The Realization Gap

The savings your automation program reported are probably not the savings your CFO saw.

A program identifies 40,000 hours a year of manual work in a finance function. It builds the automations. The steering committee slide says "40,000 hours released." Twelve months later the function's cost base has moved by a fifth of that. Sometimes less.

Where did the rest go? Three places, every time. The work was released to nobody: twenty analysts each got back ninety minutes a day and filled it with other work, some of it useful, none of it measured. The exception queue was renamed, not removed: the automation handled the clean seventy percent, the messy thirty still needed a person, and the person was still there. And nobody had decided, before building, how the value would be converted. Absorb growth without hiring? Avoid a backfill? Redeploy to something with an output? Each needs an owner and a date. Without one, "hours released" is a number on a slide.

The distance between reported and realized value is the Realization Gap. It is not a technology problem. Most automation programs are technically fine. It is a decision that was never made. The firms selling AI right now are about to repeat this at a larger scale, because the tools are better and the accounting is the same.

The fix is boring. Before you automate anything, price the manual work, decide the conversion path, name the owner, and get finance to sign. Then build. Then measure the cost line, not the hours.