be treated as health information. All of it deserves more care than a sales pipeline or an inventory list.

That care has to be built in from the start. Before any document goes near an AI system, I would want clear answers to a few plain questions. Does each measure use only the data it needs? Where totals are enough, are identities masked? Who can see individual records, and is that access logged? Where is the data stored and processed, and could it be used to train a model outside your control? If a supplier asked you tomorrow to show how their information is handled, could you?

A system that cannot answer those questions should not be handling your employees' records, however impressive its dashboard. A system that can answer them often turns out to be more private than the spreadsheets it replaces, which travel by email and live in many inboxes.

Start with your last verification

If you are considering AI for your B-BBEE or ESG work, I would start with a modest question: where did we lose points, or nearly lose them, at our last verification? The answer usually points to one or two places, often supplier certificates or skills evidence, where reading and checking would make the most difference. Begin there, prove the value, keep a person confirming every figure and widen it only when the controls are working.

AI will not make your company more transformed. That still depends on who you hire, who you buy from and who you share ownership with. What it can do is make sure the work you have already done is counted, in full and on time.

If you would like to talk through where AI could sensibly help your scorecard, and where it should stay out, write to us at sales@bsvision.co.za. We are happy to start with your last verification report and work from there.