Forecasting a long tail service parts catalog
Exponential smoothing was being applied uniformly to a catalog where most items moved a handful of times a year, producing confident forecasts of zero.
- Items on a fitted method
- 100%Segment appropriate model selection replaced a single default
- Manual override volume
- ReducedOverrides concentrated on items where they measurably helped
- Output type
- ProbabilisticQuantile forecasts feeding safety stock calculation directly
The situation
Over seventy percent of the active catalog had fewer than twelve demand events per year.
The planning system applied one statistical method across every item regardless of pattern.
Planners had lost confidence and were overriding the system on volume that did not need it.
What we did
- 1
Segmented the catalog by demand frequency, variability, and criticality before selecting any model.
- 2
Applied intermittent demand methods and bootstrapped lead time demand distributions to the slow moving segment.
- 3
Produced quantile forecasts rather than point estimates so inventory policy could consume the uncertainty directly.
- 4
Instrumented forecast value add so the effect of each manual override became visible by planner and segment.
Other engagements
All work- Network Design
Consolidating a nine site distribution footprint
A network that had grown by acquisition was carrying duplicate coverage in three regions. Optimization showed the footprint could contract without losing next day reach.
- Simulation
Sizing an automation investment before signing
A vendor proposal promised throughput that the demand profile never actually required, and understated the staffing needed on the days that mattered.
- Inventory
Releasing working capital without cutting service
Every echelon held a full safety buffer sized against the same demand variability, so the network was paying for the same protection three times.