9 min read
Key takeaways
- A target CPA is a payback decision, not a platform default.
- Customer LTV after discounts, measured to a horizon finance will wait for, is the number that can actually repay media.
- A cohort of 1,000 new customers makes the leak visible in euros rather than opinions.
- If day-30 (D30) value cannot cover CPA, the budget shrinks even when new-customer volume looks healthy.
- Customer relationship management (CRM) and media must share one LTV definition, or the target means nothing.
Why does day-one CPA mislead?
First purchases include people who came for the welcome offer and people who will still be buying in week four. If you set CPA to 80% of day-one net revenue, you have priced media as if the discount did not exist and as if nobody churned. Both assumptions fail.
The honest input is customer LTV to a horizon finance will actually wait for, often D30 or D90, net of discount cost, refunds and failed payments. 'Month three' in the title is shorthand for that test: the target should still hold once the D30 cohort has closed and the D90 data starts to arrive. The worked numbers below use D30, because that is the earliest point most finance teams will sign off. A longer horizon than finance will wait for is a story; a shorter one is a coupon.
What is the payback equation?
Target CPA ≤ (LTV at horizon − variable fulfilment costs you will not recover) × hit rate. The hit rate is not a feeling. It is the share of cohorts that have historically reached the horizon value. If half your markets miss their D30 forecast, you cannot use 100% of forecast LTV as the cap.
| Input | Worked value | Note |
|---|---|---|
| D30 customer LTV | €72 | After discounts, refunds and failed payments |
| Unrecoverable fulfilment | €6 | Payment fees and fraud checks not in net revenue |
| Hit rate of D30 target | 70% | Share of markets that actually land here |
| Implied cap | (72 − 6) × 0.70 = €46.20, rounded to €46 | This is the target CPA, not €72 |
If media is currently at a CPA of €61, it is €15 too expensive, even though a slide might call €72 of LTV comfortable cover. Comfort is not payback.
What does a cohort of 1,000 new customers show?
Take 1,000 new customers bought at a €55 CPA. Media cost is €55,000. Discount cost averages €18 per customer. The day-7 (D7) repeat-purchase rate is 38%. D30 net revenue across the whole cohort, before discount cost and including the 62% who don't buy again in the first week, is €41 per customer.
| Line | Per customer | On 1,000 |
|---|---|---|
| Media | €55 | €55,000 |
| Discount | €18 | €18,000 |
| D30 net revenue before discount | €41 | €41,000 |
| Gap at D30 (41 − 55 − 18) | −€32 | −€32,000 |
Month three will not rescue this if the missing €32 has to come from a thin tail of repeat orders. Either CPA falls, the discount falls, or D30 value rises through product and CRM. Celebrating the 1,000 new customers only gives the loss a friendlier name: growth.
Now rerun it at a €40 CPA and a €12 discount, with the same D30 net revenue of €41: 41 − 40 − 12 = −€11 per customer, or −€11,000 on the cohort. The gap narrows from −€32 to −€11, but it does not close. Cheaper media and a smaller discount are not enough on their own; the remaining €11 has to come from D30 value, through product and CRM, before this market earns more budget. The customers did not change. The allocation did, and it has not yet gone far enough.
How often should the target move?
When the horizon data says so, not when the platform auction does. A weekly CPA that chases the cost per thousand impressions (CPM) will spend the D30 surplus before D30 arrives. We lock a target for each market for 30 days, then rebase it on the cohort that has just closed, which is why paid acquisition and retention and CRM have to run on one plan.
For the partner version of the same maths, read what an affiliate revenue-share deal is actually worth.
Read next: Server-side tracking: what breaks and what to fix first. Or see how we run paid acquisition against payback.