Field Note 02 · ROI & Productivity

Productivity Is Not ROI

A productivity story is the easiest thing to tell about AI and the hardest thing to bank. Saved time is not a return until someone captures it.

Daniel S. Wipert · 2026

The most common claim made for AI inside organizations is a productivity claim. The model drafts in seconds what took an hour. The assistant makes the team faster. The developers report that they ship more. These statements are usually true, and they are almost never returns.

A return requires that a specific benefit be captured, somewhere, net of what it cost to produce. Saved time is not captured value on its own. Time freed from one task becomes value only if the freed capacity is redeployed to work the business can monetize, or if a cost is actually removed. Time that fans out into more slack, more meetings, or more output no one needed is a productivity story with no line in the accounts. The hour was saved. Nothing was captured.

The distinction is not pedantic. Cost enters the flow automatically: every call, every review, every retry accrues whether or not anyone decides it should. Value does not. It appears only where someone draws a boundary around a specific piece of the business, this workflow, this period, this measured outcome, and answers a plain question honestly: what did this cost, and what did it return, net. Outside such a boundary a value claim floats free of any test, and untested claims default to the optimistic reading.

A productivity gain becomes a return the moment it crosses a boundary that someone owns. A role that is not backfilled. A queue that is cleared, releasing the revenue that was waiting behind it. A vendor line that actually falls. A cycle time that shortens in a way a customer pays for. Each of these is a captured benefit with a name against it. The difference between this and a team that feels faster is the difference between a return and a feeling.

The operating principle: require every AI value claim to name its boundary. Where is the benefit captured, over what period, measured how, net of what cost, and who is accountable for the answer. A claim that cannot name its boundary is a productivity story, not a return.

This is where most AI programs quietly lose money. They collect one hundred percent of the cost, which accrues by default, and an unknown fraction of the value, which accrues only by design. The gap between the two is invisible on any dashboard that measures usage. It is visible only at a boundary someone was accountable for drawing, and in most organizations no one is.