Guide - Unit Economics and Commercial Decisions
Date: 2026-09-28
Build the model once, in three levels — per order, per customer, per business — then answer every commercial question at the level it belongs to. Most expensive mistakes are right arithmetic at the wrong level: a discount judged on revenue per order, a channel judged on its first order, growth judged on a lifetime ratio while the cash runs out.
Unit economics is the profit or loss attached to one unit of the business — an order, a customer — and the discipline of deciding commercial questions by working out what one more unit does to it.
This guide builds the model the CRO guides lean on and then uses it for the decisions they keep running into: whether a change is worth building, discounts, price, channel spend, promotional cadence. Same invented retailer as Guide - Ecommerce CRO, extended: UK direct-to-consumer, 120,000 sessions and 2,880 orders a month, £62 average goods value, 45% cost of goods, 12% returns.
The three levels
| Level | Unit | The number | Answers | Classic wrong-level mistake |
|---|---|---|---|---|
| Per order | one order | contribution per order | is this sale worth having? | judging a discount on revenue or AOV instead of contribution |
| Per customer | one acquired customer | contribution over time against acquisition cost | is this customer worth acquiring, and when is the money back? | judging a channel on first-order return, or on a lifetime figure nobody has observed |
| Per business | one month | operating profit and cash | can we afford it, and does it change the P&L? | approving a positive-ratio plan that runs out of cash before it pays |
A decision is usually answered at one level and checked at the next one up. A discount is decided per order and checked per customer (does it bring worse customers?). A channel is decided per customer and checked per business (can we fund the payback?).
1. Build the model
Copy this, then replace every figure with your own. Every input comes from finance or the order system, never from estimates — the whole model inherits the accuracy of its worst input.
LEVEL 1 — PER ORDER source: finance, order system
goods revenue £62.00
shipping revenue (blended) £1.10
───────
total revenue £63.10
− cost of goods (45% of goods) £27.90
− payment fees (1.9% + 20p) £1.40
− pick and pack £1.20
− carriage £4.95
───────
contribution, order kept £27.65
contribution, order returned −£11.05 fees, pick/pack, carriage both ways lost
blended at 12% returns £23.01 ← contribution per order
LEVEL 2 — PER CUSTOMER source: cohort data, ad spend
new customers / month 1,296 (45% of 2,880 orders)
marketing spend / month £38,000 all acquisition spend, incl. agency and creative
blended CAC £29.32 38,000 ÷ 1,296
paid CAC £38.00 38,000 ÷ 1,000 customers from paid channels
cumulative orders per customer month 0 3 6 12 24
1.00 1.18 1.32 1.52 1.70
cumulative contribution £23.01 27.15 30.37 34.98 39.12
↑
LTV(12m) £34.98 · LTV(24m) £39.12 contribution-based, observed cohorts only
payback vs blended CAC ≈ month 5 (between 27.15 and 30.37)
payback vs paid CAC ≈ month 21 (between 34.98 and 39.12)
LEVEL 3 — PER BUSINESS, PER MONTH source: management accounts
revenue £181,728 2,880 × £63.10
contribution £66,269 2,880 × £23.01
− marketing £38,000
─────────
contribution after marketing £28,269
− fixed costs £22,000 staff, premises, platform, software
─────────
operating profit £6,269 ≈ £75,000 a year · 3.4% of revenue
marketing efficiency ratio 4.78 181,728 ÷ 38,000
What to notice before using it:
- Operating profit is £75,000 a year on £2.2 million of revenue. Everything below is measured against that. The free-delivery threshold test in Guide - Ecommerce CRO §14 — conversion up 10%, contribution down £11,700 a year — would have cost 15% of the year’s profit while every dashboard reported a win
- The first order never covers acquisition. £23.01 against a £29.32 blended CAC: every new customer is £6.31 down until they come back. Repeat purchase is what funds growth here — Repeat Purchase Rate
- Paid customers take about 21 months to pay back, against 5 on the blended figure. The blended figure is flattered by organic customers the spend didn’t acquire — Customer Acquisition Cost · Payback Period
- LTV is bounded and observed. 24 months, from cohorts that are 24 months old, on contribution. An unbounded, modelled LTV would look better and be a forecast — Customer Lifetime Value · Cohort Revenue
The level-1 block is worked line by line in Guide - Ecommerce CRO, including why a returned order is −£11.05 rather than zero. The concepts behind each line: Contribution Margin · Cost of Goods Sold · Fixed and Variable Costs · Return Rate and Reverse Logistics.
