The slide is the one every automation sponsor hopes to present. Output per employee is up by a fifth since the mobile app went live, and the head of retail at a regional lender is pleased, because the strategy calls for her staff to move off simple transactions and into lending to small businesses. The capacity is plainly there. What the slide leaves out is where it went, and the study below suggests the honest answer can be more of the same work.

Four researchers studied a real version of this inside a large Latin American bank.1 The March version of their abstract names all four and puts the bank at roughly 450 branches and 10,000 workers.2 The setting is unusually clean. The bank first rolled out a mobile app in stages, automating routine transactions, and later redefined roles and performance criteria to prioritize complex tasks such as lending to small and medium businesses. Neither step displaced anyone.1 That sequence lets the researchers watch what the technology did to the work before the incentives changed, and then what happened afterward.

A caution on the source before the findings. This is an unpublished working paper, presented at a Tinbergen Institute seminar in March and at the Society for Institutional and Organizational Economics conference in July, and it has not been through peer review. Everything below comes from the authors' public abstract, whose wording shifted between the two presentations. That is normal for work in progress, and a reason to hold the details loosely.

What a fifth more room bought

The app worked as a productivity tool. Output per worker rose by roughly 20 percent.1 The authors write that automation, for all the capacity it creates, "does not, on its own, lead to task upgrading."1 Workers kept concentrating on routine tasks and did not shift toward the high-value complex ones.1

A gain of a fifth, with nobody leaving, means the same people producing a good deal more, and on the abstract's account the extra went into the same routine work. Consider how that looks from a branch. A worker who has been evaluated for years on routine transactions keeps closing them, and now closes more of them in the same day. Nobody has told her the new capacity was meant for something else, and the measures that decide her standing still reward what she was already doing. That is my reading of the mechanism. The paper's own finding is the plainer one, that workers continued to concentrate on routine tasks.

The change that moved the work

Reallocation arrived with the second step. After the bank redefined roles and performance criteria to put weight on complex tasks, the authors report that "Task reallocation occurs only after the firm explicitly reallocates priorities."1 By then the app had already supplied the capacity, and what changed things was a new definition of the job and of how it would be judged. The abstract does not separate those two changes, so I would treat them as one lever.

Capacity that automation frees goes back into the old work until someone redefines the job and how it is judged.

Who moves last

Even after roles and criteria changed, adjustment was markedly slower among workers with long tenure, which the authors read as firm-specific know-how creating inertia.1 The abstract's account of managers changed between versions. In March it said high-productivity managers sped up the shift mainly among newer workers, widening gaps within teams.2 By July it described high-quality managers as lowering the cost of adjusting and easing the move toward complex work.1 My read is that the second step needs a plan for the long-tenured staff specifically, because neither version suggests a good manager alone closes that gap.

The line missing from the business case

Three weeks ago I wrote about an AI scribe study and ended on the question of what the time it handed back was for. This study, from a different industry, points toward an answer: on its evidence, the capacity goes back into the work people are measured on until someone changes the job and the measure. Where a business case counts the freed hours as savings, or as a promise of redeployment, and stops there, it has priced the capacity and left its use to whatever people were already being judged on. The order in the study, technology first and criteria second, left a stretch in which the extra capacity went into more of the routine work that the later redesign set out to move people away from. The study suggests one more line item: what will the performance criteria reward once the simple work is gone? If the answer is the same set of measures as before, I would bet the capacity gets spent on the same kind of work. If the answer is a new set, someone senior has to write it, and the workers with the longest tenure will need the most help getting there.

So when your last automation project reported its gain in output per person, who was assigned to decide what those people were now supposed to be doing with it?