Demand Planning for Wholesalers: Forecasting Intermittent and Lumpy Demand

Wholesale distribution has a forecasting problem that retail often doesn't: a huge share of the catalog sells intermittently. Many SKUs go weeks with zero demand, then a customer orders a case — or a pallet. Averaging that pattern produces a number that's wrong almost every week. Here's how to think about forecasting lumpy, intermittent wholesale demand.

Why wholesale demand is hard to predict

Wholesalers serve businesses, not walk-in shoppers, so demand arrives in irregular, chunky orders driven by a handful of customers. The catalog is long, much of it slow-moving, and a single large account can swing a SKU's monthly total. Classic moving-average and reorder-point methods assume smooth, frequent demand — exactly what wholesale isn't.

The trouble with averages on intermittent items

Apply a moving average to an item that sells 0, 0, 0, 12, 0, 0 and you get a steady trickle that never matches reality. You either carry buffer stock for demand that mostly isn't there, or you're caught flat when the order lands. Across thousands of long-tail SKUs, that's a lot of trapped cash and a lot of misses.

Forecast at the right grain — and segment the catalog

Effective wholesale forecasting starts by segmenting items by volume and variability, then matching the method to the pattern:

• Fast, stable movers: standard statistical / ML forecasts with tight service levels.
• Intermittent items: methods built for sporadic demand (Croston-style and modern ML approaches) that model when demand occurs, not just how much.
• Customer-driven SKUs: incorporate known account behavior and large-order signals instead of burying them in an average.

Account for lead-time variability, not just demand

For wholesalers, supplier lead time is half the equation. Modeling its variability per supplier lets you protect availability where lead times are long or unreliable, and free up cash where they're short and steady — instead of buffering everything to the worst case.

Deliver it into the systems your buyers use

Better math only helps if buyers act on it. The forecast and recommended buys should land in your ERP's purchasing workflow or the tools your team already uses — not a separate system nobody opens. That's how forecasting actually changes what gets ordered.

Where to start

You don't need to boil the ocean. Pick a problem category — often the intermittent long tail — and prove a better forecast against your own history first. A short data audit will tell you which segments are dragging on cash and service, and where AI forecasting will move the needle fastest.