2. Check it before trusting it
The inputs that are usually wrong, in order of how much damage they do:
- Returns counted as zero. A returned order costs money. Leaving returns out overstates contribution by the whole returns line — here £4.64 an order, 20%
- Gross margin used instead of contribution. Gross margin ignores fees, fulfilment and carriage, and makes every discount and channel look affordable — Gross Margin
- Fixed and variable muddled. A warehouse salary that doesn’t rise with volume isn’t a per-order cost; carriage is. Get it wrong and you either refuse profitable volume or chase unprofitable volume — Fixed and Variable Costs
- CAC numerator missing things. Agency fees, creative, affiliate commission. And retention spend wrongly included, which inflates CAC and hides retention economics
- Repeat customers counted as new. Ad platforms reacquiring existing customers make CAC look lower and new-customer volume higher — New vs Returning Customer Acquisition
- LTV on survivors. Computed on still-active customers, it excludes the ones who left — Survivorship Bias. And it’s a mean over a heavily skewed distribution, so segment it — Skewed and Heavy-Tailed Distributions
Then reconcile top-down. Level 3 of the model should land within a few per cent of the management accounts for the same months. If the model says £6,300 a month and the accounts say £1,000, a cost is missing — find it before using the model for anything. Marketing efficiency ratio — total revenue ÷ total marketing spend — is the cross-check no platform can inflate: if it falls while every channel reports improving returns, the channels are claiming the same customers — Marketing Efficiency Ratio.
3. Use it — the decisions
| Decision | Decided at | The calculation | Check at |
|---|---|---|---|
| Is this change worth building? | per order | extra orders needed to repay the cost | per business |
| A discount or promotion | per order | volume uplift to break even; share that’s incremental | per customer |
| A price change | per order | volume you can lose and still break even | per customer |
| More spend on a channel | per customer | marginal, incremental CAC against contribution over time | per business |
| Promotional cadence | per business | sale-week gain minus pull-forward | per customer |
| Free-delivery threshold | per order | contribution per session, not AOV | per customer |
Is this change worth building?
Break-even before the work, not after. A change costing £14,000 — the checkout rebuild from Guide - Ecommerce CRO §9 — to repay within a year:
orders needed £14,000 ÷ £23.01 = 608 a year = 51 a month
as a lift 51 ÷ 2,880 = +1.8% relative conversion, sustained
Then ask whether +1.8% is plausible and detectable. Plausible: it’s modest. Detectable: at site level, a 1.8% lift is far below what this site can power a test for — Guide - Statistics for CRO. So the decision can’t rest on a test; it rests on the strength of the diagnosis, and that’s worth knowing before the work starts — Break-Even Analysis.
In plain terms: the question isn’t “would this help?” — almost anything might. It’s “does it need to help by more than we could ever confirm?”
A discount — the welcome code
Proposal: 10% off the first order for new customers.
no code 10% code
goods revenue £62.00 £55.80
fees (1.9% + 20p) £1.40 £1.28
contribution, kept £27.65 £21.57 = 27.65 − 6.20 + 0.12 fee saved
contribution, returned −£11.05 −£10.93 smaller fee lost on a return
blended at 12% £23.01 £17.67 ← −£5.34 per first order
new-customer volume needed to break even: 23.01 ÷ 17.67 − 1 = +30.2%
New-customer orders must rise by nearly a third just to stand still — and that’s if every extra customer is genuinely extra. They aren’t. Most of the 1,296 people who’d have bought anyway now get £5.34 off:
cost 1,296 existing new customers × £5.34 = £6,918 a month given away
needed £6,918 ÷ £17.67 = 392 genuinely incremental
new customers a month
The code has to create 392 new customers a month who wouldn’t otherwise have come — 30% more — before it earns anything. Find out with a holdout: a random share of eligible visitors don’t see the code — Incrementality Testing.
Then check one level up. Discount-acquired customers typically repeat less. If code customers reach 1.45 orders by month 24 rather than 1.70, their LTV(24m) falls by the difference on every subsequent order — track their cohort separately from day one — Cohort Analysis.
Count the discount once. It either reduces first-order contribution, as here, or it’s added to CAC — never both, which double-counts it, and never neither, which is the common error. Discount Impact on Margin has the general arithmetic; Discounting Strategy the case for targeting narrowly.
A price change
The same arithmetic in reverse. A 5% price rise on goods:
current +5%
goods revenue £62.00 £65.10
contribution, kept £27.65 £30.69 = 27.65 + 3.10 − 0.06 extra fees
contribution, returned −£11.05 −£11.11
blended at 12% £23.01 £25.68 ← +£2.67 per order (+11.6%)
orders you can lose and still break even: 1 − 23.01 ÷ 25.68 = 10.4%
A 5% price rise pays unless it loses more than 10.4% of orders. Put as elasticity — the percentage change in volume divided by the percentage change in price — the break-even is 10.4 ÷ 5 ≈ 2.1. If demand falls by less than 2.1% for each 1% of price, the rise makes money.
The asymmetry is the whole lesson. Margin amplifies price in both directions: 10% off a first order needs +30% volume; 5% on survives −10%. It’s why price rises are under-tested and discounts over-used — Price Elasticity · Margin versus Volume. Test the rise properly, and watch repeat rate as well as conversion — Price Testing.
