From just-in-time to just-in-case: How overseas warehouses are rewriting inventory and fulfillment logic

The Analysis

Overseas warehouses are being sold to sellers as a logistics upgrade. From where I sit they are a balance sheet decision wearing a logistics costume. Whether 2 to 3 day delivery beats 7 to 15 days is not the interesting question; of course it does. The interesting question is how much of your working capital you are willing to convert into a delivery promise, at what price, and what that promise is worth on the day the platform changes what it rewards.

Three numbers carry the argument. Fulfilment through an overseas warehouse runs 2 to 3 days against 7 to 15 days shipping in from China. In some Southeast Asian markets the local delivery standard has tightened from 72 hours to 48. And the reason the arithmetic moved at all sits on the corridor: Red Sea rerouting adds 10 to 14 days to Asia-Europe transit, with peak-season composite increases reaching as high as 40%. Two of those numbers sit on the revenue side. One sits on the cost side. Most sellers file all three under logistics and stop thinking.

I have been on the buying side of enough of these decisions to know how they fail. They almost never fail on the warehouse invoice. They fail on the stock that went into the warehouse, in a jurisdiction it cannot easily be moved out of, costed against a duty assumption nobody re-tested.

The first move is to take the cost structure apart, and it usually takes ten minutes to discover that two people in the same meeting are comparing different things. A forward-stocking model carries four cost buckets, not one. The goods cost does not change. The cross-border move changes shape: a container into a warehouse is cheaper per unit than a parcel to a consumer, but it is lumpy and it is paid months earlier. Local fulfilment is the bucket everyone compares, because it is the only one that appears on the warehouse quotation. And the cost of certainty is the capital tied up in stock plus the write-down on whatever does not sell. Sellers spend the meeting on the third bucket and sign on the basis of it. The decision actually turns on the fourth.

Here is the arithmetic I would put in front of a seller doing 3,000 orders a month on one hero SKU. These are assumptions, set by me to make the structure visible and not market data, and any reader should substitute their own. Assume a retail price of $39.90, a landed goods cost of $12, and a direct-from-China fulfilment cost of $6.80 an order all-in, covering parcel, last mile and the write-off on returns that never come back. Assume the forward model costs $4.90 an order all-in once the inbound container, storage, pick and pack, local last mile and local return handling are counted. The forward model wins $1.90 an order. At 3,000 orders that is $5,700 a month.

Now the bucket that never makes the deck. Assume the forward model needs 75 more days of stock than the direct model, because the container has to be produced, moved, cleared and received before the first order ships. At a $12 goods cost and 3,000 units a month, 75 days is roughly 7,500 units, or $90,000 of working capital parked in a foreign warehouse. At an 8% cost of capital that is $7,200 a year, $600 a month. The honest comparison is $5,700 of margin against $600 of carry plus the risk sitting on 7,500 units. The carry is not the problem. The 7,500 units are.

Scale it and the structure holds with different stakes. Assume 30,000 orders a month at the same $1.90, which is $57,000 a month of margin. Assume the same 75 days, which is now 75,000 units and $900,000 of working capital, $72,000 a year at 8%. Margin is still ten times carry. On paper the large seller should forward everything. In practice the large seller is the one who gets hurt, because 75,000 units of a hero SKU is also 75,000 units of exposure to one design, one size curve and one platform's ranking algorithm. The small seller's mistake is under-investing. The large seller's mistake is concentration.

The platform changes the shape of the maths, and almost nobody models it correctly. Ranking and promotion eligibility are not a slope, they are a threshold. A listing that clears the local-fulfilment requirement gets the traffic. One that does not gets what is left. So the marginal value of forward stock is not constant; it jumps at the line. Below the line you are buying a small delivery improvement. Above it you are buying access to the demand pool. Sellers build a linear cost-benefit model, conclude that the extra warehouse cost is roughly offset by the delivery gain, and stop there. The correct model has a cliff in it, and the cliff is the entire investment case.

The mechanics are not subtle. Overseas-warehouse listings get higher search ranking. Promotions are restricted to local warehouses. One platform requires a local returns warehouse as a condition of selling in the category at all. And in some markets the local delivery standard tightened from 72 hours to 48. Read those four together and the overseas warehouse stops being an option and becomes an entry ticket. An entry ticket is worth paying for, but only up to the point where the ticket still gets you in, and that point is set by somebody else's policy update.

