Balancing Stockouts Against Overstock: Five Common Replenishment Model Pitfalls

Two faces of the same coin

The awkward truth of replenishment management: stockouts and overstock usually coexist, and they are often two results of the same wrong decision — wrong totals, wrong allocation, wrong timing all leave both empty shelves and dead stock on the books. Optimizing one side in isolation almost inevitably worsens the other.

Five pitfalls recur across replenishment model implementations.

Pitfall 1: uniform parameters

"One safety-stock rule for the whole catalog" is the most common shortcut. SKUs differ wildly in volatility, stockout cost and margin structure; uniform parameters leave high-volatility items stocking out while stable items tie up cash. Fix: tier parameters by ABC class × coefficient of variation (CV), with at least 4–6 parameter bands.

Pitfall 2: the average trap

"Average daily sales × replenishment period" is the first step of most calculations and the most treacherous. Two SKUs both averaging 50 units/day — one ranging 45–55, the other 0–200 — need completely different policies. Fix: every computation must carry both the mean and the dispersion (standard deviation / CV). A model that only uses averages assumes a deterministic world.

Pitfall 3: promotion-contaminated history

Promotion spikes blended into history make moving averages and exponential smoothing extrapolate peak events as normal demand — producing overstock after the event. Conversely, zero-sales days during stockouts are read as true demand, biasing forecasts down and causing the next stockout. Fix: tag and clean history before modeling — promotion days excluded or modeled separately, stockout days marked as censored demand. Cleaning rules should be scripted and reviewable, not remembered.

Pitfall 4: one-sided KPIs

Measure only stockout rate and buyers over-order; measure only turnover and they under-order into stockouts. Fix: pair stockout rate with turnover days (or dead-stock share), with category-specific balance points — fast movers tolerate lean stock; long-lead custom goods must carry more.

Pitfall 5: parameters set once, never revisited

Safety stock computed once, coefficients fixed once, then untouched for six months — while markets, channels and supplier lead times drift. Models don't fail suddenly; they fail gradually, and by the time anyone notices, the damage is done. Fix: a review cadence — monthly forecast-bias direction checks, quarterly service-level calibration against realized stockouts, seasonal-coefficient re-estimation after every major promotion — with every parameter change logged.

The goal is neither "zero stockouts" nor "zero inventory" — both are unaffordable — but minimizing the combined cost of stockouts and overstock under an explicit capital constraint. Accepting that moves the model from chasing correctness to serving decisions.