Order-up-to replenishment means ordering, at each decision, exactly enough to bring the stock position back to a target level. It is the generic form underlying periodic review and min-max, and the policy most planning systems actually compute. It is remarkably efficient for the stage applying it. Its blind spot lies elsewhere: recomputed each period on a forecast that moves, it amplifies the variability it passes upstream.
Why it matters
Order-up-to is the reference policy of inventory theory. Under fairly broad assumptions, ordering enough to reach a well-chosen target level is the best possible decision for the stage taking it. It is also the policy that most planning engines produce in practice, whether they call it that or disguise it as a pair of bounds.
Its mechanics nonetheless carry a major side effect, and that is the whole subject of this page. The target level is not fixed: it is recomputed each period from a forecast that itself updates with the latest sales. The order placed is therefore not the demand observed, it is the demand observed plus the adjustment of the target level. A good week raises the forecast, hence the target, hence the order, well beyond what that week actually consumed.
The signature above makes this visible. The light curve is the demand received, perfectly stable in its amplitude. The golden curve is the order placed in response. At zero lead time, orders already swing more than sales. At long lead times they come loose entirely: upstream receives a signal that bears little resemblance to what the market actually asked for.
This is the most rigorous explanation of what is called the bullwhip effect. No player behaves irrationally, none exaggerates on purpose: each applies the policy that is optimal for itself, and the amplification arises from their superposition. The problem is not in the people, it is in the structure of the decision.
The second panel of the signature gives the order of magnitude, and it is striking. With a five-period forecast window, a lead time of four periods multiplies the variance passed upstream by five; a lead time of eight multiplies it by more than ten. These factors compound from one stage to the next, which is why a third-tier supplier sees erratic orders while final consumption is calm.
Two levers follow directly, and both are structural. The first is the lead time: shortening it does not merely lighten stock, it reduces amplification quadratically. The second is the smoothing window of the forecast: reacting to the latest observation maximises amplification, smoothing over a longer history strongly dampens it. Many organisations do exactly the opposite, chasing the most recent sales as closely as possible.
The expert lesson is to measure before debating. The ratio between the variance of the orders you place and that of the sales you receive is a simple indicator, computable on twelve months of history, and rarely tracked. When it exceeds two or three, the company is not enduring its market’s volatility: it is manufacturing it, then invoicing it to its suppliers in the form of lead times and prices.
The mechanism
Each period, you update the forecast, derive a target level covering the lead time and the period, then order the difference between that level and the stock position. A simple rewriting of this rule is extremely illuminating: the order placed equals the period’s demand, plus the change in the target level between two periods.
That second term is the culprit. If the forecast rises, the target rises, and you order more than was sold. Since the target level is proportional to the window to cover, lead time included, this correction is multiplied by the lead time. A modest change in the forecast thus turns into a considerable change in the order.
Theory gives a precise bound on this amplification. It grows with the square of the lead time relative to the length of the smoothing window. In other words, doubling the lead time amplifies far more than proportionally, while lengthening the forecast window dampens in the same proportions. These are two levers for action, not fatalities.
A third path exists, often the most effective, and it belongs not to calculation but to information. If the upstream stage receives final demand directly instead of having to infer it from the orders placed on it, it no longer needs to reconstruct a distorted signal. This is the foundation of data-sharing and vendor-managed replenishment arrangements: they do not improve the policy, they remove the cause of its drift.
The traps
The policy is excellent for the stage applying it, and that is precisely what makes it dangerous. With everyone optimising at home, the whole chain starts oscillating without any player being at fault. Judging a replenishment policy on local performance alone ignores what it costs the rest of the network.
Shortening the forecast window to gain responsiveness mechanically maximises amplification. You believe you are sharpening control, you are only injecting noise into orders. Useful responsiveness comes from shorter lead times, not from over-reacting to the last point of sale.
A target level updated every period produces orders that change constantly, including when the change carries no economic meaning. This nervousness disorganises upstream, which ends up disbelieving the signals it receives, and covers itself in turn with lead time and stock.
The rollout
Compare, over a year, the variance of orders placed with that of sales received, family by family. This single ratio says whether the company endures volatility or manufactures it.
Test the effect of a longer forecast history on order stability. The gain on amplification is often far larger than the loss of responsiveness, which is generally illusory anyway.
Lead time weighs quadratically in amplification. Every week saved on replenishment produces a double effect, on safety stock and on the variability passed upstream.
Pass suppliers your stock withdrawals or final sales rather than orders alone. This is the only lever that removes the cause of amplification instead of dampening its effect.
Define a change threshold below which the target level is left untouched. This deliberate insensitivity reduces order nervousness without measurably degrading service.
Neighboring concepts
From knowledge to action
The ratio between the variance of your orders and that of your sales takes a day to compute and rests on three levers. Our Inventory & distribution file measures it and brings it back to its structural minimum.