In short: Vendor managed inventory transfers the replenishment decision for a customer location from the customer's planner to the supplier's, and the supplier then owns the availability outcome. Four things have to be agreed before it runs: what data comes across and how often, the min and max policy and who may change it, who owns the stock and when title passes, and how availability gets measured. Total inventory across the two companies usually falls while the supplier's own balance sheet stock can rise, which is why the commercial terms matter as much as the policy engine. The arrangement is only as good as the location level stock and sell-out feed behind it, since stale or partial data produces confident wrong replenishment.
The first month goes well. Your planner is now setting the replenishment for the customer's regional depot, the orders are smoother than the ones the customer used to place, and everyone says so in the review. In week six a store group runs a local activity nobody told you about, the depot drains in four days, the min and max levels have not moved since go live, and the customer's supply chain director sends an email asking why availability dropped. It is a fair question now, because the decision that caused it was yours.
That transfer is the whole substance of the arrangement. Everything else, the feeds, the portal, the policy engine, is machinery for executing a decision that used to sit on the other side of a company boundary and now sits on yours.
The four things that have to be agreed
Contracts for this vary enormously and the workable ones settle the same short list.
Visibility at the level you replenish. On hand stock and sell-out for each location you are responsible for, at item level, with a stated as-at time. Aggregate national stock is not enough to run a depot policy, and a national number with a depot split derived from a percentage is worse than aggregate because it looks usable.
A replenishment policy you own. Min and max, or a target cover with a review period, or a base stock level. Which one matters less than the fact that it is written down, it is yours to change, and the customer knows the rule rather than just seeing its output.
Agreed boundaries on that policy. Maximum stock the customer will accept at a location, minimum they expect you to maintain, the drop sizes and delivery days available, and the categories of change you can make unilaterally against the ones that need their agreement. Without these, every unusual week becomes a negotiation.
Who owns the stock, legally and financially. Title transfer on delivery, or consignment with title passing on sale, or something in between with a time trigger. This is the term that decides most of the economics and it frequently gets settled last, by people who were not in the operational discussions.
Why total system inventory falls while yours can rise
Removing an ordering layer removes an amplification stage. Forrester described the dynamic in 1961 and Lee, Padmanabhan and Whang formalised the causes in Management Science in 1997, and one of those causes is order batching by an intermediate party. When you replenish the customer's depot directly against consumption, that batching stops happening. Disney and Towill's 2003 paper in the International Journal of Production Economics modelled the effect specifically for this arrangement and found the bullwhip contribution from the removed layer largely disappears. Waller, Johnson and Davis reported the same direction empirically in the Journal of Business Logistics in 1999.
Total inventory across the two companies usually falls, and where that reduction lands depends entirely on the ownership term. Work through the arithmetic on a simple case. The customer holds four weeks at their depot, you hold six weeks of finished goods to serve their orders plus everyone else's, so the pair of you hold ten. Under a well run programme the system needs eight, because the depot buffer no longer has to cover your order lead time plus your batching plus their own reorder cycle. If the terms put those eight weeks under your ownership, including the stock physically sitting in their building, you now carry eight where you carried six. The system improved by two weeks and your working capital got worse by two.
Consignment makes this explicit in the accounts. Under both IFRS 15 and ASC 606, goods held on consignment remain the consignor's inventory and revenue is recognised when control passes to the end customer rather than when the pallet arrives. So consigned stock stays on your balance sheet, days inventory outstanding rises, and if payment terms run from the sale rather than from delivery, the receivable clock starts later too. None of that makes consignment a bad structure. It makes it a priced structure, and the price belongs in the commercial conversation rather than being discovered by the finance team in the second quarter.
Title on delivery avoids all of it and gives up some of the operational benefit, since a customer who owns the stock will eventually start second guessing the levels you set. The middle option, title transferring after an agreed number of days or at an agreed stock age, splits the exposure and is worth more attention than it usually gets.
The data quality dependency is the whole thing
The policy is only as good as the position it thinks it is replenishing. Two failures do most of the damage and they compound.
Latency. A stock feed with a two day lag means every decision is made against a position that is two days old. On a stable item with four weeks of cover this is irrelevant. On a fast mover with six days of cover it is a third of your reaction time gone, and during any period of unusual demand it is the difference between responding and reporting. Establish the worst case lag rather than the average one, because the days that matter are the days when the feed also happens to be late.
