Africa’s AI Capacity Push Needs Workflow Proof, Not Training Counts

By Dr. Gleb Tsipursky

Africa is entering a new phase of artificial intelligence development.

In July 2026, the African Telecommunications Union (ATU) and the United Nations Office for Digital and Emerging Technologies (UN-ODET) announced a continent-wide collaboration aimed at strengthening AI capacity and digital public infrastructure across Africa.[1]

The initiative includes capacity building for policymakers, developers and public administrators, while also addressing responsible AI, digital identity, payments, data exchange, open-source ecosystems and technologies adapted to African languages and institutions.

That investment could help governments, businesses and developers create technologies that are more relevant to African needs.

But there is a danger in measuring progress through the easiest numbers: people trained, workshops delivered, certificates issued, software licences distributed and pilot programmes launched.

Those figures measure activity.

They do not necessarily measure whether AI improves actual work.

Productivity is not the same as transformation

Recent workplace research illustrates the distinction.

Gallup’s State of the Global Workplace 2026, drawing on U.S. employee data for this particular AI finding, reported that 65% of employees in organisations using AI said it had improved their productivity and efficiency, while only 12% strongly agreed that AI had transformed how work gets done across their organisation.[2]

That difference matters.

An employee may use AI to draft an email faster, summarise a document or prepare a presentation without the organisation itself becoming fundamentally more productive or capable.

Africa’s AI capacity-building programmes should learn from this distinction before large-scale training initiatives risk becoming certificate factories.

The objective should not simply be to teach people how to use AI.

It should be to demonstrate that they can use AI to improve a real workflow safely, repeatedly and measurably.

Every AI training programme should end with a real workplace project

Every publicly supported AI capacity programme should include a supervised workplace project.

Participants should apply an approved AI tool to a recurring task within a real organisational environment.

That might mean:

• analysing service requests;
• preparing a procurement summary;
• translating public information;
• identifying payment anomalies;
• improving inventory analysis;
• assisting with customer service;
• reviewing operational data; or
• helping a small business respond more efficiently to customers.

A named human reviewer should check the AI-generated output and remain accountable for the final result.

The point is not merely to prove that AI can perform a task.

The goal is to understand whether using AI actually makes the workflow better.

Four questions every AI project should answer

1. Did AI improve a real outcome?

Time saved matters, but it should not be the only measure.

Organisations should also examine:

• accuracy;
• service quality;
• turnaround time;
• customer experience;
• error rates;
• employee productivity;
• operating cost; and
• the amount of rework required after AI has produced an output.

An AI system that generates a report in two minutes rather than two hours creates little value if an employee then spends another two hours correcting it.

Productivity should therefore be measured across the whole workflow, not just the AI-assisted step.

2. Where did human judgement remain essential?

AI can draft, classify, summarise and identify patterns.

But people still need to determine whether an output is appropriate for the relevant legal, cultural, linguistic and institutional context.

This becomes particularly important when AI contributes to decisions involving:

• public benefits;
• healthcare;
• credit;
• recruitment;
• education;
• legal processes; or
• other high-impact activities.

An organisation should therefore document not only what AI performed successfully, but also where a human needed to intervene and why.

That information is valuable because it reveals which activities can responsibly be automated and which still depend on professional judgement.

3. What failed?

AI capacity-building programmes need protected mechanisms for reporting failure.

Participants should be encouraged to document:

• fabricated information;
• inaccurate responses;
• biased recommendations;
• privacy risks;
• cybersecurity concerns;
• weak local-language performance;
• inappropriate automated decisions; and
• unofficial workarounds employees develop when approved tools fail.

Concealing these problems creates what might be described as shadow AI: employees using tools and processes outside approved governance because the formal system does not meet their needs.

Hiding failures also prevents institutions from learning.

A failed AI pilot can therefore be valuable if it prevents a much larger and more expensive failure later.

4. Can the improvement be repeated?

A successful demonstration by one technically confident employee does not prove that an organisation can use the same workflow reliably at scale.

A ministry, bank, university, hospital or small business needs more than an impressive demonstration.

It needs:

• clear operating instructions;
• approved data practices;
• defined human responsibilities;
• escalation procedures;
• quality controls;
• monitoring;
• periodic review; and
• an understanding of when the AI system should not be used.

That is the difference between an AI experiment and an organisational capability.

The model should adapt to the organisation

Not every organisation needs the same type of AI implementation programme.

A national ministry may conduct a structured six-week pilot with formal risk assessment, data governance and management approval.

A small enterprise might test a single customer-service or inventory workflow for several days.

The scale can vary.

But the evidence should answer the same basic questions:

What changed?

What remained unreliable?

