Why this matters
A healthy blended cost per order is reassuring and nearly useless for deciding where money goes next. It averages cheap early orders with expensive late ones and hides channels whose last share of spend lost money. The marginal returns and forecasting guide explains why the next pound almost always buys less than the average one. Here the skill is not the arithmetic but knowing which figures can bear a decision, which need a test first, and which costs never reach the marketing line.
The merchant and its spend curves
Cobblebrook Tea Merchants is an invented online loose-leaf tea retailer. It buys paid search and paid social and runs an affiliate programme of content sites and voucher-code and cashback sites. Finance has agreed a target cost per first order from contribution margin, treating repeat purchase as a safety margin. Spend levels come from comparable months, with orders counted in the merchant's own order system.
Cobblebrook Tea Merchants: spend and orders by channel
Target cost per first order: £22, which is also the contribution a first order earns before marketing cost.
| Channel | Monthly spend | Orders | Average cost per order |
|---|---|---|---|
| Paid search | £3,000 | 300 | £10.00 |
| Paid search | £4,500 | 390 | £11.54 |
| Paid search (current) | £6,000 | 430 | £13.95 |
| Paid social | £2,000 | 80 | £25.00 |
| Paid social (current) | £3,000 | 135 | £22.22 |
| Paid social (last autumn) | £4,000 | 188 | £21.28 |
Affiliates, current month: voucher and cashback sites 280 attributed orders, content and review sites 120. Commission and network fees cost £4.50 per attributed order. Voucher codes give an average discount of £3.00.
Last spring a network dispute paused the voucher and cashback sites for a month. Their attributed orders fell almost to zero; total site orders came in about 60 below forecast.
Blended view the owner sees: £10,800 of cost for 965 attributed orders, about £11.19 per order. Finance books commission in cost of sales and discounts against revenue, so its marketing line shows only £9,000.
Social was cut back last autumn because its average looked the worst of the three.
From average to marginal cost per order
The marginal cost of a step is the extra spend divided by the extra orders it bought. Compare it with the target cost per acquisition (CPACost of each conversion.), never with the channel's own average. Past steps carry seasonal noise, so read each figure as a range. Attributed orders are claims, not causes: a channel whose orders would have arrived anyway costs far more per incremental order than it reports, which IncrementalityConversions caused by marketing. evidence exposes. The marketing profit and loss guide adds that commission and discounts are costs even outside the marketing line.
Task
- Work out the marginal cost per order of each budget step for paid search and paid social.
- Work out the cost per incremental order of the voucher and cashback sites, including the discounts.
- Rank every step and channel against the target, and say which spend is below the line and which above it.
- Propose next quarter's monthly allocation inside the same total cost, and estimate the change in orders and in contribution after marketing cost.
- Write a stop rule for each channel: the evidence that would make you step it back or hold it.
- Name the budget test that should confirm the largest move, with its primary metric and guardrails.
Worked answer
Cobblebrook: marginal costs and the reallocation
| Step | Extra spend | Extra orders | Marginal cost per order | Against £22 |
|---|---|---|---|---|
| Search £3,000 to £4,500 | £1,500 | 90 | £16.67 | Below |
| Search £4,500 to £6,000 | £1,500 | 40 | £37.50 | Well above |
| Social £2,000 to £3,000 | £1,000 | 55 | £18.18 | Below |
| Social £3,000 to £4,000 | £1,000 | 53 | £18.87 | Below |
| Voucher and cashback sites | £1,260 fees + £840 discounts | about 60 incremental | about £35 | Well above |
Proposed monthly allocation: search £4,500 (down £1,500, about 40 fewer orders); social £4,000 (up £1,000, about 53 more orders); voucher and cashback sites removed from commission and public codes withdrawn (£1,260 fees and £840 discounts saved, about 60 fewer orders); content and review affiliates unchanged.
Contribution after marketing cost, at £22 per order: search +£620, social +£166, voucher change +£780, about £1,566 a month better, with about 47 fewer orders. Of the £1,760 of cost freed, £1,000 funds the social test below and £760 is returned to finance until the test reads.
Search's last step is the clearest waste: the average still looks excellent, but the final share of spend buys orders well over target. Social is the opposite: the worst average on the sheet, yet both recorded steps came in under target and the curve has not visibly flattened. The affiliate figure is the trap. Its reported cost is the lowest of all, but the pause showed most voucher orders would have happened anyway, and with discounts counted each incremental order costs well over target. A healthy average return on ad spend (ROASRevenue generated per unit of ad spend.) on that line describes credit, not cause.
Stop rules. Search: raise it only if a fresh step test shows the next step under target; cut further if the lower step drifts above it. Social: if its marginal cost over a full review period passes target, step back; go no higher without the test. Voucher sites: if total orders fall clearly more than the paused month suggested, reinstate selected sites on a lower rate. Content affiliates: keep, and queue a holdout, because they are untested.
The largest move, and the one with no history above the current range, is pushing social beyond last autumn's level. Use the budget increase incrementality test: raise social by half in matched test regions, leave control regions alone, and judge the result on marginal cost per incremental order against the target, with blended cost per order and branded search volume as guardrails. The budget allocation guide explains how the result then feeds the next quarterly split.
Common mistakes
- Cutting the channel with the worst average. That is how Cobblebrook starved social last autumn.
- Trusting the affiliate average. Attributed orders from code sites are mostly demand that already existed.
- Leaving discounts and commission out. They sit outside the marketing line but are costs of the decision.
- Promising more orders. This reallocation buys profit with fewer orders; tell the owner.
Check your reasoning in the marginal returns and budget reallocation checkpoint, or revisit the intermediate digital marketing level if the channel jobs behind these budgets feel unfamiliar.