Unreported loss. Shrink, damage, unrecorded transfers between the customer's own locations, receiving errors that never got adjusted. Each of these leaves stock in your system that is not on the shelf. Your policy sees enough cover and does not order. The location runs out. Because you own the decision now, the conversation about that gap is a conversation about your performance, and you will be arguing about a number you did not measure. In categories where shrink runs at a few percent of volume, an unadjusted feed will produce a persistent phantom position within a quarter.
There is a mechanical check for both. Take your calculated on hand position for each location and compare it to their reported position over ninety days. Plot the drift. A feed that is arithmetically consistent shows a flat difference; a feed with unreported loss shows a widening one, and the slope of that line is your shrink rate whether or not anyone will confirm it. Where the drift is material, the fixes are a scheduled reconciliation against physical counts, a location level allowance built into the policy, and a written rule about who absorbs the difference.
Ask three questions before you sign anything. How often is stock counted at these locations, what happens to the variance when it is, and does the feed reflect the adjustment. If the answers are annually, it gets written off centrally, and no, the feed will drift and you should size the buffer for it.
Trust, and where the resistance actually comes from
The Barilla case that Janice Hammond wrote for Harvard Business School in 1994 is still the clearest published account of why these programmes fail, and the part most people skip past is the internal opposition. Distributor reluctance is in there and it is the half everyone remembers. The other half is that Barilla's own sales organisation objected, because their incentives were built on volume shipped, and a programme designed to smooth shipments and take stock out of the channel directly reduced the number they were paid on.
That mechanism has not changed. If your commercial team is compensated on sell-in and the programme's purpose is to stop sell-in from being lumpy, you have designed two parts of your own company to work against each other, and the part with the quarterly target will win. Check the incentive structure before the systems requirements. It is a shorter conversation and it predicts the outcome better.
On the customer side the reluctance is about control and about information. Handing over replenishment means accepting availability risk from a party whose interests are only partly aligned, and sharing location level sell-out means giving a supplier a view of their business that has commercial value elsewhere. Both concerns are reasonable. The usual answer is a written scope on what the data may be used for, plus a service level agreement with consequences that run in both directions.
Pricing the service into the terms
This is a service with a cost structure: a planner, a system, additional stock, obsolescence risk, and the availability liability. Programmes that are not priced become an unfunded expectation, and the expectation grows.
Four instruments do most of the work, and most agreements use more than one.
An explicit service fee or margin adjustment, which is the cleanest and the hardest to get agreed, because it makes the cost visible at exactly the moment the customer is being told the programme saves money.
A commitment in exchange, usually range, volume or shelf space. A range commitment given in return for running the customer's replenishment is a real trade and it is easier to agree than money.
A service level agreement with symmetric consequences. You accept a penalty for availability failures caused by your decisions. They accept that failures caused by unnotified activity, feed outages, or unreported loss are excluded, and the exclusion list is written down rather than argued about later.
Terms on the exit. What happens to consigned stock if either side walks, who owns obsolete and short dated stock, and how long the wind down runs. This clause costs nothing to negotiate at the start and is expensive to negotiate at the end.
Where this stops
The programme transfers a decision. It creates value only where you make that decision better than the customer was making it, and that is a question with an answer rather than an assumption.
You have some real advantages. You see demand for your items across every customer, so a pattern emerging in one account is visible to you before it is visible to them. You know your own supply constraints and lead times, so you can shape orders around them instead of reacting to them. You see your own promotional plans earlier than they do.
You also have blind spots they do not. You cannot see the competing lines fighting for the same shelf, the planogram change scheduled for next month, the local events that move a single store, or the customer's own commercial decisions about which categories get pushed this quarter. On items where those factors dominate, the customer's buyer, with worse data and better context, will beat your policy.
There is a test that settles it before either party commits. Ask for thirteen weeks of location level stock and sell-out history, run your intended policy against it as a shadow, and compare the orders you would have placed to the orders they actually placed, scoring both against what actually sold. If your policy does not clearly win on that history, you have not earned the contract and taking it will cost you money and reputation at the same time.
Two boundaries worth naming. Deciding how much total stock to hold and at which echelon is a network question with its own method, and this arrangement is about who decides at one node rather than where the buffer belongs overall. And the shadow test needs a period that includes at least one promotion and one supply disruption, because a policy that only proves itself on quiet weeks has proved very little.
Ask your largest candidate customer for thirteen weeks of location level stock and sell-out history, run your own replenishment policy against it, and see whether you would have beaten the orders they placed themselves.