Ask a logistics operator what revenue a customer generated last quarter and you will get an answer in minutes. Ask what that customer cost to serve, the actual resources their freight consumed, and you will get an allocation: total cost spread across shipments by weight or revenue share. Allocations are accounting, not economics. They systematically overcharge the easy freight and subsidize the difficult, because difficulty is precisely what the averaging erases.
True cost to serve is granular and causal. This shipment used this leg capacity, triggered this handling, consumed this share of a constrained resource on a peak day, carried this probability-weighted cost of failure and re-delivery. Two shipments identical on a rate card can differ in real cost by a factor of 2 once timing, geography, service level, and variance exposure are counted.
Why the number is hard
Producing it requires reconstructing the causal path of every shipment through the network: which assets, which legs, which touches, at what marginal versus fixed cost, under what congestion state. It also requires the probabilistic layer, because part of a shipment's true cost is the risk it imports: tight windows on volatile lanes carry expected failure costs that never appear on an invoice but always appear in the annual expediting bill. This is an estimation and modeling problem before it is a reporting problem, which is why ERP systems, built to allocate rather than to model, have never produced it.
The average hides the expensive customers.
Assemble the true cost of one delivery.
Price to the blended average and the network quietly subsidizes its hardest deliveries.
What changes when you have it
Cost to serve reprices the customer portfolio. Some accounts everyone valued turn out to be subsidized; some unglamorous ones turn out to be the margin engine. Sales incentives shift from revenue toward contribution. Pricing shifts from cost-plus-on-averages to floor prices that reflect what the freight actually consumes. Network design shifts too, because the loss-making flows finally become visible enough to fix or exit. Very few numbers change an operator's behavior at every level from the quote to the board. This one does, which is why the work of producing it honestly is worth doing.