Tools aren't enough. Capability wins.

Most companies are treating AI as a better tool. Faster drafting, a copilot in the inbox, a note-taker in every meeting, a chatbot bolted to the website. Each one is real. Each saves a little time. Together they add up to far less than anyone expected — because a tool makes the existing work faster, and the existing work was never the opportunity.

This is the difference between two ways of thinking, and it decides the return before a single tool is bought.

A tool slots into a process you already run. It makes one step cheaper or quicker and leaves the shape of the work intact. That is the ceiling of tool-thinking: incremental efficiency inside a cost base that never changes. Same roles, same hand-offs, same bottlenecks, done slightly faster. The org chart doesn't move. Six months on, the copilots are still running and no one can find the gain in the numbers.

A capability is different in kind, not degree. It changes what the work can be. When a task that used to take a skilled person two days — reading everything, synthesising it, drafting the options — can be done in an afternoon under that person's direction, you haven't sped up the old process. You've changed what the person is for. The work worth designing isn't 'the same report, faster'. It's 'what does this function do now that the slow, expensive part is cheap and quick?'

Treat that as a tool and you bolt it onto the process you already had. The draft comes out faster, then waits three days in the same approval queue it always did. The bottleneck was never the drafting. This is why bolting capability onto an unchanged process under-delivers, and it's structural rather than bad luck: the constraint sits somewhere the tool never touched, so the output of the system as a whole barely moves.

The evidence is blunt about this. Around 88% of companies now use AI in some form; roughly 6% are getting meaningful money out of it. That gap isn't explained by which tools they bought — everyone has broadly the same models. Of twenty-five factors McKinsey tested against profit impact, the one that mattered most was redesigning the work itself, and the firms seeing real returns are about three times more likely to have done it. The money isn't in adopting AI. It's in rebuilding the work for what the capability now makes possible.

So the first question is not which tool to buy. It's what to point the capability at. And the answer isn't a function or a use case — it's a number. You start from the one metric that matters, the number that would change the business if it moved, and you ask what the work would look like if it were designed from scratch to move it, given what AI can now do. The capability sits at the heart of that design, not patched onto the edge of the old one.

That is the whole reposition. Not 'add AI to what you already do'. Redesign what you do with the capability at its heart, pointed at the one thing you've decided matters. The efficiency gains still come — but they arrive as a by-product of a better-designed system, not as the point of it.

A tool makes the old work faster. A capability changes what the work is.