Gross margin by customer is easy to produce and almost meaningless in isolation. It captures price and product cost and stops there, ignoring that one customer takes full pallets monthly to a single location while another takes mixed cases three times a week to eleven sites, demands a two-hour delivery window, and returns four per cent of what it buys.
Both might show a twenty-eight per cent gross margin. One is comfortably profitable. The other is being subsidised by the first, and nobody in the business can see it because the costs that distinguish them are pooled in an overhead line and allocated on revenue.
What allocation on revenue conceals
Allocating logistics and service costs as a percentage of revenue guarantees the answer will be wrong, and wrong in a specific direction: it makes high-revenue customers look expensive to serve and low-revenue customers look cheap, regardless of what either actually consumes. Since cost drivers in distribution are overwhelmingly about frequency, drop size, location and complexity rather than value, revenue allocation inverts the truth for exactly the accounts where the answer matters most.
If your cost allocation basis is revenue, your cost-to-serve analysis is a restatement of your revenue analysis wearing a different hat.
The drivers that actually separate customers
A workable model needs relatively few drivers. The temptation toward exhaustive activity-based costing should be resisted. It produces models nobody maintains. Six or seven drivers usually capture the great majority of genuine variation.
- Order frequency and average drop size, the dominant transport cost driver in most distribution networks
- Delivery location and distance from the serving node, including access constraints and congestion
- Order composition: full pallet, layer, mixed case or piece pick, which drives warehouse cost by an order of magnitude
- Service requirements: delivery windows, appointment scheduling, waiting time, tail-lift or driver-assist requirements
- Returns rate and reverse logistics handling
- Commercial servicing: order management effort, dispute resolution, bespoke reporting and packaging
- Inventory carried specifically for that customer, including bespoke items and consignment stock
Building it without a two-year project
Cost-to-serve has a reputation as a heavyweight exercise, largely because it is often attempted as a finance-led activity-based costing programme covering the entire cost base. A supply chain-led version scoped to distribution and service costs is achievable in six to eight weeks.
- 01Take the transactional data as it is. Shipment records, order lines, warehouse activity logs and freight invoices contain nearly everything required. Do not wait for a data cleansing project.
- 02Build unit cost rates for each driver from actual expenditure, cost per drop, per pallet handled, per case picked, per kilometre by lane, per return processed.
- 03Apply the rates to each customer's actual activity profile over a full seasonal cycle. Twelve months, not a representative quarter, because service profiles vary seasonally in ways that matter.
- 04Reconcile the total against the general ledger. If the model does not tie to actual spend within a few per cent, the rates are wrong and every conclusion drawn from it will be challenged and defeated.
- 05Only then cut the results by customer, channel, region and product family.
The distribution you should expect
Almost every cost-to-serve analysis produces the same shape. A minority of customers generate the large majority of profit. A middle band is roughly break-even. And a tail (often twenty to thirty per cent of accounts) destroys value, sometimes substantially.
The distribution is usually more extreme than management expects, and it very often includes at least one strategically important, high-revenue, prestigious account sitting firmly in negative territory. That discovery is the entire point of the exercise, and it is also the moment the analysis is most likely to be suppressed.
What to do about it, in order of preference
The reflex response (raise prices or fire the customer) is the least useful. Cost-to-serve is a supply chain diagnosis and the best responses are operational.
- 01Change the service profile before changing the price. Consolidating three weekly deliveries into one, shifting from case picking to layer picking, or moving the account to a nearer serving node frequently converts a loss-making customer to profitable with no commercial conversation at all.
- 02Introduce order-size incentives. Minimum order quantities, free-delivery thresholds and drop-size pricing align customer behaviour with cost structure and are usually accepted without difficulty because they present as a discount rather than a penalty.
- 03Price the service explicitly. Charge separately for delivery frequency, special handling, expedited orders and bespoke packaging. This is the most transparent approach and it lets the customer choose whether the service is worth its cost, which is often the fastest route to them declining it voluntarily.
- 04Renegotiate commercial terms with the analysis in hand. A conversation grounded in a driver-level cost model is different in kind from one grounded in an assertion that an account is unprofitable.
- 05Exit, last and rarely. Occasionally correct, particularly where a customer's service demands are structurally incompatible with your network, but it should follow the other four options rather than pre-empt them.
The lasting benefit of a cost-to-serve model is not the one-time list of unprofitable accounts. It is that pricing, network and service decisions afterwards get made against a driver-level view of what serving actually costs, which changes the quality of commercial decisions long after the original analysis has been superseded.
