Everyone has the tools. Almost no one has the focus.

The hard part of AI is no longer getting hold of it. Capability is everywhere — more models, platforms and vendors than any leadership team can evaluate, arriving faster than anyone can absorb. The constraint has moved. It isn't what AI can do. It's deciding what to point it at.

Most programmes answer that question by spreading thin. A pilot in every function. A tool for every team. A working group, a licence, a proof of concept somewhere in procurement. It reads as progress because it looks like activity — but motion gets mistaken for momentum. Twelve months in there are forty experiments, three of which anyone can still name, and no movement on the numbers the business is actually run against.

This is a structural pattern, not a failure of effort. When capability is abundant and cheap, the path of least resistance is to acquire more of it. Every demo is persuasive. Every function has a case. Nothing forces a choice, so no choice gets made, and the spend fans out across everything at once — which often ends up largely the same as pointing it at nothing.

The firms that capture value do the opposite, and it's almost boringly simple to state: they pick one metric that matters, and they move it.

One meaningful metric. Not a scorecard, not a set of objectives cascaded across the org. One number that would change the shape of the business if it moved — the cost to serve a customer, the time from first enquiry to signed contract, the share of revenue that renews. Everything else gets framed through it. Technology earns its place by whether it moves the metric. Process gets redesigned where it constrains the metric. People are trained on the work that moves the metric. The question stops being 'where could we use AI?' and becomes 'what moves this number, and what does AI change about that?'

From the metric, you map outward. The number has drivers; the drivers have inputs; those inputs connect to systems, teams and decisions that no one had previously drawn on the same page. That map — the metric at the root, expanding to everything that moves it — is what keeps single-minded focus from collapsing into a narrow point solution. Focus stays singular. The map keeps it whole. You are not automating one task in a corner. You are moving one number by redesigning everything that touches it.

This is the part most efficiency drives miss. Focus without the map produces a point solution: a faster version of one task, bolted onto a process that still runs the way it always did. The map without focus produces the sprawl we started with: forty pilots, no movement. You need both — one point of focus, and a full picture of what feeds it.

And moving the number isn't a technology exercise. It runs across five layers — vision, governance, technology, culture, process — and efforts that touch only one of them stall. The metric twitches, then settles back. That is the usual fate of the tool-first pilot: a real gain in a demo, gone by the next quarter, because nothing around it changed.

None of this makes the work simple. Realising value from AI is genuinely complex. What makes the complexity tractable isn't a bigger programme or a longer tool list. It's the refusal to chase all of it at once.

Everyone has the tools now. That was the hard part for about a decade; it isn't any more. The scarce thing — the thing that separates the few programmes that pay for themselves from the many that quietly don't — is the discipline to choose what matters and leave the rest.

Capability is settled. Focus is the work.