Where to put your second fulfilment node in India

Once a brand accepts that it needs another node, the location gets decided badly. A broker sends a rent quote. A founder has a home city. A partner has spare space and offers a rate. Those are inputs. None of them is the decision.

Site selection for a fulfilment node is an analytical exercise with about six inputs, and it can be run in a week on data you already own.

Demand centroid versus supply origin

A node sits between two gravity fields. Customers pull it one way. Suppliers pull it the other. Which one wins is decided by the ratio of your inbound freight cost to your outbound freight cost.

Inbound moves in bulk, palletised, at a low cost per unit. Outbound moves as thousands of small parcels priced by zone and by weight. For most D2C brands the outbound side is several times larger, so demand wins and the node belongs near customers.

The exceptions are real though. If your product is heavy or bulky, if your order volume is still modest relative to your inbound tonnage, if you are import led and clearing through one port, or if your category suffers inbound damage that you need to inspect at receipt, supply origin gets a much heavier weight.

India has tight supplier clusters, and that geography matters. Apparel out of Tirupur or Ludhiana, ceramics out of Morbi, engineering goods out of Rajkot, jewellery out of Jaipur, electronics assembly around the NCR belt. Placing a node far from your cluster adds an inbound leg and days of lead time to every replenishment. Lead time is not free. It converts directly into buffer stock, and buffer stock is cash.

The pincode share analysis that should drive it

Pull twelve months of delivered orders by destination pincode. Delivered, not placed, and flag the cancelled and returned volume separately because you will want it later.

Roll the pincodes into clusters. A few dozen clusters will usually account for the large majority of your volume, and the exact number depends on how concentrated your category is. Work at cluster level. State level views hide everything worth knowing, and raw pincode level is too noisy to read.

Weight the clusters by what actually drives your cost. Order count if your freight is parcel and rate driven. Billed weight if you are in a heavier category. Then overlay contribution, not revenue, so you do not build a network around thin orders.

Now the step most analyses skip. Correct for suppression. Your current demand map is shaped by your current promise. A region you serve slowly is already under ordering, so it looks smaller than it is, and the analysis quietly recommends that you keep ignoring it. Read traffic share, add to cart share or marketplace glance share against order share for the same clusters. Where the interest is high and the conversion is low, you are looking at demand your network is refusing.

Then, for each candidate city, compute two numbers against your existing node. The volume weighted transit day, and the volume weighted freight. Not the distance to a centroid. A centroid is a geometry answer to a commercial question, and in India it usually lands you somewhere with no carrier density.

Proximity to carrier hubs, not just to customers

The address does not ship. The lane does. Being in a city is not the same as being on the network.

Ask four questions about every shortlisted address. How many carriers run a same day pickup there, and at what cut off. Where is the nearest sorting hub, and does your parcel go through one hop or two before it enters line haul. If you ship air for light and high value goods, how far is a cargo terminal that actually accepts your category. And does the industrial belt around that address have late cut offs, because a location an hour outside the city can quietly cost you sixty minutes of cut off.

Cut off time is frequently worth more than distance. An hour of cut off is often a full day on the customer promise, and a full day on the promise is worth more than the freight difference between two comparable addresses. Brands negotiate hard on rent per square foot and never ask what time the last van leaves.

Cost optimal and promise optimal are different answers

These are two objective functions and they do not produce the same city.

Cost optimal minimises volume weighted freight. It drifts toward your largest clusters and toward cheap property, because that is where the rupees are.

Promise optimal minimises the number of customers who see a date beyond your target. It drifts toward the edge of your coverage gap, because that is where the dates are ugly.

Decide which one you are solving for before anyone opens a spreadsheet, and write it in the brief. If the node exists because the south sees a long date, then a site that saves the most freight and leaves that date unchanged has failed, no matter how good the model looks.

Why it is usually not your biggest city

The marginal value of a node is what changes for customers who are currently served badly. Your biggest city is almost certainly served well already, or it is where node one sits.

The largest metros are also the most expensive on rent and labour, the tightest on quality supply, and the most competitive on warehouse leasing.

The answer is often the second or third city of an under served region, or a satellite location on a highway or rail corridor just outside a metro, where property is cheaper, labour is easier to hold, and carrier line haul still passes through. The test is not the size of the city. It is how much of your badly served volume comes inside your target window when the origin moves there.

Shortlist three, then test before you sign

Score three candidates on the same sheet. Volume weighted transit, volume weighted freight delta, carrier cut off, labour availability and attrition, inbound lead time from your main supplier cluster, room to expand, and exit terms.

Then route real orders through the leading candidate for a season using a temporary arrangement, and read what happens to the delivery date, the conversion rate and the return to origin rate in those clusters. A lease is a three year opinion. A season of live data costs far less and is much harder to argue with.