A distributor with a central warehouse feeding branches has a multi-echelon network: stock at the centre and stock at the branches protect the same customers, and a decision at one level changes what the other level needs. Multi-echelon inventory optimization (MEIO) is the family of methods that set stock across those levels together instead of treating each site as its own independent supplier.
Where the idea comes from
Clark and Scarf showed in 1960 that for a serial system the right quantity to control is the echelon stock (a site's stock plus everything downstream of it), and that base-stock policies on echelon stock are optimal under their assumptions. Most of what followed, including the one-warehouse, many-retailer results, builds on that observation.
The model behind most MEIO software
Commercial MEIO tools are largely built on the guaranteed-service model of Graves and Willems. Each stage quotes a guaranteed service time to the next; demand is assumed bounded; safety stock is placed to cover the net replenishment time at each stage. The model is elegant and solvable at scale, and it produces the familiar picture of pooling risk at the centre and holding less at the branches.
Its assumptions deserve attention in distribution. Bounded demand and guaranteed interstage service are exactly the assumptions a port closure or a supplier outage violates. When the centre cannot receive, the branch's guaranteed service time from the centre is fiction, and the safety stock placed on that basis is short.
Network optimization as an alternative
A different approach models the network directly: purchases arriving at the centre, transfers on approved lanes to branches, receiving and storage limits at each site, cash and credit shared across all of them, and demand and supply drawn as scenarios rather than bounded. Each review re-solves the whole network. Risk pooling appears as a consequence (the optimizer holds at the centre when that serves several branches cheaply) rather than as a design rule, and disappears when transfers are slow or capacity-limited.
This is heavier computationally and needs an operator's actual lanes and constraints, which is why it suits a distributor with a handful of sites more than a global manufacturer with hundreds of stages.
What to ask any MEIO vendor
Does the model know the transfer lead time and receiving capacity between sites, or does it assume instant availability? Does it treat two suppliers through one port as two sources? What happens to the placement when the upstream lane fails? And can it show the value of the multi-echelon placement separately from the value of a transfer it would have made anyway? Counting the same saving twice is easy in this area.