Definition
Intermittent demand forecasting addresses series where positive demand occurs irregularly, separated by periods with no demand. It asks how frequently demand arrives and how large it is when it occurs. This differs from estimating sales hidden by stockouts: a genuinely quiet period can occur even when an item is fully available.
Why It Matters
- Slow-moving replacement parts can remain commercially important despite selling on relatively few days.
- A daily average alone does not describe the timing uncertainty that affects replenishment.
- A Commerce Intelligence OS should distinguish sparse demand from missing observations before recommending inventory changes.
How It Works
- Build a regular SKU-location time series. Mark unavailable periods and missing source records separately from confirmed zero demand.
- Compare simple benchmarks with methods suited to intermittent series. Croston's method separately smooths positive demand sizes and the intervals between them, then uses their ratio as a demand-rate forecast.
- Evaluate over the replenishment horizon with rolling historical cutoffs. Avoid percentage errors that divide by zero; inspect bias and operational stock consequences alongside forecast error.
- Review obsolescence, promotions, and changing purchase frequency. The original Croston method has known bias and does not itself provide a full predictive distribution; validate the chosen approach for the actual assortment.
Ecommerce Example
Context: Illustrative example: a replacement seal sells three units on two days during a thirty-day, fully stocked month.
Recommended move: The observed average is 0.2 units per day. Preserve the pattern of two demand occasions instead of interpreting that average as a daily fractional order.
Why it matters: Use longer history and supplier lead time to evaluate replenishment scenarios. These hypothetical counts do not establish a recommended stock quantity.
iKawn Framework
Observe
The iKawn framework retains zero-demand periods and their availability context.
Model
Represent purchase frequency and quantity as distinct planning evidence.
Evaluate
Compare candidate forecasts at the operational decision horizon.
Decide
Keep demand estimates separate from service targets and inventory rules.
Concise Summary
Sparse demand requires explicit treatment of zero periods and arrival timing. Evaluate intermittent forecasting methods against the replenishment decision rather than choosing a method from the series label alone.