One demand, three buffers: how stock chains double the cost of inventory

A plant holds two weeks of safety stock of finished goods. The regional hub that draws from that plant holds another two weeks. Every depot below the hub holds one and a half weeks. Together, the network carries 5.5 weeks of protection against a single uncertainty in customer demand. In many such networks, three weeks of stock could deliver the same fill rate.

Nobody decides on this double storage. Each location sets its buffer on its own numbers and has no idea how much stock is lying a step above or below it.

Each buffer in the chain is set according to the fluctuation it can see. The depot sees customer orders. The hub sees the sum of the depots' replenishment orders. These orders arrive in lumps and swing more than the customer demand behind them. The plant sees the hub's orders, which come in even bigger lumps. All three levels are thus reacting to the same disturbance in the underlying demand, the disturbance grows as it travels up, and all three hold extra stock for it.

Service targets make the problem worse. Suppose a depot aims for 98 per cent availability and the hub also targets 98 per cent for its supply to the depots. The hub is then protecting once again the service level the depot has already secured from its own stock. Each figure looks reasonable when viewed separately, so such internal targets are seldom questioned.

The data shows three signs. First, the total days of cover across all levels is far greater than the longest lead time in the chain. Second, stock at the hub grows old while the depots run short of the same items. That means the stock is at the wrong level: too much at the top and too little where customers place orders. Third, the safety stock settings at different levels were last reviewed in different years by different teams. No single sign proves double storage, but when all three are present, a proper review of the network is warranted.

How should the buffer be set? Start from the customer-facing level, because the real service promise is made only there. First decide what fill rate each depot must reach and within what replenishment lead time. Before that, group the items. Fast-moving goods sold at every depot behave differently from slow items stocked at only two locations, so one rule should not cover both. Then move upwards.

Each higher stage is given a committed service time, that is, a promise of how many days it will take to fulfil an order from the stage below. A hub that promises two days needs enough stock to keep that promise most of the time. If the same promise is stretched to five days, the hub can hold less stock but the depots must hold more. This trade-off is real and worth costing, because stock higher up is usually cheaper to hold and more flexible to move elsewhere, while stock lower down is close to sales. Double counting ends when a single service target is set at the edge of the network and the committed times of all stages are set together. Multi-echelon inventory management software does this calculation for thousands of item and location pairs, because a spreadsheet falls short beyond a few dozen.

How would the remedy look for one item? Take an example item with the 5.5 weeks of cover described above. If the buffers are set together, they could be split as one week at the plant, one and a half weeks at the hub and about one week at each depot. The depot's fill rate stays where it was. Total stock falls by more than a third, and the cash released once the old stock at the top is sold shows up in the first quarter.

Why should supply capacity figure in the same discussion? It is assumed that the plant can refill the buffer on schedule. A line already running at its limit cannot refill an emptied hub within the assumed lead time. The real replenishment time then stretches and all the buffers below fall short. Constraint-based supply planning software checks a replenishment plan against what the constrained resources can actually produce before the plan is accepted.

Inventory settings and supply plans should therefore be reviewed together. The new targets coming out of the multi-echelon software should reach the supply planning software in the same cycle. The result is a network that holds less stock and can genuinely refill the stock it holds. This month, review one product family. Compare the sum of cover at all levels with the longest lead time, and the size of the double buffer becomes clear. Where the sum is twice the longest lead time or more, a large part of the excess stock is due to the double buffer.