Set your Q4 targets from payback because platform history no longer holds still
Three platforms changed how results are counted or delivered this year. A payback-based method for setting CPA and ROAS targets that does not depend on last quarter's reported numbers.
Most teams set next quarter's CPA or ROAS target by looking at the last quarter's reported result and nudging it. That worked when the reported result meant the same thing from one quarter to the next. This year it has not. Meta changed what counts as a click-through conversion in March. Apple added ad positions in App Store search that you cannot bid on separately. Google changed how budget-limited Target CPA and Target ROAS campaigns bid starting August 17, rolling out over several weeks. If your Q4 plan starts from "what did we get in Q2," you are anchoring on numbers produced under different rules.
Start from what a customer is worth
The alternative is to work backward from payback, which is independent of what any platform reports. The method has four steps.
- Define the value of a paying user over a horizon you can defend, such as 90 days of net revenue after store fees, refunds and the costs of serving them. Use cohort data from your measurement partner or billing system, not platform-reported revenue.
- Choose a payback requirement. This is how long you can wait to earn back acquisition cost. A business with cash constraints may need 30 days, while one with patient capital may accept 6 months. Make it an explicit number.
- Derive the maximum cost per paying user that meets the requirement. If a user returns $60 net in 90 days and you need payback within that window, $60 is a ceiling, and your target should sit below it with a margin for error.
- Translate the ceiling into the platform's currency. Each platform optimizes toward a different event, such as an install, a trial start, or a purchase. Convert the ceiling into a cost per that event using your own funnel rates between events.
Suppose, hypothetically, that 20 percent of installs start a trial, and 30 percent of trials convert to paid, so 6 percent of installs become paying users. If your ceiling is $60 per paying user, the cost-per-install ceiling is $3.60. With a safety margin of 20 percent, you would aim for roughly $2.90. The numbers are round and illustrative, but the logic is what matters: the target is derived from your economics, not from a past report.
Why history is a poor anchor this quarter
Each change affects the reference point in a different way.
- Meta. Meta said in March that click-through attribution would count only link clicks, with other interactions moving to a new engage-through category. That can shift reported conversions and cost per result without any change in user behavior. A Q2 baseline may be measured on different rules than your Q4 reading.
- Apple Ads. Apple said that additional ads would appear in search results and that advertisers cannot choose a position. Cost per tap and conversion rate may now be blended across positions that behave differently.
- Google. Budget-limited Target CPA and Target ROAS campaigns that were delivering well inside their targets are meant to move closer to them. A historically low CPA from such a campaign is not a promise of what it will cost next quarter.
None of this means platform numbers are useless. It means they are inputs to compare against your payback ceiling, not the definition of success.
Use platform history for direction, and your own data for the level
Keep using platform reports for relative questions: which creative performs better, which keyword set is cheaper, which country is stronger. Differences within a platform and a period are more reliable than absolute levels across periods. Use the cohort data you own for absolute questions: are we profitable, and how profitable?
A practical check is to reconcile weekly. Take installs and paying users per channel from your measurement partner, divide channel spend by them, and compare with each platform's own cost per result. If the ratio between the two moved over the summer, a definition changed and you should adjust targets by that ratio rather than by the platform's number alone.
Set targets in tiers
A single target for everything hides differences in intent and value. A simple structure works well:
- Brand and high-intent campaigns: the tightest target, because the users are already close to buying and the incremental question is addressed separately.
- Generic and prospecting campaigns: a tight target, close to the payback ceiling, with a review after two weeks.
- Experiments: a deliberately looser target on a small budget, with a stated end date, so that you learn without distorting the main numbers.
Change one thing at a time in Q4
Seasonal demand and competitor pressure move costs in the fourth quarter. If you also change targets, structures and creatives all at once, you will not know what worked. Set the new targets in the first half of October, hold them for at least two weeks, and put every other change behind them.
The takeaway: build targets from payback and cohort data you own, use platform history only for comparisons within a period, and annotate every definition change so Q4 results can be read honestly.