African AI after the pilot: ask who will keep the system running

An AI pilot needs someone who can explain what happens next: who maintains the data pipeline, investigates a bad result and teaches a colleague to take over? For an African business commissioning a system, or a UK client working with an African development team, these questions belong in the project brief. A polished demonstration does not answer them.

GIZ and research firm Briter's October report, AI Made in Africa: Voices from AI businesses across Africa, provides a reason to ask. It describes companies developing local talent while struggling to recruit experienced specialists and pay for infrastructure. Briter's report page is dated 29 October; GIZ's landing page displays 15 September. The downloaded report itself says October 2026. This is a reading of that evidence, rather than a new launch announcement. Read the full report

The distinction between the study and the whole market matters. Researchers mapped 311 AI businesses and analysed 85 unique business responses collected between April and July 2026. Their methods section calls the sample non-exhaustive, notes uneven geographic coverage and says AI-native companies are more heavily represented in the survey than in the mapped dataset. Individual questions also have different response counts. Those limits prevent a country-wide hiring verdict.

Ask who owns the learning

Figure 22 makes one part of the problem concrete. Among 82 charted responses about the biggest challenge in attracting and retaining AI talent, limited qualified talent accounts for 37%, salary competition 32%, international competition 16% and infrastructure constraints 15%. Three responses labelled Other are excluded from the chart. These are the respondents' reported constraints, not measured vacancy rates or evidence that a particular supplier cannot deliver.

The report describes companies combining experienced recruitment with local development, including structured internal training. Its Danaya case study describes investment in young Ivorian engineers alongside experienced technical leadership. PING has not interviewed the company or evaluated its staff. The useful question for a prospective customer is what that learning arrangement means for the work being commissioned, including who reviews a junior engineer's decisions and who can maintain the result.

Make handover part of the brief

PING's suggested check starts with one ordinary maintenance task. Ask the team to describe how a new colleague would diagnose an incorrect output, find the relevant data and decide whether to change the system. Name the person responsible for review, the documentation they need and the boundary between a routine correction and a change requiring approval. This is a proposed commissioning method, not a test PING has conducted or a guarantee that documentation makes a model reliable.

Give learning a compute budget

Training people also needs working infrastructure. The report identifies cloud and compute as a major development expense; Figure 24 states a 72-respondent base and excludes Other and Prefer not to disclose responses. A percentage from that question should not silently become a percentage of all 311 mapped firms. For a project plan, ask what computing access is needed for practice, evaluation and maintenance, who controls it and what happens when an introductory credit ends. Do not assume a credit establishes a sustainable operating cost.

Before a pilot becomes an ongoing service, put four answers beside the delivery date:

• Who maintains the system, and who can take over when that person is unavailable?
• What review and supervised learning time is included in the work?
• Which data, evaluation records and operating instructions transfer at handover?
• Who pays for the computing needed to investigate faults and keep the service running?

For an African team, these questions can make mentorship and maintenance visible in the scope. For a UK client, they create something specific to discuss with the supplier beyond a demonstration or a broad talent claim. The report does not establish that internal training guarantees retention, jobs or better model performance. Its findings support a more useful conversation: name the people, resources and handover arrangements that the proposed service actually needs. Explore more systems and businesses in PING's Innovation desk.