A campaign on €1,000 a month brings in 40 conversions at €25. Will €2,000 bring 80? Almost never: it will bring 65, at €31 on average… and the last 25 will have cost €40 each. That gap between the average CPA and the marginal CPA is the central idea of budget planning, and the one almost no report shows you.
Diminishing returns
Every campaign buys the most profitable auctions first (the ones Smart Bidding rates as most likely) and, as it gets more money, moves into worse auctions: more generic terms, weak hours, more expensive positions. The spend → conversions relationship is a curve that flattens out:
Conversions
^ ____________
| _______/
| ______/
| ___/
| __/
| __/
|/
+---------------------------------------> Spend
Three zones: at the start, every euro goes a long way (good auctions going unbought); in the middle, it performs normally; at the end, almost nothing (the market is bought, or what is left does not convert).
Average CPA versus marginal CPA
| Definition | What it answers | |
|---|---|---|
| Average CPA | Total cost ÷ total conversions | Is the campaign profitable overall? |
| Marginal CPA | Cost of the last conversions (Δcost ÷ Δconversions) | Is the next euro profitable? |
The marginal is always worse than the average in the flat zone. A campaign with an average CPA of €30 and a €45 target looks like it has room to grow; if its marginal CPA is already €60, every extra euro loses money even though the average still looks healthy.
How to draw a campaign's curve
From your own data
When the budget has changed in the past (or using the weekly history of spend and conversions), each week is a point (spend, conversions). In a spreadsheet: sort by spend and look at how conversions grow. With 8-12 points at different budgets you can see the shape. The limitation: seasonality and other changes muddy the points; use comparable periods.
With the Performance Planner
Google simulates each campaign's curve using its auction data (lesson 3): it shows estimated conversions at different spend levels. It is the "official" curve, with all the caveats of any forecast.
With a budget experiment
The clean way: a copy of the campaign with a +50% budget in a 50/50 experiment for 3-4 weeks. The extra conversions divided by the extra spend are your measured marginal CPA.
Deciding with the marginal
The next euro goes to the campaign with the best marginal CPA, not the best average CPA. Monthly procedure:
- For each campaign: average CPA, lost IS to budget, an estimate of the marginal CPA (your own curve or the Planner).
- Discard the ones that are not losing IS to budget (they cannot absorb any more) and the ones whose marginal is above target.
- Among the rest, raise in steps (10-20%) on the best marginal; wait two weeks; recalculate.
- Pull budget out of the ones sitting in the flat zone with a bad marginal, and move it to the ones still in the growth zone.
Curves and Smart Bidding
With a target CPA/ROAS, the algorithm already turns down auctions with an expected CPA above target: the curve "cuts off" by itself at the target and the budget is not spent in full. That is why a campaign with a target does not spend all its budget: it has reached its marginal. Raising the budget there does nothing; raising the target does (it buys worse auctions, on purpose).
Mistakes
- Judging growth by the average CPA and spotting the decline too late.
- Raising the budget in one jump (×2): you land in the flat zone in a single step and the average CPA suffers for weeks.
- Ignoring that the curve moves: competition and seasonality change its shape every month.
- Cutting the campaign with the best marginal because "it spends a lot".
💡 Ninja trick: the Suite's Smart Bidding (SBNS) makes its budget moves with exactly this logic: for each campaign it reads how well the target is being met, the lost IS to budget and the CPA trend across the last increases (the slope of the curve), and it moves money in small steps towards wherever the next euro performs, stopping when the marginal gets close to the target. The curve, recalculated every night.
What you should remember
- Spend → conversions is a curve that flattens out: diminishing returns.
- Marginal CPA (the CPA of the last conversions) decides the next euro; the average only tells you whether the whole thing is profitable.
- Draw it from your own data, the Planner or a budget experiment.
- Raise in steps, towards the best marginal; with a target, the curve cuts itself off.