Ghanaian managers have adopted artificial intelligence far faster than their organisations have learned to govern it. New research from Ashesi University, drawing on a survey of 162 managers and 15 interviews, shows what that gap is already costing, and what leaders can do about it before regulation forces their hand.
Why we went looking
Ask a manager in Accra whether they use Artificial Intelligence (AI) at work, and the answer is almost always yes. Ask what their organisation’s rules are for using it, and the conversation usually stops.
That gap is what prompted this study. Over the past three years, AI has moved from novelty to routine in Ghanaian offices faster than any technology before it. Reports told us adoption was rising. Nobody could tell us what managers were actually doing with it, which tools, for which decisions, with what safeguards.
So, we asked them. Between February and May 2026, we interviewed 15 senior managers across nine sectors, then built a survey from what they told us and sent it to 162 managers. More than half are chief executives, founders or senior executives, and two-thirds are people managers with over a decade of experience.
It seems this is the first study of its kind in Ghana, and its findings are partly uncomfortable.
What we found
Adoption is ubiquitous, and nobody ordered it. Some 83% of managers use AI weekly or more often, and only 1 respondent out of 162 has never used it. However, 66% started out of personal curiosity, not because an employer asked them to, and16% began because their organisation required it. Hence, the transition/adoption was led from below.
It is not a side tool. More than half (54%), say their last important piece of work was partly AI-generated, and 42% had consulted AI before an important business decision within the previous week.
Managers treat AI like a colleague, not a calculator. Nearly 8 in 10 describe it as a thinking partner or a capable assistant; only 9% call it strictly a piece of software. That matters more than it sounds. We extend trust to colleagues that we do not extend to software tools.
The work has changed, leading has not. Asked how AI is changing leadership, 7 in 10 chose one of two positions: that AI is a powerful tool and what good leadership requires has not fundamentally changed, or that AI changes some things while the human core of leading people remains. One executive summed up the mood:
I do not know about change, but enhancement. AI has shaped my approach to daily activities. But it has not changed my leadership style. Not yet.”
— Senior Manager, energy sector
Something is happening to “thinking”. While this finding was not altogether surprising, the numbers were higher than we expected. A clear majority, 61%, say their team members are producing work that looks more polished but shows less original thinking than before. Nearly half say colleagues are avoiding the hard analysis altogether and letting the AI tool do it.
Some of my team members have totally outsourced the thinking process. The reports look beautiful, but I can tell, there is nothing original in them.”
— Chief executive, energy sector
The part that should worry boards
Now the uncomfortable part. We asked managers whether their organisation had six basic safeguards: a written policy on approved tools, rules on what data may go into them, a named person accountable, mandatory human review before AI-assisted work goes out, training before access, and guidance on which tasks AI may be used for.
On every single one, a majority said “No”.
Scored together, the average organisation has roughly a third of the governance it needs, and 6 in 10 sit at the bottom of the scale. Only 10% report a policy that genuinely changes what people do. And 35% say plainly there is nothing at all. Each person decides for themselves.
We don’t have the systems in place yet, but we’re using the tools every day.”
— Senior manager, civil society sector
The consequence is measurable. Almost a third of these managers (32%) have put confidential organisational data into a public AI tool because a deadline was tight and no approved alternative existed. Remember who is answering here: half this sample sits in the C-suite. Because people under-report behaviour their employer would forbid, the real figure is almost certainly higher.
Two points of context matter here. The gap we measured is specific to AI. Many of these organisations already have data security programmes under way; what has not caught up are the controls built for AI in particular. We also did not benchmark the sample against regional or global comparators, and nothing in the data marks it out as an outlier. These findings reflect a global AI governance gap, driven by the sheer pace of AI adoption, rather than a regional deficiency.
Responses to two questions stood out. The highest-rated statement in our entire survey was not about productivity or jobs: firstly, 54% of Ghanaian managers are troubled that foreign companies control the tools they depend on with no accountability to Ghana or Africa, and secondly, 48% have begun deliberately limiting their own AI use, not out of technophobia, but to stop their judgement from weakening.
What should leaders do as their first, next step?
The findings point to a clear need for action. Based on what we heard and saw in the data, we suggest five practical steps that organisations can start with immediately.
- Write the one-page list. Which AI tools are approved, and what must never go into them (except where a tool is configured for secure handling of that data): client information, staff records, board papers, and unpublished financials. One page. Circulate it this week. This is the single measure that addresses the problem of confidential organisational data being entered into a public AI tool, because it replaces a prohibition with an answer.
- Make human review non-negotiable. Nothing AI-assisted leaves the organisation without a named person having properly read it. No exemption for urgency; urgency is precisely when the mistakes happen. This is currently the weakest of the six safeguards.
- Train people before you give them access. Most organisations that train at all teach staff how to write prompts. Teach them how to check the answers as well. When people are using tools that nobody in the building can interrogate, the ability to recognise a wrong answer becomes one of your most important safeguards.
- Name one owner. One person, by name, responsible for AI questions, policy updates and incidents. Not “I.T.”, and not “management”. A person. Half of organisations cannot currently name one.
- Protect the moment of judgement. For decisions that matter, form your own view first and write it down, then consult the tool. It costs nothing, and it addresses the strongest risk pattern we found. Half of your managers are already improvising some version of this alone, and so legitimize and shape the habit.
There is also a bigger opportunity. 70% believe that the most successful leaders in 2030 will stand out not by using AI best, but by doing what AI cannot: building trust, exercising judgement and taking responsibility.
That is the human advantage worth protecting.
Note on the use of AI tools: The team used Claude, ChatGPT, and Gemini to assist with research, analysis, editing, and brainstorming while retaining full responsibility for the final work.
The authors are the research team behind Leading with AI in Ghana, a study supported by Ashesi’s Center for Executive Education. Esi E. Ansah is Executive Director of the Ashesi Center for Executive Education. Prince A. Acquaye lectures in the Law Department at Ashesi University. Kofi Dadzie is a technology governance consultant and board director. Dirk Kleine is an AI expert and professor from Germany. Ayorkor Korsah is Head of the Department of Computer Science and Information Systems at Ashesi University, and Ishmael Asiedu is a research assistant on the study.
