Splitting stock across nodes: what actually breaks

The day stock lands in a second location, your operating model changes. Not the warehouse, the model. Before the split, availability is a single number and every order has one obvious answer. After it, availability is conditional and every order carries a decision.

Brands prepare for the lease and the racking. They rarely prepare for this.

What changes the moment stock is in two places

Your national stock number becomes fiction. A customer in Chennai does not see four hundred units. She sees the forty at the node that can reach her in the window you promised, or she sees a worse date. Availability is now a function of location, and your website is probably still reporting a single figure.

Every order needs a routing rule, and routing rules need an owner. Cheapest node, nearest node, node with the full basket, node with the earliest cut off remaining today.

Stock accuracy roughly doubles as a workload and reconciliation more than doubles, because units in transit between nodes belong to neither one. Purchase orders now need a destination decision at the time of raising, which pushes a network question upstream into buying. Returns arrive where the reverse network drops them, which is not where the order shipped from.

Forward deployed stock versus central reserve

This is a network question, not the pick face and racking question inside a single building. Here it means how much of your stock is pushed out to serve a region, and how much sits centrally uncommitted.

A pure split deploys everything. Maximum proximity, zero flexibility. Every forecast error becomes a physical problem.

A mother and satellite structure keeps a reserve at the main node and pushes stock out in response to what actually sold. It gives up a little proximity and buys back the ability to be wrong. For most brands below a substantial revenue base, this is the right shape.

Use the reserve deliberately. It is your answer to forecast error. If nothing is held centrally, every miss turns into an inter node transfer, which is the most expensive way to correct a plan. New launches, seasonal lines and unproven SKUs stay central and earn forward deployment only after they show velocity in the region.

Split ratios follow velocity, not fairness

Even splits are the default, and the default is wrong.

Run two gates. Eligibility first, then ratio.

Eligibility asks whether the SKU should be at that node at all. The floor is a function of your replenishment interval and your case pack, not a fixed number of units. If the node cannot move a case pack within a replenishment cycle, deploying it there does not improve service. It relocates dead stock and adds a transfer to your future.

Ratio starts with that node’s share of regional demand and then adjusts. Adjust up for lead time to replenish that node from central, because a slower link needs more cover. Adjust up where regional demand is choppier. Adjust down for short shelf life, high return rates and anything seasonal that you may need to consolidate later.

Express the outcome in weeks of cover, never in a percentage split. Two nodes at the same weeks of cover are balanced. Two nodes at the same unit share are usually not.

The commonest failure here is the long tail deployed everywhere. Tail SKUs have thin and lumpy demand. Splitting them multiplies the buffer requirement and still produces the outcome where both nodes are out of stock in the same week. Tail stays central.

The split shipment problem and what it costs

A multi line order where no single node holds every line has four bad options. Ship two parcels. Hold the order until one node can complete it. Transfer a unit and then ship. Or ship the whole order from the node that has everything, which is usually the far one.

Splitting looks cheapest and is not. You pay a second shipping charge, a second pack, and often a second minimum billable weight, which is punishing on light items. You double the tracking events, so you roughly double the chance of a where is my order contact. You create two return windows and two exposures to return to origin. And the customer reads a partial delivery as a mistake even when both parcels arrive on time.

Track split shipment rate weekly. It is the single best measure of whether your placement matches how customers actually buy from you.

Here is the specific correction most brands miss. Placement must respect basket affinity, not just velocity. If your merchandising pairs one SKU with another through bundles, cross sell modules or a subscription pack, those SKUs have to live together at every node that holds either of them. Ranking SKUs individually by velocity and deploying them independently is exactly how you manufacture a split shipment rate you then blame on the warehouse.

Rebalancing is a symptom, not a tool

Inter node transfers cost freight, two handling events, documentation, and time during which the stock is unsellable while it continues to age.

One planned seasonal consolidation is normal. Routine monthly rebalancing is not a capability, it is a diagnosis. It says your split ratios are wrong, your eligibility floor is too low, or your buying is not committing to a destination early enough.

Measure transferred units as a share of units dispatched and watch the trend. When it rises, fix the allocation upstream at the purchase order rather than downstream with a truck. The worst pattern of all is transferring stock to rescue a stockout when quoting a slightly longer date from the other node would have served the customer at a fraction of the cost.

The safety stock penalty of being in more places

The same SKU in more locations needs more total buffer for the same service level. The standard planning shape is a square root relationship, which puts two locations at roughly one and a half times the buffer of one, and three at roughly one and three quarter times. It applies to the buffer, not to cycle stock.

The practical consequence is a conversation, not a calculation. Your inventory days rise in the quarter you split, and someone will ask why turns fell in a period when sales grew. The answer is that you bought proximity with cash. That is a defensible trade if you said it in advance and indefensible if it surfaces in a review.

Which is the argument for making deployment earned. Per SKU, per node, against a floor, reviewed on a stated cadence. Stock does not have a right to be everywhere.