The numbers can all look good. They say little about what has happened to the work.

Processing time falls. Output rises. Staff spend less of the day on work nobody wanted to do in the first place. A business case that began as a projection starts returning figures that hold up under scrutiny, and an implementation that was an experiment eighteen months ago becomes simply how the work is done. There is every reason to call this a success, and in most respects it is one. But a dashboard reports what was measured, and what was measured was the thing the organisation set out to change. It is quieter about who now sees the difficult cases, about where review has replaced production, about which judgements have moved to another desk or another pay grade, and about what the people coming up behind no longer encounter often enough to learn from.

These consequences are shaped by decisions made in different rooms. Strategy identifies the opportunity. A business case prices the gain. Workforce planning models roles and headcount. Technology teams configure and integrate. Change programmes establish new ways of working. Risk looks for the ways it might fail. Every one of those functions can reach a defensible decision within its own terms, on its own evidence, against its own criteria - and the cumulative effect on judgement will still be nobody's paper. Organisations do not fail to plan Al adoption. They plan it exhaustively. The redesign of work is divided among functions that are never required to see it whole, and no single approval in the sequence is an approval of the thing that actually happened.

Al-enabled redesign can remove work outright, but it can also redistribute it. People may do less of the ordinary run of work and more of what begins when the ordinary run breaks: checking, interpreting and resolving the cases that do not fit. A sequence of bounded tasks can become oversight across several streams at once. The saving is booked where the time was taken out. The demand that replaces it can surface somewhere else entirely - another role, another team, another cost centre, another year's operating conditions - where nobody is looking for it, because nothing there was supposed to have changed.

Consider an illustrative claims operation in which the exception rate falls from ten per cent to four. On any reasonable reading that is a good result, and it is the result the system was bought to deliver. Now suppose volume quadrupled over the same period. The rate has improved by well over half. The people handling exceptions are dealing with sixty per cent more difficult cases than before, and difficult cases are not merely more numerous - they are what the working day now consists of, one after another, with the ordinary work that used to sit between them gone. Both numbers are true. Only one of them is going to the board.