A study of 502 senior enterprise AI decision-makers published on 12 November found that only 33 per cent had met or exceeded the business case for their most significant AI programme. Ninety-six per cent reported significant governance gaps, and 78 per cent said governance failures were a bigger threat to scaling AI than technology failures.
That is the gap worth paying attention to. Adoption is no longer the hard part. Running the thing properly is.
The Numbers Behind the Gap
What Actually Changed
Three things from the last few weeks, with dates, because a lot of AI commentary blurs a six-month-old announcement into this morning's news.
12 November — the value gap got measured. HFS Research and Rackspace published findings from 502 enterprise AI decision-makers at companies above $100 million in revenue, surveyed in August and September. Only a third met their business case. The detail that matters most: 44 per cent had no single consistent approach to running AI in production, and the study attributes the shortfall to fragmented accountability rather than weak technology.
20 October — Dubai put skills behind the mandate. Dubai Chamber of Digital Economy signed agreements with Nebius Academy and Lovable to build agentic AI skills across 14,000 member companies of its Business Groups and Business Councils, with training delivered through Dubai Chambers Academy. It sits under the Dubai Economic Agenda D33 and follows the Crown Prince's initiative to move Dubai's private sector to agentic AI within two years.
1 October — the UAE adoption picture. "Unlocking the UAE's AI Potential 2026", from Strand Partners with AWS and the UAE Artificial Intelligence Office, found 72 per cent of UAE businesses now use AI, up from 53 per cent a year earlier. But only 31 per cent of adopters deploy what the report classes as advanced AI, and 48 per cent name skills shortages as the main obstacle — ahead of money, at 31 per cent.
The Pattern Underneath
Read together, those three say the same thing from different angles.
Almost everyone has adopted AI. Very few have operationalised it. The organisations getting value are not the ones with better tools — they are the ones who decided who is accountable, what the rules are, and how they would know whether it is working.
One figure in the HFS study makes this concrete. Where AI was embedded in standard operations, 56 per cent met their business case. Where it was running across several functions without that grounding, 28 per cent did. Same technology, twice the success rate, depending entirely on whether it had been made part of how the company actually runs.
Where AI is embedded in standard operations, 56 per cent met their business case. Scattered across functions without it, 28 per cent did.
And the thing holding UAE companies back is not budget. It is that the people using these tools have not been taught to use them well — which is a solvable problem, and a cheaper one than most of what is being spent on platforms.
The AI Readiness Checklist
Ten questions. If you cannot answer one of them in a sentence, that is where your next month goes. Most companies find they fail between four and six on a first pass, and that is normal.
1. Do you know what AI is already being used here?
Not what you approved — what people are actually using. Staff adopted these tools individually long before most companies had a policy. Ask the question plainly and without consequences, or you will get a clean answer that is not true.
2. Is one named person accountable?
Not a committee and not "IT". The HFS study puts fragmented accountability at the centre of the value gap. If three people are responsible, nobody is.
3. Is it written down what must never go in?
Client data, salary information, unreleased financials, anything under NDA. One line each, in words a new joiner understands. This is the cheapest item on the list and the most commonly skipped.
4. Which decisions must a human make?
Hiring, firing, credit, pricing, anything with a legal or safety consequence. Write the list before someone has to defend an automated decision they cannot explain.
5. Can anyone tell whether it is working?
If you did not record how long the task took before, you cannot show it got faster. Most AI business cases fail here — not because the gains were absent, but because nobody captured a baseline.
6. What happens when it is wrong?
Who notices, who fixes it, who tells the client. Models are confidently wrong about numbers, dates and citations, so the question is not whether this happens but whether you find out before the customer does.
7. Have the people using it been taught how?
Forty-eight per cent of UAE businesses say skills are the main obstacle. Access is not training. Most people are using these tools at a fraction of their capability and do not know it, because nothing tells you what you are missing.
8. Is it in the actual workflow, or beside it?
This is the 56-versus-28 question. A tool people must remember to open is a pilot. A tool inside the process they already follow is operational.
9. Do you know what your vendors do with your data?
Including the AI features quietly added to software you already pay for. Your obligations to your clients do not stop at your own systems.
10. When do you look at this again?
A date in the calendar. This field moves faster than annual policy cycles, and an AI policy written once and filed is a policy about last year's tools.
What to Do With a Bad Score
Do not start a programme. The organisations that get this wrong almost always respond by buying a platform, and platforms do not supply accountability, rules or judgement.
Items 1, 2, 3 and 4 can be done in a fortnight by one person with authority, and they cost nothing. They are also the four that most reduce your exposure. Item 5 takes a week of recording how long things currently take, and it is what makes everything afterwards arguable with a finance director.
Item 7 is where training genuinely helps, and it is worth being specific about what kind. The parts that pay off are the ones that do not expire with the next model release: how to brief a model properly, how to recognise when its answer is wrong, and what should never be handed to it at all.
Where to Start
Our AI courses in the Skills Hub cover exactly those three things, in twenty to thirty minutes each. Prompt Engineering for briefing properly. AI for Everyday Work for the tasks it is genuinely good at and the ones where it quietly wastes your time. Using AI Responsibly for the decisions it should never be left to make — which is checklist items 3, 4 and 6 in practice.
For a whole team, or for the governance side specifically, our AI for Business programme runs three days on your own material and your own processes.
The companies pulling ahead right now are not the ones with the newest tools. They are the ones who can answer those ten questions.
Sources
- HFS Research and Rackspace — enterprise AI value study, published 12 November 2026
- Dubai Chamber of Digital Economy agentic AI agreements, Fast Company Middle East, 20 October 2026
- Unlocking the UAE's AI Potential 2026 — Strand Partners, AWS and the UAE AI Office, reported 1 October 2026
- UAE Artificial Intelligence and Data Authority, Al Tamimi & Company — background on the regulatory picture
Three AI Courses, Free While the Hub Is New
Prompt Engineering, AI for Everyday Work, and Using AI Responsibly. Twenty to thirty minutes each, something to do in every lesson, and a certificate at the end. Built around the judgement that survives the next model release.
Start with the AI Courses