There is a board paper on your screen titled “Our AI Strategy”. Twelve slides. A few use cases, a competitive line, an investment ask, a tasteful diagram. The board read it, nodded, and approved it. Three months on, you are looking at the same deck and you cannot point to a single decision that is now made differently because of it. The document was fine. It just never told anyone what to do.
That gap is where a lot of AI mandates come undone. The slide gets approved, everyone feels progress has been made, and the business runs on Monday exactly as it ran the Friday before. If you have been handed the job of getting AI into the business, your instinct that a strategy will not be enough is correct. What the business needs is an operating model, and it helps to be able to name the difference out loud.
What is an AI operating model?
An AI operating model is the working structure that decides how AI actually changes the business day to day. It sets out which decisions move from human to AI-assisted and which stay human, how a decision made at speed remains explainable, who owns each AI capability, and the rhythm on which the board reviews it. A strategy describes the destination. The operating model is what people pick up and act on each morning.
The distinction matters because the two are easy to confuse. A strategy can be excellent and still inert, because describing where you want to go is a different act from deciding how the work changes to get there. Practitioner analysis from Tribe AI frames the operating model as how an organisation structures, governs, and deploys AI in daily operation, rather than as the document that announces an intention to do so.
Why doesn’t the strategy slide change anything on its own?
The strategy slide changes nothing because a use-case catalogue is a list of wishes, and a wish is not an instruction. The deck says the business would like AI to speed up quoting and tidy the data. It rarely says who now decides differently, what oversight applies, or who is accountable. Without those answers, nobody acts, and the deck becomes a record of good intentions.
This is not a rare failure. Research from MIT found that roughly 95% of AI pilots show no measurable profit-and-loss impact, and a common reason is precisely this gap. The tool gets bought, the slide gets approved, and the operating model that would have turned either into a changed decision never gets built. Adoption without that structure is how an initiative joins the failure majority rather than the small group that shows a return.
BCG’s research adds useful context. For around a third of companies, AI is still not a top-three priority, and roughly 15% of chief executives sit in what BCG calls the “Followers” group, holding AI to pilots and small-scale improvements. Those organisations often produce strategy documents that read well and then stall, because the document was never the thing that was going to move the business.
Where will you actually meet the operating model?
You meet the operating model in five dimensions that mature together, drawn from Prolego’s AI operating-model framework. They are capabilities, the AI that solves real problems; infrastructure, the plumbing to run it safely; talent, the people who maintain it; governance, the oversight that keeps it honest; and tools, the third-party resources that speed things up. The point is to advance all five a little, not to perfect one.
Underneath those five sit four practical questions the model has to answer. Decision rights settle which calls move to AI-assisted and which stay with a human. Oversight settles how a decision made at speed stays explainable after the fact, which the NIST AI Risk Management Framework structures through its govern, map, measure, and manage functions. Ownership names who runs each capability rather than leaving it orphaned. Cadence sets the review rhythm.
That cadence is worth pinning down early, because it is where the board’s attention lives. A quarterly review against the five dimensions, supported by the weekly and monthly rhythms a team needs to stay aligned, gives everyone the right meeting for the right purpose. The CorpDev guidance on meeting cadence puts it plainly, structured rhythms keep a team aligned and prevent the drift that comes when an emerging-technology effort has no fixed place on the calendar.
When should you push back on the brief, and how?
Push back when the board has asked for a strategy document and you can see that a document alone will leave the business unchanged. The moment to do it is before you start writing, not after you have produced a deck that disappoints. The trick is to reframe rather than refuse, because a flat “this won’t work” lands as resistance, whereas a better-shaped proposal lands as competence.
The reframe that works is organisational change rather than technology. Bain & Company put the underlying point well, AI is not a technology upgrade, it rewires how the enterprise creates value and how work gets done. Prolego make the same observation from the funding side. Executives rarely win large board backing by talking about technology, but they regularly win it by talking about how the business itself will change. Lead with the change, and the resources tend to follow.
In practice that means proposing a small number of projects that move all five dimensions at once, then measuring progress against those dimensions rather than promising a single ambitious return by a fixed date. You think big and start small. The board hears a credible plan to mature the business, you avoid committing to a number you cannot yet defend, and the work actually changes how something gets decided. That is the whole game.
Related concepts worth knowing
A few ideas sit close to this one and are worth holding in view. Decision rights, the question of which calls move to AI-assisted and which stay human, is the spine of any honest operating model. Oversight and explainability, how a fast decision stays defensible later, is the governance dimension in working clothes. Cadence, the review rhythm, is what keeps the model alive rather than letting it ossify into another slide.
The operating-model thinking also feeds directly into what comes next for the delegate. It becomes a prioritised, defensible roadmap when you turn it into a sixty-day plan, and it becomes a test of nerve when you have to explain all of it out loud in a first board update. Get the operating model clear in your own head now, and both of those later tasks become a great deal easier. If you want a second pair of eyes on the structure before you take it upward, book a conversation.
