The Kilometre Between the Certificate and the Payroll

A training programme is not the same thing as a labour market. The current African AI-skilling story is quietly failing to hold that distinction in view, and the summer's reporting is where the gap starts to show.

What 99.6 percent measures

Nigeria's Federal Ministry of Communications, Innovation and Digital Economy has been publishing a stream of updates on its Three Million Technical Talent programme. In its May 2026 alliance note with Microsoft, the ministry reports 1.87 million registrations across all 774 local government areas, roughly 135,000 fellows trained across three cohorts, and a 99.6 percent completion rate on programme assessments. In its first Microsoft Elevate wave, the partnership hit close to 9,000 trained against a target of 7,000, with 86,000 additional engagements. On its own placement page, 3MTT lists a Talent Registry offering roles at ₦150,000 per month for about 3,000 fellows, and the UN Development Programme has separately confirmed 3,000 more placements through its Nigeria Jubilee Fellows arm.

Those figures deserve to be taken seriously. 1.87 million registrations across every local government area of Nigeria is not trivial in a country whose youth unemployment estimate ranges from the World Bank's official five percent to the 53 percent figure carried in the State of the Nigerian Youth Report. The state has decided the labour-market problem is fundamentally a skills-supply problem. Real budget and real administrative capacity are being spent on the answer.

The 99.6 percent completion figure, though, does what most training-programme completion figures do. It measures the wrong thing, loudly. Assessment completion tells you that fellows who stayed to the last screen clicked through to the last screen. It tells you nothing about retention six months later, about applied output, about employability, about whether the person now solves a class of problem they could not solve twenty weeks earlier. The ministry knows this. The same rollout has spawned five separate placement bridges: the NJFP arm, the Airtel NextGen Fellowship running across 45 learning centres, a Hello.CV recruiter deal targeting 20,000 fellows, the Talent Registry, and Microsoft's career fair held in Lagos in mid-July 2026. If completion were the outcome, none of these bridges would need building.

The upstream tier is being built out at the same time. In late July 2026 the Ministry of Education, through a signed agreement with ICEDT Consult, launched a nationwide AI capacity programme for 11,700 teachers in the Federal Unity Colleges, following a pilot in six colleges across the six geopolitical zones. That is a different tier of the pipeline entirely, upstream of the 3MTT fellow. It will produce the second-order measurement question of the decade: whether teachers trained to teach AI actually teach it, and whether their students carry the skill into a labour market able to absorb it.

The IMF's colder arithmetic

On 21 July 2026 the International Monetary Fund published a departmental paper by Martin Schindler and Andrew Tiffin's team, titled Unlocking the Potential: AI in Sub-Saharan Africa. The baseline scenario is the sentence to read. Under today's infrastructure, today's skill levels, and today's rate of AI adoption, the region's productivity lift over the next decade is estimated at 0.2 percent, contributing about 0.4 percent to cumulative GDP growth. The alternate scenario, a 2.1 percent productivity lift and around 4 percent GDP contribution, requires the region to move on electricity, connectivity, regulation and skilled labour supply in a coordinated way it has never yet moved on anything.

The Fund's own companion editorial is plainer than most Bretton Woods prose. The young workforce of the region will either find more productive jobs, or watch the global productivity gap widen further. There is no soft middle in that sentence. The baseline scenario is not a warning about failure that might one day come. It is the current-conditions read of what a serious body of macroeconomists expects if the training numbers keep growing while the infrastructure and placement plumbing does not.

Hold the two datasets in one frame. Nigeria is training 135,000 fellows in three cohorts. The IMF baseline says the aggregate productivity impact of AI adoption for sub-Saharan Africa is 0.2 percent over a decade. These numbers do not contradict each other. They describe two ends of the same pipe. Training builds inputs. Whether the inputs move the output depends on whether they land in jobs that use them, on infrastructure that supports the jobs, and on economies that can absorb the output. That is not a training-programme question. It is a placement-infrastructure-and-market question, and it is being reported as if it were the first.

Nairobi is measuring a different thing

Kenya has taken a different route. UNESCO's Global Skills Academy in the TVET sector reached more than 5,300 beneficiaries and over 200 enrolled teachers across 21 TVET institutions and the Kenya School of TVET by end of May 2026, with Machakos University as an earlier anchor. Moringa School's free AI upskilling initiative announced this year targets 3,600 Kenyan youth by December. Both are order-of-magnitude smaller than Nigeria's national programme, and both are structured to look more like a bootcamp cohort than a state literacy drive.

The comparison isolates the variable that Nigeria's headline number cannot see. In heavier-duration Kenyan private programmes of the ALX class, community-reported completion rates run well below half. That range has been openly discussed in local sector reviews for a couple of years, without an official rebuttal from the operators. Both numbers are true. Nigeria's 99.6 percent reports whether a fellow finished a bounded state curriculum. The Kenyan private figure reports whether a fellow spent nine to twelve months absorbing a much heavier programme end-to-end. A programme that asks less will always report a higher completion rate. That is not a scandal. It is a caution about the meaning of the number.

What honest reporting would look like

Absa, together with Microsoft Elevate and Women in Tech, has now expanded its ElevateHer AI programme across nine African markets after the South Africa launch in September 2025, reporting more than 10,000 learners reached in the first phase. The interesting feature is not the badge count. It is that Absa's own hiring desk sits at the end of the pipe. When the training is paid for by the same organisation that will hire the graduate, the completion signal starts to be reported alongside conversion, retention and salary lift, because the bank has an internal reason to want the honest figure.

The Naspers and Prosus report on Africa's AI Imperative names the underlying diagnosis in one line. The continent accounts for less than one percent of the global AI economy while carrying the world's youngest population, and the same pattern shows up in infrastructure, data, talent, capital and policy: "activity without integration, ambition without execution, intent without coordination." That describes what a training programme looks like when it is optimised to report its own scale rather than to move an economy.

The critic worth reading on Nigeria's specific case is ITEdgeNews on 3MTT's financial and quality difficulties. Their two claims: that partnering organisations expected to guarantee post-training placement often lack the network and the finance to do so, and that many graduates have not yet secured roles that make meaning of the training. Both deserve engagement on the merits. On the first, the ministry has responded. The NJFP arm and the Hello.CV partnership exist precisely because the private placement pipe has been thin, and building that plumbing is the right instinct. On the second, the ministry has not published cohort-level employment outcomes, and until it does, the critique stands.

The metrics an honest programme would carry are known and unglamorous. Every cohort should publish, at six and twelve months, the fraction placed in a role that uses the training, the median earnings before and after, the retention rate at eighteen months, and the true dropout rate during training rather than only the completion rate on the terminal assessment. These are not novel demands. They are the standard set a decent labour economist would ask of any active labour-market programme in the OECD. The reason they do not yet appear in the 3MTT reporting is not bad faith. It is that they take longer, cost more, and rarely flatter the headline.

If you finished your final 3MTT assessment this morning in Kaduna, congratulations, and I mean that. The assessment is not nothing. Now find the person who employs someone doing the job you trained for. Ask them what a six-month contract with you would look like, and what they would need to see from you first. Take the answer they give. That is your actual completion rate.

Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan, an Energy AI startup, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.