What value was created?

What risks appeared?

Who checked the result?

Africa needs AI capability, not simply AI users

This approach could strengthen Africa’s position in the global AI economy. As AfricaBusiness.com has previously examined in the South African context, successful AI adoption depends not only on access to models, but also on the strength of the infrastructure, skills, data, governance and systems supporting them.

The continent needs more than consumers who know how to prompt imported AI systems.

It needs professionals who can:

• evaluate AI systems;
• redesign workflows;
• measure outcomes;
• identify risks;
• manage data responsibly;
• understand when human judgement is necessary;
• adapt technology to local contexts; and
• build locally relevant solutions.

Workplace projects help create that capability because they connect technical knowledge with institutional reality.

Knowing how to use an AI application is increasingly becoming a basic digital skill.

Knowing where it creates value, where it fails and how to integrate it responsibly into an organisation is a considerably more valuable capability.

Measure time to competence, not just training completion

Governments, universities, companies and development partners naturally track the number of people who complete AI training.

But completion should be only the beginning.

A more useful indicator would be time to demonstrated competence.

For example:

How long does it take a participant to use AI effectively in a real workflow without requiring extensive correction or intervention?

That measure shifts attention away from attendance and towards capability.

The same principle can be applied at organisational level.

Instead of asking:

How many employees have received AI training?

leaders should ask:

How many business processes have been measurably improved through responsible AI use?

Every major AI programme should publish an outcome scorecard

Governments and funders should consider requiring major AI capacity-building programmes to publish outcome scorecards.

These should report:

• which workflows participants tested;
• what baseline performance looked like;
• what changed after AI was introduced;
• how much time or cost was saved;
• whether accuracy improved or deteriorated;
• what problems emerged;
• where human intervention remained necessary;
• what safeguards proved effective; and
• whether the improvement could be repeated.

Aggregated findings could help countries and institutions compare lessons without exposing confidential organisational or personal data.

Over time, this could create an increasingly useful evidence base showing where AI actually works in African operating environments.

Negative results should also be rewarded

Funders should not expect every pilot to become a success story.

A project that demonstrates that an AI system performs poorly in a local language, creates unacceptable privacy risks or produces unreliable outputs may have generated extremely valuable information.

Discovering those weaknesses during a controlled pilot can prevent expensive deployment failures.

If programme managers believe funding or reputation depends on every project being presented positively, they have an incentive to hide precisely the evidence that responsible AI adoption requires.

Successful AI policy therefore needs a culture in which well-documented failure is treated as learning rather than embarrassment.

Africa’s diversity makes workflow testing essential

Africa is not a single AI operating environment.

A system performing successfully in one country, language, regulatory environment or infrastructure setting may perform differently elsewhere. This is also why debates around building sovereign AI capacity in Africa increasingly focus on local data, governance, infrastructure and the ability to adapt technology to national and regional conditions.

Language alone creates substantial variation.

So do:

• connectivity;
• electricity reliability;
• institutional capacity;
• data availability;
• regulatory frameworks;
• digital literacy; and
• local business practices.

Capacity-building programmes therefore need feedback from the places where people actually use AI.

A model that performs well in an international benchmark may still struggle with the vocabulary of a local government office, the accents encountered in a contact centre, the documentation used by an SME or the linguistic complexity of a healthcare consultation.

Infrastructure also matters. AfricaBusiness.com has previously examined how reliable and lower-carbon energy is becoming critical infrastructure for Africa’s digital future as AI, data centres and digital services expand.

This is why workplace evidence matters.

From certificate counts to institutional capability

Africa’s AI capacity push represents an important opportunity.

Training people to understand and use AI is necessary, but training counts alone cannot demonstrate successful digital transformation.

The stronger test is whether organisations can show, workflow by workflow:

where AI created value;

where it introduced risk;

where human judgement remained necessary;

what failed;

and

whether the improvement could be repeated safely at scale.

The strongest AI strategy will not produce the largest pile of certificates.

It will produce institutions that know how to use AI responsibly, measure what it changes and explain which human remains accountable for the result.

About the Author

Dr. Gleb Tsipursky is a behavioural scientist, CEO of Disaster Avoidance Experts and author of The Psychology of AI Adoption at Work: From Resistance to Results, published by Georgetown University Press in 2026.

Sources and Information

[1] United Nations Office for Digital and Emerging Technologies / African Telecommunications Union. African Telecommunications Union and UN-ODET Collaborate on AI Capacity and Digital Public Infrastructure, 3 July 2026.

[2] Gallup. State of the Global Workplace 2026. The AI workplace figures cited in this article refer to U.S. employee data.

Image credit: AI-generated