Imagine introducing a forecasting tool and telling your team they must accept every recommendation unchanged. Now imagine letting them make a small adjustment.
That small difference can change whether they use it.
In three studies, Dietvorst, Simmons and Massey found that people were more willing to choose an imperfect forecasting algorithm when they could modify its predictions. Even tightly limited adjustments helped. Within the ranges tested, allowing larger changes did not reliably increase uptake further.
There is a catch: people’s adjustments often made the algorithm’s forecasts worse. But giving them some control encouraged enough algorithm use to improve performance compared with relying on their own judgment.
For managers, this suggests testing a defined adjustment option rather than making adoption an all-or-nothing choice. Track whether overrides improve results; feeling in control does not guarantee better decisions.
For my research, the distinction matters: this paper supports a link between limited control and willingness to rely on an algorithm. It does not establish that human oversight increases accountability. Those outcomes need to be measured separately.
The evidence comes from forecasting experiments, not modern generative-AI deployments, so its lessons need testing in today’s business workflows.
Source: Dietvorst, Simmons & Massey (2018), Management Science, “Overcoming Algorithm Aversion”. Published online in 2016.