RCEP gets quoted at sellers as 90% of goods moving tariff-free within the region, and I do not trust that number as a planning input. It is a coverage statistic about the agreement, not a statement about your SKU. What matters at the line-item level is whether your tariff line is in the schedule, whether the regional value content test is met, and whether you can document the cumulation at the time of clearance. A forward plan built on the assumption that duty is zero, with no SKU-level origin file behind it, is a plan with an unpriced liability inside it. Duty on forward stock is paid at import, in cash, before you know whether the season works. If the origin claim fails later, you have financed the mistake for the whole cycle.

The Red Sea number gets misused more than anything else in this file. Rerouting adds 10 to 14 days to Asia-Europe transit, and the standard response is to add 10 to 14 days of safety stock. Wrong instrument. A mean shift is plannable: if the corridor is consistently 14 days longer, you place the order 14 days earlier and your safety stock does not move. Safety stock covers variance around the mean, not the mean itself. What the rerouting actually did was widen the arrival distribution, and widening is what you insure against. Buyers who added two weeks of stock to cover a two-week mean shift have paid for inventory twice and covered the wrong risk. The cheaper fix was an earlier cut-off date with the factory.

The part that never makes the business case is reversibility. Stock in a domestic warehouse can go to another channel, back to a factory, or into liquidation. Stock in an overseas warehouse in a market you have decided to leave is a different asset: duty has been paid, VAT may only be recoverable against local sales, and moving it to another country usually means exporting it and starting the duty cycle again. That is a fixed commitment inside a variable business. The warehouse fee is monthly and cancellable. The duty is not.

Size decides which risk binds. A seller doing a few thousand orders a month should not run a multi-country network; the fixed cost of each node is real and the volume cannot amortise it. The right shape at that size is one node, one hero category, and a hard rule on how much working capital may sit abroad. The measure I use is forward inventory days as a share of total inventory days: above a third, the seller has bought a delivery promise with money it needs for the next buy. A mid-size seller, tens of thousands of orders and two or three categories, is where the network question starts to pay, because one backend with routing rules gives optionality and not just efficiency. The failure at that size is signing a single-country commitment because the unit price looked better, then discovering that a single-country commitment is a single-country business.

Category decides whether the forward model is a margin engine or a write-down machine. High-return categories, the 3C and beauty end of the range, take most of the benefit and most of the risk in the same decision. The benefit is that a local returns address turns a lost sale into a resale, and a re-listed unit is worth far more than a written-off one. The risk is that the same return behaviour which makes local handling valuable also means forward stock has to be sized against net demand rather than gross. Low-return, high-cube categories, furniture and bulky home goods, get less from the delivery promise and pay much more for storage, because space is billed and turns are slow. One stocking rule across both is how a catalogue ends up with the wrong inventory in the wrong node.

Southeast Asia is not a market. It is a set of markets with different rules, different platforms and different seasonal calendars. That is why the regional networks are being built: one operator covers 10 countries and regions with more than 20 warehouses, another runs 9 warehouses across the region, a third is building 5 in a single country. Those counts are the supply side responding to the same threshold logic the seller is responding to. The lesson for a buyer is not to follow the network. It is that being in the wrong node now costs more than being in two nodes, because the operator can route and you can move a listing faster than you can move a container.

Timing is where planning discipline earns its keep, because forward stocking has its own lead time and it is longer than people assume. Factory production, inland to port, ocean, import clearance, warehouse receiving, and only then the first order. Against a transit that is 10 to 14 days longer than it used to be, a peak-season stocking plan has to be taken a full quarter ahead, not a month. On a quarterly calendar that means the plan for the next peak is being written while this season's sell-through data is still incomplete. That is the structural problem with forward stocking, and no amount of platform data fixes it: you commit capital against a forecast you can only test after the commitment has become irreversible.

Three horizons move on different clocks. Inside the current quarter the only lever is mix: which SKUs are already in the node, which are in transit, which can still be redirected. Over one to two quarters the lever is the warehouse contract and the node count, and that is where the money is decided. Over a year the lever is the sourcing base, because a forward model in three markets is really a decision about which factories and which origins you want to keep. Sellers usually negotiate at the first horizon and commit at the third, which is the wrong order.

