How do you handle team resistance to a tool that changes their forecasts?

The most effective approach is not to settle the debate upfront but to make it measurable. Forecast value added (FVA) continuously compares the accuracy of the automatically generated forecast with that of the final forecast after human overrides.

The research is more nuanced than the usual narrative. A study of roughly 147,000 forecasts by Robert Fildes, Paul Goodwin and Shari De Baets, published in the International Journal of Forecasting, found that manual adjustments improve accuracy for only slightly more than half of items, with a clear asymmetry: upward revisions generally hurt performance, while downward revisions, especially large ones, improve it. In other words, human judgment adds value, but rarely where teams instinctively think it does, and optimism costs more than caution.

The debate then stops being about ego and becomes a question of scope. The most advanced organizations end up limiting override rights to the segments where people truly outperform the model, and they get there after several iterations, not on day one.