Insights
What 65 executives taught me about AI training in a power company
Ahmed Amrousy trained 65 executives at Kahraba, Egypt's largest private-sector power generation and distribution company, across three runs of the same 16-hour workshop. It was rated 4.8 out of 5, and the CEO and nine senior managers later booked a separate advanced session. These are the five things the engagement taught him about corporate AI training — including why the data boundary has to come before the first exercise rather than after it, and what running the same workshop three times revealed that running it once would not have.
Last updated:
Kahraba is Egypt’s largest private-sector power generation and distribution company. When the engagement began, I expected the hard part to be the technology. It was not.
Sixty-five executives later — across three runs of the same 16-hour workshop, rated 4.8 out of 5 — these are the five things I would tell anyone planning corporate AI training in a serious organisation.
1. Run it more than once
The conventional advice — mine included — is to start at the top. Kahraba happened the other way round, and it taught me something I did not expect.
We ran the same 16-hour workshop three times, 20 to 23 executives per run. The first run is where you discover which exercises actually land. By the third, the weak sections had been cut and the strong ones extended. A format delivered once is a hypothesis; delivered three times to comparable groups, it is evidence.
And the leadership session came after — the CEO and nine senior managers booking eight hours on delegated AI work once the organisation had already proved it could use the tools. That order has a quiet advantage: the leadership conversation was about extending something real, not authorising something speculative. I would still usually argue for leadership first. But a pilot that earns the leadership session is a legitimate second route, and this engagement is why I now say so.
2. The data boundary comes before the first exercise, not after it
In a power company, a large share of what an executive handles daily cannot go into a public AI tool under any circumstances. My instinct beforehand was that this would constrain the training.
The opposite turned out to be true — provided the boundary conversation came first.
We spent the opening session agreeing explicitly what could and could not leave the organisation. Once that was settled and written down, the room could work on genuinely real material for the rest of the programme. Had we left the boundary to the end, every exercise would have run on sanitised examples, and sanitised examples are why most corporate training is forgotten by Friday.
The boundary is not a restriction on the training. It is the precondition for the training being real.
3. Seniority changes the question, not just the pace
Early on I assumed the executive session would be a compressed version of the team session. That was wrong, and I changed it.
Senior people are not primarily asking how do I use this. They are asking:
- Which of these investments is real and which is theatre?
- What do I demand from the people who report to me?
- Where does accountability sit when an AI-assisted decision turns out badly?
- What do I stop funding?
Those are not advanced versions of the practitioner questions. They are different questions, and they need different exercises — which is why Menova now runs AI for Executives and AI for Business as genuinely separate programmes rather than long and short versions of one.
4. Scepticism in the room is an asset
Engineers in a power company are professionally sceptical, and they should be. Several participants arrived expecting to be sold something.
The most productive move was to lead with the failure modes — to demonstrate, early and deliberately, a confident wrong answer, and to explain why it happened. Establishing that the instructor will name the limitations bought more credibility than any capability demonstration could have.
If a room is not sceptical, that is the harder audience. Uncritical enthusiasm produces the unverified output that damages an organisation later.
5. What people rate is what they built
A 4.8 out of 5 was not awarded for the content. It was awarded because participants spent most of the programme rebuilding their own recurring work, on their own documents, and left holding a written plan for their own area.
That is the whole design principle: 70% hands-on, and a demonstration every ninety minutes that genuinely surprises the room — a task that normally takes an afternoon, completed and verified in four minutes, on the organisation’s own data.
Nobody rates a lecture 4.8. They rate the afternoon they got back.
The pattern, generalised
Run it more than once. Boundary before exercises. Separate programmes for separate seniorities. Lead with the failure modes. Make people build.
None of that is specific to energy. It is what I would now recommend to any organisation of any size that wants AI training to survive contact with the following Monday.
Ahmed Amrousy is the founder of Menova, Executive Education Instructor at the American University in Cairo’s Onsi Sawiris School of Business, and Head of Marketing & PR at Nile Air. The full Kahraba case study is linked below.