The warehouse quotation is a poor instrument for comparing operators, and buyers keep using it. Storage is quoted per unit per month or per pallet per month, which rewards the operator for holding stock and penalises the seller for turning it. Receiving, put-away, pick and pack and return handling are billed separately, and the return line varies most because it scales with a rate the seller controls only indirectly. The comparison that works is cost per fulfilled order at a stated return rate and a stated turn, not cost per pallet. Ask for the quote in that unit form. An operator who cannot produce it is telling you something about how they will behave in month seven.

Returns deserve their own line, because in the high-return categories they are the difference between the model working and not working. A local returns address plus relabelling plus re-listing turns a return into a second sale at a lower fulfilment cost. A process that ships the unit back to China turns the same return into a total loss plus freight. The number to govern is not the return rate, which you influence slowly, it is the recovery rate on returned units and the days it takes to get a unit back to sellable. Both are operational, both respond to contract terms, and both are what a warehouse operator will quietly let slip when nobody is measuring.

If a platform reweights its ranking away from local fulfilment, or relaxes the delivery standard back toward 72 hours, forward stock does not become slightly less valuable. It becomes a stranded asset inside a quarter. The traffic gate is the reason to pay for the warehouse. If the gate opens, you are paying a fixed cost for a benefit that no longer exists. That is why the commitment should be sized to survive the gate moving, and why a multi-year warehouse contract signed against a 48-hour rule is the worst trade in this market.

If tariff uncertainty resolves, it resolves one way or the other and the forward position is asymmetric. A resolution that lowers duty makes the stock you imported at the higher rate more expensive than the competition's next container, and duty already paid is not refunded because the rate later fell. A resolution that raises duty makes the forward stock cheap in hindsight and rewards whoever moved early. You cannot win both. What you can do is split the bet: forward the SKUs whose tariff line is stable and whose origin is documented, and keep the tariff-exposed SKUs on a shorter leash even when the delivery promise is worse for them.

If the corridor normalises and Asia-Europe gives back the 10 to 14 days, part of the case for holding stock abroad weakens, because what you were buying was certainty of arrival. This is the slowest of the three and the easiest to plan for, because it shows up in transit data before it reaches your cost line. The point is not to predict it. The point is to know which line of the business case it hits, so that when it moves you re-run one number instead of rebuilding the model.

Before any of this is signed I want one page with a screen on it. A SKU goes forward only if it passes three tests. It sits in the part of the range that carries the contribution, which for most catalogues is a minority of SKUs and a majority of revenue. Its return rate sits below the store's own weighted average, because the forward model magnifies whatever the return behaviour already is. And its demand is predictable enough that a quarter-ahead commitment is a plan rather than a bet. Everything that fails goes to domestic consolidation and ships on demand. The output is usually a small number of SKUs carrying most of the forward investment, which is the opposite of what a warehouse operator will propose to you.

The agreement should read like a service contract, not a lease. Delivery attainment against the local standard, measured weekly, with service credits that escalate. Return processing days from receipt to relist, same structure. An exit clause with notice short enough to matter, and a stock-transfer mechanism that says who pays to move what, and where. None of that is exotic; it is the difference between renting space and buying a promise, and only one of those deserves a premium. Over the top of it sits a cash guardrail, because forward stocking fails quietly: total inventory days and forward inventory days as a share of them, reviewed monthly against the buy plan, and a cash conversion cycle not allowed to lengthen by more than the margin gain justifies. A programme that adds thirty days to the cycle while the delivery promise adds less is buying traffic with the seller's own balance sheet, and that works only while growth lasts.

There is bargaining room that most sellers leave on the table, and it comes from the operator needing volume in a node as much as the seller needs the node. A commitment that is real but staged, one category first, a defined volume threshold before the second market, is worth more to an operator than a headline number that never materialises, and it should be priced that way. Ask for the second and third markets at the first market's rate, conditional on volume. Things I would not do: forward the whole catalogue to clear a platform threshold, because the threshold is cleared by the listings that qualify and not by the units that sit; compare operators on storage price, because storage is the smallest line and the easiest to discount; and sign a multi-year term against a delivery standard a platform can change in a policy update.

What I would sign, sitting on the buying side of the table, is a one-season commitment with a defined SKU list, a staged second market, service credits tied to the 48-hour attainment and the return recovery rate, and an exit that costs me a quarter rather than a year. That is not caution. It is the only position that lets me take the traffic gate seriously without betting the balance sheet on a rule I do not write. The warehouse is not the asset. The right to change my mind is.

↑ Back to the key points

— By Grace Yang