This practical method uses fictional examples. It does not report a client engagement, production trace or measured campaign improvement.
Define the decision before sorting by CPA
I would review an APAC push or pop placement at the level where I can actually act: source, zone, GEO, device or OS, and acquisition period. A zone average can hide a change in traffic mix. For push, creative exposure matters too. For pop, check the arrival and landing-page path instead of borrowing push click-through benchmarks.
First write the paid outcome and its denominator. Use reconciled billed spend in one currency, with fees treated consistently. Estimated tracker cost is provisional. These event definitions belong in the review:
- Deposit event
- A recorded deposit action. One person may generate several events; retries must be deduplicated.
- Unique depositor
- A distinct person who deposited in the defined population. This can include returning depositors.
- FTD
- A person's first qualifying deposit under the advertiser's rules, deduplicated by person. A tracker event label alone does not prove this.
- Approved FTD
- An FTD accepted after the required validation. Keep pending and rejected outcomes separate.
For an affiliate CPA agreement, I would compare billed spend with the commission actually approved under that agreement. Deposit amounts are not revenue. A cheaper deposit event does not establish cheaper approved acquisition or profit.
Give the cohort time, with a spending boundary
Group people by acquisition time, then count their outcomes at the same age. A conversion-date report can mix today's spend with deposits from older clicks. Record the attribution window, timezone, conversion lag and approval lag separately.
Use the observed lag distribution and the partner's validation schedule to choose a review age. There is no universal 24-hour maturity rule. Pending approvals make the economics provisional even when the deposit window has closed.
Set a per-placement test cap, a total campaign cap, a daily pacing limit and a maximum amount still exposed to pending outcomes. Write the next review time and stop condition before buying more traffic. Hold means hold the decision; it does not mean unlimited spend. At the cap, stop new buying and let the existing cohort mature.
Choose the smallest useful action
| Action | Evidence to look for | Next step |
|---|---|---|
| Pause | Broken tracking, a quality issue, or the agreed risk cap reached. Or mature economics fail the prewritten rule. | Stop new exposure. Record whether this is a safety stop or a performance decision; specify what would justify reopening. |
| Hold | Too few mature outcomes, approval backlog or a recent mix change. | Freeze bid and creative. Continue only inside the remaining cap, or pause delivery while waiting. |
| Bid test | A credible cost or delivery constraint, with tracking and quality intact. | Change only the bid within a preset range. Hold budget, targeting and assets steady. Recheck volume and quality because auction mix may change. |
| Scale test | Several comparable mature cohorts meet the economic and quality rule, with enough evidence for the planned risk. | Increase one budget control by a bounded amount. Keep a rollback level and measure the additional traffic separately. |
A higher bid is a test of price and access to inventory. A higher budget only helps when budget limits useful delivery. Neither guarantees proportional conversion growth. If automation already adjusts bids or excludes slices, record that behavior before adding manual rules.
PropellerAds documents Target CPA slices by zone, GEO and OS, with creative included for push. Its controls depend on the campaign setup. Treat the table above as a review method, not a claim that every source exposes identical controls. Official Target CPA documentation.
Worked example: four placements, four decisions
Fictional planning example. The following USD figures, counts and thresholds are invented to illustrate judgment. They are not James's campaign results or recommended defaults.
Suppose the agreement pays $40 per approved FTD and the buyer sets a $30 media-cost target, leaving room for other costs and uncertainty. The fictional plan caps the review batch at $300 and each placement at $90. Its cohort review waits seven days plus the partner's approval close. Those timings are assumptions for this example only.
| Placement | Spend / approved FTDs | Judgment |
|---|---|---|
| A | $90 / 0; cohort closed | Pause at the risk cap. CPA is undefined with zero outcomes. This does not prove the placement can never work. |
| B | $60 / 1; three FTDs pending | Hold. Current approved CPA is $60; if all pending FTDs were approved it would be $15. Neither endpoint is a forecast. |
| C | $60 / 2; cohort closed | $30 CPA meets the target in this small batch. Consider a bounded bid reduction only if comparable prior cohorts support the cost hypothesis. |
| D | $60 / 3; cohort closed | $20 CPA is a scale candidate, not a scale verdict. Three outcomes alone do not support a reliable ranking. |
The batch has spent $270, so only $30 remains available under its cap. For B, I would stop new delivery while approvals resolve. For D, I would require the planned evidence across mature cohorts before authorizing a new scale budget. I would not spread the remaining $30 across every tempting test.
If C qualifies for a bid experiment, choose one allowed bid step and keep all other settings fixed. If D later qualifies for scaling, change only its budget cap and compare the marginal cohort after it matures. Log either action separately; a blended before/after CPA can conceal weaker new traffic.
Leave a decision someone can audit
Record the placement and cohort, event contract, billed cost, mature and pending counts, action, cap, review date and reversal condition. Add the leading alternative explanation: delayed approvals, different placements within a slice, creative fatigue or broken attribution.
Tracking assigns credit under a rule. It does not show how many people would have converted without the advertising. Keep an attributed CPA decision separate from a claim of incremental growth.