In short: A reorder point looks at one location and is blind sideways, so a site can trigger a purchase while the same item ages toward write-off four hundred kilometres away. The check worth adding is whether any other site holds stock surplus to its own needs that could cover the requirement before the order is approved. Any network generates more transfer candidates than can be executed and most are individually small, so consolidate them by lane and evaluate the truck rather than the line, because a single-line transfer almost never pays. The obstacles are organisational rather than mathematical, which makes them predictable and worth naming before you build anything.
A reorder point looks at one location. Stock at this site has fallen below the trigger, so generate a replenishment. That logic is correct in isolation and it is blind in exactly one direction: sideways.
Meanwhile the same item is sitting in surplus at another site four hundred kilometres away, ageing toward a write-off. The business is about to spend cash buying stock it already owns, and it will spend cash again later writing off the copy it forgot about.
Nothing in a location-by-location policy can see this, because the policy never looks across the network. It is one of the more reliable sources of quick cash release in a distribution business, and it requires no forecasting improvement at all.
The operations research literature has a name for both halves of this and the distinction is worth carrying into the design. Paterson and colleagues reviewed the lateral transshipment models in the European Journal of Operational Research in 2011 and separate reactive transshipments, where a site that has already stocked out pulls from a neighbour, from proactive redistribution, where stock moves on a schedule before anybody is short. The two have different economics and different approval paths, and running them through one queue is a reliable way to get the proactive moves rejected, because they always look less urgent than the emergencies sitting next to them.
The decision, stated properly
Before approving a replenishment for site A, ask whether any other site holds stock that is surplus to its own needs and could cover the requirement.
Surplus needs a definition that is more careful than on-hand quantity. A site holding four hundred units is not surplus if its own projected demand over the horizon plus its required buffer consumes all four hundred. The right measure is projected excess: on-hand plus incoming, minus projected demand over the transfer lead time plus its own safety requirement. Anything above that is genuinely available.
The horizon in that projection is the parameter that decides whether the answer is any good, and it gets set carelessly. It has to run at least as long as the transfer lead time plus the source site's own replenishment lead time, because that is how long the source stays exposed before it can restore anything it gave away. A three day transfer out of a site that replenishes on a ten week cycle needs a ten week horizon on the surplus test. Set the horizon to the transfer time alone and you will generate transfers that look free, with the evidence arriving two months later as a shortage at the site that helped.
Then the transfer is worth doing when the cost of moving it is less than the cost of buying new, adjusted for a few things people forget:
The freight cost of the lateral move, which for a partial pallet on a dedicated run can easily exceed the value of the goods.
The handling cost at both ends, which is small per unit and not zero.
The purchase cost avoided, which is the obvious benefit.
The holding cost avoided at the source site, which is the benefit people forget and which is often larger than the freight.
The obsolescence risk avoided at the source, which for a product with a shelf life or a lifecycle is frequently the largest term of all.
That last item is why the arithmetic favours transfers more often than intuition suggests. Moving stock from a site where it will expire to a site where it will sell converts a write-off into a sale rather than merely saving a purchase.
Ranking the candidates
Once you generate transfer candidates across a network, you will have far more than you can execute, and most of them are individually small. Two things make the list usable.
Consolidate by lane. Twelve separate transfer recommendations from Rotterdam to Antwerp are one truck, and the freight economics only work when you look at them together. Group candidates by origin and destination, then evaluate the consolidated move rather than each line. A single-line transfer almost never pays; a consolidated one frequently does.
The arithmetic on a real lane makes the case. Take an item costing 4.20 with a 22% annual carrying rate, moving between two sites where a part-pallet consignment costs 310 in less-than-truckload freight and handling runs 0.35 a unit at each end.
A typical candidate line is 90 units, worth 378. Evaluated on its own: purchase avoided 378, holding avoided at the source over an expected four months 28, against freight of 310 and handling of 63. Net benefit 33. That does not survive fifteen minutes of a planner's attention, let alone the meeting where somebody asks whether it was worth doing.
Now group eighteen of those lines onto the same lane. That is 1,620 units, about thirteen and a half pallets, and a shared truck run carries them for 600. Freight per line falls to 33. Each line now nets 309 instead of 33, and the eighteen together release 5,560 rather than 594. Same stock, same sites, same eighteen decisions, and the only thing that changed is whether they travelled together.