More spend on a channel
Proposal: another £8,000 a month on paid social. The platform forecasts 160 more new customers.
as the platform claims after a holdout (60% incremental)
new customers 160 96
marginal CAC £50.00 £83.33
cumulative contribution − CAC month 0 12 24
platform view −£26.99 −£15.02 −£10.88
incremental view −£60.32 −£48.36 −£44.22
Even on the platform’s own numbers, these customers haven’t paid back after 24 months. The average paid CAC is £38; the next £8,000 buys customers at £50, because the cheapest audiences were bought first — Marginal Analysis. After a holdout shows only 60% of them were caused by the spend, it’s £83. The answer is no, unless something one level up changes it: much better repeat rates for this channel’s customers, which cohort data would have to show.
Three rules that come out of it:
- Decide on marginal CAC, never average. The average includes customers you already had
- Decide on incremental customers, never platform-attributed ones. Platforms credit themselves for customers who were coming anyway — Incrementality Testing · Geo Holdout Tests
- Check the cash one level up. Level 3 makes £6,269 a month. Customers that take 21+ months to repay are funded out of that, or out of borrowing — Payback Period · Channel Mix
Promotional cadence
Proposal: a 20% sale for one week each quarter.
contribution per order at 20% off 27.65 − 12.40 + 0.24 fees = 15.49 kept
returned −£10.81; blended = £12.33
normal week 664.6 orders × £23.01 = £15,293
sale week 664.6 × 2.5 = 1,661.5 orders × £12.33 = £20,486
───────
apparent gain +£5,193
pull-forward the two following weeks run 20% below normal
2 × 0.20 × 664.6 × £23.01 = −£6,117
───────
net −£924
A sale week with 2.5× the orders loses money once the orders it borrowed from the following fortnight are counted — and returns usually run higher on sale orders, which this ignores. The bigger cost is invisible in any single sale: customers learn the calendar and wait, turning full-price demand into discounted demand permanently — Promotional Cadence.
The pull-forward is measurable: compare the weeks after a sale against the same weeks in a year without one, or against a holdout group that wasn’t emailed.
Free-delivery threshold
Already worked in full: Guide - Ecommerce CRO §14 reads a threshold test that raised conversion 10% and lost money. Setting one from the basket distribution in the first place: Shipping Thresholds. The rule is the same as everywhere here — judge it on contribution per session, never on AOV.
4. The decision sheet
One per commercial proposal, before any work or spend. It takes ten minutes, and it’s what turns “I think this will work” into a statement someone can check later:
DECISION SHEET
Proposal 10% welcome code for first orders
Level per order; check per customer
Calculation first-order contribution £23.01 → £17.67 (−£5.34)
Break-even +30.2% new-customer orders, all incremental
= 392 extra new customers a month
Must be true the code creates ≥ 392 customers/month who wouldn't have come;
their 24-month repeat stays near 1.70 orders
How we'll know holdout: 20% of eligible visitors never see the code, 8 weeks;
code-cohort repeat tracked separately at 3, 6, 12 months
Kill criterion incremental customers < 250/month at week 8, or code-cohort
month-3 repeat below 1.10
Owner / date
The “must be true” line is the point. It converts the proposal into claims that can turn out false, and it’s what makes the decision reviewable after the event — the same move as the decision rules in Guide - Running an Experiment.
5. Keep it current
- Refresh quarterly, and whenever carriage rates, payment fees, supplier costs or the returns policy change. A model nobody has updated is a fiction the business is spending against
- Re-cut LTV by acquisition cohort each quarter. Blended LTV moves with the channel mix; cohort curves show whether customers are actually getting more or less valuable — Cohort Revenue
- Keep the per-channel and per-category versions separate. Blended contribution hides categories that lose money on every order; blended CAC hides channels that never pay back
- Log each decision sheet with its outcome. The model gets better from being wrong in writing
How commercial decisions go wrong
- Revenue or AOV as the success metric for anything that trades basket size for order count — thresholds, discounts, bundles, payment plans
- Gross margin in the model instead of contribution
- Returns left out, so every change that raises returns looks free
- Blended CAC for channel decisions, and average CAC for marginal ones
- Platform-attributed customers treated as incremental
- An unbounded LTV justifying today’s spend with next decade’s assumptions
- The ratio without the cash. A healthy LTV:CAC and an 18-month payback is a business that can grow itself out of money — LTV to CAC Ratio
- The sale judged on its own week, with the pull-forward and the trained waiting uncounted
- No break-even before the work, so nobody notices the change needed a lift no one could detect
The short version
- Build three levels — per order, per customer, per business — from finance’s numbers, and reconcile level 3 against the accounts
- Contribution, not revenue or gross margin; returns as a cost, not a zero
- Decide each question at its level, and check it one level up
- Break-even before the work: the lift, volume or customer count the proposal needs
- Discounts need much more volume than they look; price rises survive much more loss than they look
- Channels on marginal, incremental CAC, against bounded LTV and payback
- One decision sheet per proposal, with what must be true and how you’ll know
Related: Guide - Ecommerce CRO for the per-order model inside a CRO programme · Guide - Running an Experiment for testing any of these · Guide - Statistics for CRO for whether a lift is detectable. Cross-domain views: CRO · Growth Engineering.