2023 · ongoing·Data Scientist
Staffing forecasts the ops team actually uses
Replaced gut-feel staffing with a forecast that ops trusts — because they could see the why.
-31%
Forecast error
-18%
Overtime cost
100%
Adoption
Context
Weekly staffing was set by intuition, leading to costly over- and under-staffing. Past forecasting attempts failed because ops didn't trust black boxes.
Approach
- 01Co-designed the model interface with ops leads so outputs matched their mental model.
- 02Built an interpretable forecast with explicit seasonality and event adjustments.
- 03Shipped a what-if view so planners could test scenarios themselves.
Impact
Adoption hit 100% within a month — the explainability was the feature. Forecast error dropped 31% and overtime spend fell 18%.
ForecastingVisualizationStakeholders