By Ayoub El Machatt
AI adoption in African businesses requires more than access to software. For companies whose employees already use generative AI, the next challenge is to turn individual experimentation into reliable workflows, with clear responsibility, appropriate data controls and measurable results.
When I meet a management team to discuss artificial intelligence, I rarely begin by asking which model they use. I ask which AI tools their employees are already using.
The answer from management is often shorter than the answer from the people doing the work. Marketing may be drafting copy in ChatGPT. A salesperson may be summarising a tender with Claude. Someone in finance may be testing a spreadsheet analysis through a personal account.
This is a pattern I encounter in my work with Moroccan companies: formal AI strategy often arrives after informal AI use. Management then has to decide how those experiments should fit into the organisation.
AI access and organisational readiness are different challenges
Access remains an important constraint. The World Bank’s Digital Progress and Trends Report 2025 identifies four foundations for AI: connectivity, computing capacity, context in the form of data, and competency through skills. The African Union’s Continental Artificial Intelligence Strategy also emphasises infrastructure, skills and responsible governance. [1] [2]
Training initiatives address part of that challenge. In February 2025, CGEM, the European Bank for Reconstruction and Development and LinkedIn announced AI Generation, an initiative targeting 1,000 Moroccan SMEs with access to online learning. The announcement offered licences valid until September 2025; it did not establish that 1,000 companies had completed an AI transformation. [3]
Inside a company, further decisions remain. Can employees upload a supplier contract? Who reviews an AI-generated sales proposal? What happens when an automated workflow produces an incorrect answer?
McKinsey’s 2026 global AI survey illustrates the distinction between individual and organisational results. Eighty percent of respondents reported improved personal productivity, while 37% said AI had contributed positively to their organisation’s earnings before interest and taxes, or EBIT. These are different measures, rather than a calculation of AI project failure. [4]
The survey is global, not an estimate of African adoption. Its relevance here is the management question it raises: how do individual gains become repeatable business results?
Make informal AI use visible
Shadow AI is the use of AI tools at work without organisational approval or oversight. In training sessions, I sometimes find that awareness of employee usage is much more developed than the rules governing it.
I do not interpret this primarily as employee misconduct. People are often trying to work faster. Problems arise when the organisation cannot see which accounts are being used, what information is being uploaded or which outputs are trusted.
ISACA’s 2026 AI Pulse Poll, drawing on more than 3,400 digital trust professionals worldwide, highlights gaps in governance, training and risk management alongside widespread AI use. Again, this is international evidence, rather than a representative survey of African SMEs. [5]
For an SME, I would begin by asking employees what they already use. Management can then establish a short list of approved tools, define information that must not enter unapproved systems, and specify which outputs require human review.
A practical policy should answer daily questions. “Use AI responsibly” does not tell an employee whether a client document can be uploaded or who must check a proposed response.
A Moroccan sales workflow: use AI where it adds value
One anonymised example from my work involved a sales team in Morocco and an initial pool of around 1,000 prospect profiles.
Applying enrichment and deeper AI analysis to every profile would have asked costly stages of the process to handle prospects that simpler rules could exclude. We instead organised qualification as a sequence:
- Start with approximately 1,000 profiles.
- Use basic filtering to retain around 150–200.
- Select approximately 50 for additional data enrichment.
- Apply deeper AI analysis to roughly 20.
- Retain human responsibility for final verification and qualification.
About 5% of the initial pool required enrichment, and around 2% required deeper AI analysis. These are approximate processing proportions, not measured cost savings or evidence of improved sales conversion.
The management change was deciding which steps could use simple rules, where AI added value and where a person still had to validate the result. The workflow became more selective and easier to control.
Design training around the work
In another Moroccan organisation, I was asked to support AI adoption across technical and commercial teams. Their recurring tasks, documents and potential use cases differed substantially.
A generic AI training programme would have missed those differences. During preparation, we mapped workflows, pain points, documents, existing tools and current AI practices by function. The intervention was then structured around those situations.
The immediate result was a shift from general AI awareness towards function-specific use cases and operating questions. No financial return had been rigorously measured at that stage, so I would not attach an ROI figure to the example.
Both cases reflect a broader lesson from my work: useful AI adoption begins with understanding the process people need to improve.
Give each AI workflow a named owner
A small company does not necessarily need an AI committee. It does need clear decision rights. I distinguish three responsibilities:
- Sponsor: approves the priority, budget and time.
- Operational owner: understands the workflow and manages its day-to-day implementation.
- Reviewer: checks sensitive outputs and controls what can be shared, published or acted upon.
One person may hold more than one role. The responsibilities should still be explicit.
Review should match the sensitivity of the task. Internal brainstorming may need light checking. Public marketing content requires editorial validation. Work involving client data, contractual commitments, financial decisions or health information calls for stricter approval and traceability.
Employees should know what information can be used, which tool can process it, who checks the result and which actions always require a human decision.
Measure results against the full cost
Licences, prompts and workshop attendance measure activity. They do not establish economic value.
Before introducing AI into a workflow, record a baseline. For example, a proposal-writing team could measure preparation time, correction time and acceptance rates. Customer-service teams could compare response times, errors and customer satisfaction.
Costs should include subscriptions, API usage, integration, training, human review and corrections. As an illustration, saving five hours of drafting while adding four hours of checking leaves a much smaller time benefit than the initial demonstration suggests.
The sales example also shows why processing fewer profiles is not enough to prove success. The business still needs to establish whether the selection produces useful prospects at an acceptable total cost.
I recommend agreeing on a stop condition before a pilot begins: what evidence would justify continuing, changing or ending it?
Build capability one workflow at a time
A distributor in Casablanca, a bank in Nairobi and a health-tech company operating across several African markets will have different requirements. Their governance should reflect their activities, resources and risks.
For a business beginning this transition, my approach is to select one useful workflow, document how it currently operates, assign responsibility, introduce appropriate controls, train the people involved and compare results with the baseline.
That experience can inform the next deployment. A growing collection of tools is less useful if nobody knows which processes improved or who is accountable when something goes wrong.
In the companies I work with, I would spend less time counting AI tools and more time examining changed workflows, data use, human review and measured results. That is how individual experimentation becomes an organisational capability.
About the author
Ayoub El Machatt leads Wave Digital Agency, working with organisations on digital marketing, acquisition and the practical integration of AI. He is the author of L’Hybride, a book examining strategy, marketing and generative AI. His contribution draws on his professional experience with management teams in Morocco.
Editorial note: The anonymised examples and approximate workflow figures were supplied by the author and have not been independently audited. The author disclosed using generative AI tools to assist research and confirmed responsibility for source verification, analysis and conclusions.
Sources and further reading
- World Bank. Digital Progress and Trends Report 2025: Strengthening AI Foundations.
- African Union. Continental Artificial Intelligence Strategy, 2024.
- CGEM. Launch of “AI Generation: Boosting 1000 Moroccan SMEs”, February 2025.
- McKinsey & Company. The state of AI in 2026: On the road to ROI, 25 August 2026.
- ISACA. 2026 AI Pulse Poll.