How many fulfilment nodes does a D2C brand need

The question almost always arrives as a capacity question. We are running out of racking, should we take a second warehouse. It is not a capacity question. Space in the same city is a procurement problem and you can solve it in a month. A second node is a network decision, and it moves your working capital, your delivery promise and your ops headcount at the same time.

Most Indian D2C brands run on one node for far longer than they expect to. Most of them should.

One node is the correct default

A single node pools everything. One pool of stock serves every order from every pincode. One receiving door, one cycle count, one team to train, one system of record. Every unit you own is available to every customer who wants it. That is the highest service level you will ever get per rupee of inventory, and no multi node design beats it on that measure.

What a single node cannot do is be close. Distance shows up in exactly two places. Transit time to the customer, and freight per order, which in India is priced by zone. From a node in the north, the west is a short lane, the south is a long one, and the north east is a separate conversation. Both the days and the rupees vary by lane and by category, because a two kilogram apparel parcel and a bulky home product do not move on the same economics.

So state the trade honestly. One node maximises inventory efficiency and pays for it in distance. Every node you add buys distance back and pays for it in inventory.

What a second node buys

Three things, in order of how reliably they show up.

Transit compression on the lanes it serves. This is the most certain benefit and the easiest to verify before you commit, because you can price the same parcel from two origins to the same destination cluster.

Freight per order on those lanes. Shortening a zone lowers the rate card. The size of the drop depends on the lane, the weight band and the contract you hold, so model it on quotes rather than on a rule of thumb.

Coverage and options. Pincodes that were technically serviceable but commercially useless become viable when the origin moves closer. Carrier choice widens, and reverse pickup timelines improve on those lanes.

There is a fourth benefit nobody lists until it matters. A single node is a single point of failure. Flooding, a labour dispute, a landlord problem or a sealed premises stops the whole business. A second node is also insurance, and that is worth something even when the freight case is thin.

What a second node charges

Inventory duplication comes first. You cannot split sixty units of a SKU across two locations and call either one stocked. Splitting forces either deeper buying or a narrower assortment at the smaller node.

Then the floor cost. Every node carries rent, a supervisor, a security arrangement, a system licence, statutory compliance and a share of your ops manager’s week, whether it ships fifty orders a day or five hundred. That floor does not scale down.

Then complexity, which is the cost people underestimate. Order routing rules that need maintaining. Two stock accuracies to defend. Place of supply and registration handling for the new state. Two returns flows, because reverse logistics drops parcels where the network drops them, not where you sold from. Inter node transfer paperwork.

Pooling versus proximity, stated plainly

Demand variability pools. Distance does not. When you hold one SKU in two places you have to protect two smaller and choppier demand streams, so the total buffer you need goes up rather than down. 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, for the same service level. Treat that as a planning check, not a law.

Proximity is worth most when your lanes are long, your category is impulse or occasion driven, and the customer is comparing delivery dates before they buy. Pooling is worth most when demand per SKU is thin and lumpy, when your assortment is wide, and when cash is the binding constraint.

The signals you have outgrown one node

Not one national order number. The useful signals are regional.

Look at whether a candidate region can absorb a full replenishment cycle of your fastest moving SKUs without the stock going stale. If it cannot, you do not have a network problem, you have a demand problem.

Look at the share of orders sitting on your longest zones, and whether that share is growing.

Look at whether one zone is dragging your freight per order away from the rest of the book.

Look at behaviour on those lanes rather than at cost alone. Higher cancellation before dispatch, higher cash on delivery refusal, higher return to origin, and worse repeat rates on long lanes all say the same thing. The promise is losing you money after the sale.

Look at where your suppliers moved. If your manufacturing base has shifted and your node has not, you are paying an extra inbound leg on every purchase order.

A decision sequence, not a formula

First, fix the node you have. A large share of second warehouse requests are really dispatch cut off problems, pick accuracy problems or carrier mix problems.

Second, measure demand at pincode cluster level rather than at state level. State views hide the only thing you need to see.

Third, test the cheap version before you sign anything. Platform fulfilment for the far region, or a temporary forward stocking arrangement for one season, will tell you what the promise change is actually worth without a lease.

Fourth, model the whole change. Freight saved, minus the node floor cost, minus the carrying cost of the incremental buffer, minus a realistic number for management attention. If it only works on the freight line, it does not work.

Fifth, decide the assortment before you decide the site. A node with the wrong hundred SKUs is worse than no node.

Sixth, write the kill criterion and the review date on the same page as the approval. Nodes are easy to open and politically hard to close.

The count is an outcome, not a target. Get the sequence right and the number falls out of it.