You trained 200 people on AI. Prove it changed anything.
A certificate is a receipt, not a result.

Every number on slide 14 is true. Not one of them answers the question. Enrolment measures how many people were told to log in. Completion measures how many clicked to the end. Satisfaction measures whether they enjoyed it. The levy claim measures that the money moved. None of those is evidence that a single person on the team can now do something with AI that they could not do before.

This is not a SkillTrainer problem or even an AI problem. It is how corporate learning has been measured for decades, and AI training has inherited the habit at the exact moment the stakes stopped being academic.

What most companies are actually measuring

Look at what gets tracked and the gap is obvious. The learning function counts activity and sentiment. The business wants capability. Almost nobody is measuring the second thing.

Read the first two together. Most people using AI at work have never been trained on it, and even the companies most serious about developing their people judge success by whether staff felt engaged and whether they stayed. Those are reasonable things for HR to care about. They are not what the CFO asked.

The third number is why this stopped being a soft question. TalentCorp’s study puts 620,000 Malaysian jobs in the highly-impacted band inside five years, and that estimate covers ten sectors that produce about 60% of GDP. For a company in wholesale and retail, food manufacturing or global business services, “did the AI training work” is now a question about whether the workforce keeps up, not whether the L&D budget was well spent.

A certificate is a receipt

Here is the uncomfortable part. Almost everything a company collects at the end of a training programme is a record of the transaction, not of the outcome.

A certificate proves the training happened. A quiz proves someone remembered it for twenty minutes. Neither proves anyone can do the work.

If you cannot name what your team can do now that it could not do in June, the invoice happened. The training did not.

The levy paperwork is the clearest example, and it is worth being precise about why. The HRD Corp claim asks whether an approved programme took place, for whom, and at what cost. That is exactly the right question for a claim, and it makes the funding straightforward for most Malaysian employers. But it was never designed to tell you whether the finance team can now build a working forecast with AI, or whether the sales team can draft a proposal in a quarter of the time. Treating a fully-claimed programme as a successful one is the same mistake as reading a paid receipt as a review.

The end-of-course quiz has the same shape. Multiple choice at the end of a module tests recall in the moment. It does not test whether the person will reach for the right approach on Wednesday afternoon, with real data, a real deadline and nobody watching. We have written before about why completion is the wrong number; the deeper issue is that the number after completion is usually missing altogether.

What proof actually looks like

If activity and sentiment are not evidence, what is? Three things, and none of them is complicated.

A task, not a quiz. The only honest test of AI capability is watching someone use it on the kind of work they do. Give a manager a messy spreadsheet and a real question and see whether they can get a defensible answer out of an AI in fifteen minutes. Give an analyst a policy document and ask for a summary a director would trust. Score the output, not the memory. If your provider’s assessment is a set of tick boxes, you are buying recall.

A map, not a percentage. “94% complete” says nothing about where the team is strong and where it is exposed. What a leader needs is a breakdown: which capabilities each team has actually demonstrated, and where the gaps are. The finance team may be excellent at engineering a prompt and weak at governance, at knowing which data must never leave the building. That is an actionable finding. A completion rate is not.

A number that moves. Capability should be visible over time, on a dashboard leadership can open without asking L&D to run a report. Where did the team start, where is it now, what improved after the last round. If the number cannot go down, it is not measuring anything.

Do these three and slide 14 changes. It stops being “212 enrolled” and becomes “the operations team can now do X, the gap is Y, here is the trend.” That is a slide the CFO can act on.

Design the proof before the programme

The companies that can answer the CFO are not the ones that trained the most people. They are the ones that decided, before anyone logged in, what “it worked” would look like and how they would know.

That usually means three decisions up front. Which real tasks each team should be able to do with AI by the end. What the baseline is today, measured on those same tasks. And who owns the number afterwards, so it keeps being looked at when the programme is over. The funding side is the easy part: training delivered through an HRD Corp registered provider is claimable against the levy your company is already paying, so the cost of doing this properly is rarely the constraint.

This is also, plainly, how SkillTrainer is built. Every course ends in an assessment that tests real application on realistic work rather than a recall quiz, leadership gets an analytics view of capability by team rather than a completion figure, and completion itself is visible rather than assumed. On the AI Productivity Power-Up course, finishing also unlocks a post-completion tutor grounded in that learner’s own assessment results, so the weakest area gets practised rather than filed. If you are equipping a whole team, AI Training covers the self-serve courses, and corporate AI training covers a scoped programme built around your own tasks and baseline.

Either way, decide the proof first. The training is only as real as the number you can show afterwards.