I have had a version of the same conversation more times than I can count. A leader — usually a CDO or the head of transformation at a large enterprise — walks me through their AI program. Pilots across three or four departments. A governance framework. A vendor shortlist. A "Center of Excellence" that was stood up six months ago. They have a number: hours saved, headcount reduction, cost avoided. The number is real. The savings are real. The work to get there was genuine.
And then they ask me whether they are on track.
I ask one question: "Which decisions in your organization are made differently today because of this program?"
The pause that follows is not ignorance. It is recognition. They know, in that moment, that the number they have been tracking is not the wrong number for what they have built — it is the wrong number for what they said they were building.
What they have built is an adoption program. What they told the board they were building was an AI strategy. The confusion is understandable. It is also consequential.
What Adoption Actually Is
AI adoption is the work of deploying AI tools and AI-augmented workflows across an organization. It is real work. It is necessary work. It frequently delivers measurable, legitimate value.
Adoption asks: where can we introduce AI capability into existing processes to make them faster, cheaper, or less error-prone?
The output of adoption is efficiency. And in the MENA context specifically, that efficiency has a compelling surface story: organizations that are simultaneously managing digital transformation and AI implementation can show boards and ministry stakeholders a continuous improvement curve. Headcount is optimized. Processing time drops. Manual document review that used to take four days takes four hours. These are not fabrications — they are real operational wins.
The problem is not that adoption is wrong. The problem is that adoption is a floor, not a ceiling. And when adoption is declared "AI strategy," the organization stops asking the harder questions. Capital is allocated. Progress is reported. Momentum becomes its own justification.
Here is what adoption cannot do on its own: it cannot change the competitive or institutional position of the organization. It makes the existing structure run faster. That is valuable — until you realize you are optimizing a structure that may not deserve to exist in its current form, while a competitor or peer institution is quietly making structurally different choices about where intelligence operates in their organization.
In my work across MENA organizations, I have watched genuinely well-resourced AI programs achieve significant efficiency gains and then discover, 18 months in, that their market position has not changed. In some cases it has weakened — because while they were optimizing internal processes, others were making decisions that changed the nature of the game.
Adoption is necessary. It is not sufficient. And mistaking it for strategy is an expensive error.
What Strategy Actually Is
AI strategy is a different class of decision entirely.
Adoption asks where AI can improve existing processes. Strategy asks which decisions should be restructured — who makes them, with what information, at what speed, and with what degree of machine versus human judgment — in order to change the organization's competitive or institutional position.
That is not a refinement of the adoption question. It is a different question.
Strategy operates at the decision layer. It is not asking "how do we do this faster?" It is asking "should this still be a human decision at all, and if so, what does that human need to be able to decide well?" It is asking "which processes do we currently run that exist to compensate for information limitations that AI can now eliminate?" It is asking "where is our slowness itself the vulnerability, and what would it mean to remove it?"
When strategy is working, the output is not hours saved. The output is decisions that were previously impossible becoming routine, or decisions that previously required senior judgment becoming delegable, or decisions that previously took weeks happening in hours — with accountability structures that match.
In practice, this means AI strategy often discovers that some of the highest-leverage opportunities are not in the obvious places. The customer service automation initiative may be well-conceived. But the real advantage might be in restructuring how credit decisions are made in a financial institution, or how infrastructure priorities are sequenced in a government body, or how a logistics company routes capacity across a region in real time. These are not process improvements. They are changes to what the organization can do — and how fast it can do it.
This distinction also tells you where AI should not operate. Strategy includes explicit decisions about where human judgment must be preserved — not because AI cannot help, but because the legitimacy, accountability, or trust requirements of the decision demand it. Adoption rarely has that conversation.
The evidence that adoption is outrunning strategy
This is not only my observation from advisory work. The ninth annual MIT Sloan Management Review / BCG global study, surveying 2,102 executives across 21 industries and 116 countries, found organizations deploying agentic AI "well before they have a strategy in place." The adoption curve it documents is the reason: traditional AI took roughly eight years to reach 72% adoption, generative AI reached 70% in three, and agentic AI reached 35% in about two — with a further 44% planning to deploy. Capability is arriving faster than the deliberation that should govern it, which is precisely the failure mode this article describes.
The scale of the adoption side is not in dispute either. Stanford HAI's AI Index records 78% of organizations reporting AI use in 2024, up from 55% the year before. Set that against the return figures — only around a quarter of AI initiatives delivering the ROI expected of them — and the gap is not a technology gap. Nearly everyone has adopted. Far fewer decided what they were adopting it for.
Two honest caveats. The MIT SMR study is co-produced with a consultancy that sells AI transformation work, so it has a commercial interest in the conclusion that you need a strategy; and the AI Index is a compilation of other organizations' surveys rather than primary fieldwork, so the underlying denominator for that 78% is not stated on the page itself. I cite them because they are the best available measurements of a pattern, not because a survey settles an argument. The direction they point is consistent, and it matches what I see in the room.
Three Questions
These are diagnostic questions. They are not rhetorical. Apply them to your own program today.
Question one: Is your primary ROI metric about time, cost, or headcount?
If yes — you have built something real, and you should account for it honestly. But understand that this is an efficiency metric, not a strategic metric. Efficiency metrics tell you that AI is working inside your existing structure. They do not tell you whether your structure is correct. A strategic AI program measures things like: speed to decision on a class of decisions, quality of decisions made with AI-augmented information versus without, organizational capabilities that now exist that did not exist before.
Question two: Could you describe, in one sentence, which decisions are made differently in your organization today because of your AI program?
Not "which tasks" — which decisions. If you cannot answer this cleanly, your program has not yet reached the strategic layer. That is not a failing — it is a precise diagnosis of where you are. Many organizations need a phase of adoption before they can identify the decision-layer opportunities. The failure is claiming you are already there.
Question three: Does your AI roadmap include explicit choices about where AI will not be used?
A strategy without constraints is not a strategy — it is a preference list. Real AI strategy includes deliberate decisions about where human judgment is preserved, protected, or elevated. Where you will not automate. Where the organizational or societal accountability requirements exceed what AI-assisted decision-making can legitimately carry. If your roadmap has no such exclusions, it has not been thought through at the level of strategy.
What Changes When You Make This Distinction
Making this distinction does not require scrapping what you have built. It requires reorienting what you are building toward.
Organizations that successfully navigate this shift typically do something specific: they preserve and continue their adoption programs — because the efficiency gains are real and fund further investment — while simultaneously standing up a separate strategic analysis process. This process does not ask "where can we deploy AI?" It asks "which decisions, restructured, would change our position?" The answers are almost never the same as what adoption discovers on its own.
In the MENA enterprise and government-adjacent context, this reorientation carries additional weight. The organizations that will matter in five years are not the ones that automated the most processes. They are the ones that made structurally different choices about how intelligence operates inside their institutions — and had the clarity to make those choices deliberately, not accidentally.
The confusion between adoption and strategy is not a technology problem. It is a thinking problem. And it is solvable, but only if you are willing to ask whether the metric you have been celebrating is measuring the right thing.
That question is uncomfortable. It is also the most important question your AI program has not yet asked.
Continue Reading
- Measuring AI ROI: Why You Are Using the Wrong Metrics — Why efficiency metrics are necessary but insufficient, and what capability metrics reveal instead.
- The AI Readiness Gap in MENA Enterprises — An in-depth analysis of the five barriers to AI adoption in the Middle East and North Africa.
- Why AI Strategy Matters More Than AI Tools — The foundational argument for strategy-first AI investment.
- Large Language Models Explained for Executives — What LLMs can and cannot do, explained without the technical noise.