The obsolescence term changes which of them stand alone. Take the same 90 units at a site selling four a month with seven months of shelf life left. That site will sell 28 before expiry and write off 62, which is 260 at cost. Move them to a site selling forty a month and the line clears entirely. The 260 lands on top of the 33 the transfer was already worth, so a move that failed the single-line test passes it comfortably.
Rank by net benefit per unit of freight capacity. This is a packing problem, and the sensible objective is to fill the truck with the moves that release the most value. Sorting by benefit density and filling greedily gets you close to optimal and is easy to explain, which matters because someone has to approve the list.
The density to sort on is net benefit divided by the pallet space the line consumes rather than net benefit alone. A line worth 400 that fills three pallets loses to two lines worth 250 each that share one pallet between them, and sorting on raw benefit gets that backwards every time. Where the truck constraint is weight rather than volume, which happens on dense goods, swap the denominator and re-sort, because the two orderings can differ substantially on a mixed load.
Set a floor so trivial moves never reach a human. Below some threshold, the review time costs more than the transfer saves.
What tends to hold it back
The mathematics is straightforward. The obstacles are organisational and worth naming because they are predictable.
Site-level performance measures. A depot manager measured on their own service level and their own stock cover has a rational reason to resist giving up inventory. They carry the risk of the transfer and someone else gets the benefit. This is the single most common reason rebalancing programmes stall, and no amount of optimisation quality addresses it. Either the measure changes to a network measure, or transfers get approved centrally with the local site's service protected explicitly by the calculation, which is the more practical route.
Trust in the surplus calculation. If a site has been burned once by giving up stock it later needed, it will refuse from then on. The calculation must respect the source site's own buffer and be seen to do so. Showing the source manager exactly what remains after the transfer, and that it still covers their own requirement plus safety, does more for adoption than any accuracy improvement.
Transfer pricing. In businesses where sites are separate profit centres, an internal transfer creates an accounting event that can make a beneficial move look bad on somebody's P&L. This is worth sorting out once at a policy level rather than negotiating per transfer.
Where it shows up in the numbers
The effect appears in three places, in roughly this order.
Purchase volume falls first, and it is visible within a cycle or two.
Write-offs fall next, and this shows up over a longer period because the avoided obsolescence would have happened months later. It is also the effect most likely to be missed in a benefits case, since it is a cost that did not occur.
Worth setting the write-off measurement up before the first transfer runs, because that is the effect people argue about afterwards. Take the write-off rate per site for the twelve months before, and track the same rate after, on the items that actually moved rather than across the whole catalogue. A benefits case built on total write-offs gets diluted by everything the programme never touched, and that dilution is how a working programme gets cancelled at its first review.
Service improves at the receiving sites, because they get stock faster than a purchase order would have delivered it. In networks with long inbound lead times, this is often the benefit that gets the attention, since a lateral transfer arriving in two days beats a replenishment arriving in ten weeks.
When not to do it
Three situations where the answer is to buy rather than move.
When the surplus is not really surplus. If both sites have the same seasonal pattern and both will need the stock in six weeks, moving it just relocates the shortage. Check that the source's projected demand actually stays below its stock through the horizon rather than only today.
When the freight is disproportionate. Low-value, high-volume goods rarely justify lateral movement. A pallet of a bulky, cheap item can cost more to move than to buy, and the calculation should say so.
When the transfer breaks a compliance or traceability requirement. Regulated goods, temperature-controlled products and anything with batch-level market authorisation may not move freely between sites or countries. Encode those constraints as hard rules rather than leaving them to the reviewer, because a recommendation that is legally impossible undermines confidence in the whole list.
One failure mode has a signature clear enough to diagnose from the transfer log alone. The same item moves from A to B in March and back from B to A in May. Each move passed its own economic test at the moment it was evaluated, and the pair of them destroyed value and paid two freight bills.
The count to run is over the last year: how many item and lane pairs saw a transfer in one direction followed by a transfer in the opposite direction within two review cycles. A non-trivial figure means the horizon on the surplus test is shorter than the cycle the stock actually moves on, and reviewing more often will make it worse rather than better.
There is also a systemic caution. If rebalancing recommendations are large and constant, that is a signal about the buffer placement rather than an opportunity to be harvested repeatedly. Persistent surplus at the same sites means the safety stock at those sites is set wrong, and the durable fix is upstream in the network configuration. Rebalancing is the ambulance; the placement calculation is the road design.
Run the analysis once as a snapshot before building anything. Take current on-hand across every site, project demand over a reasonable horizon, and count how much stock is sitting above requirement at sites that do not need it while other sites are triggering replenishment for the same item. That single number usually decides whether this is worth automating.