Choosing the optimization event for a retail app

Optimise for purchase with a value-based bid strategy, because both volume and value support it.

• Volume is sufficient. 4,000 purchases per month is roughly 130 per day, comfortably above the level at which target CPA or tROAS can learn - commonly described as at least 30 conversions in 30 days, and more is better.
• Value is available. Purchases carry an $80 average order value, which enables tROAS or maximize conversion value. When the business goal is revenue rather than installs, that is the correct lever.
• Concrete targets: $88,000 spend against roughly $320,000 of D30 revenue gives a blended ROAS near 3.6. Set tROAS a little below that to leave bidding headroom - around 3.0–3.2 - or, if the team is not ready for value-based bidding, a tCPA per purchase of about $22 ($88,000 ÷ 4,000).
• Do not use add-to-cart as the primary event now. At 12,000 per month it is a viable fallback if purchase volume were low, but it is a proxy that correlates imperfectly with revenue.
• Do not keep first open in the primary slot. It rewards opens that never monetise, which is precisely how the account can report healthy installs while finance sees weak payback.

One caveat to verify before switching: purchase values must be passed cleanly, net of refunds, deduplicated between MMP and Google Ads, and in a single currency. A tROAS target built on inflated values will overbid.

Do not edit the existing campaign's conversion action in place - that discards learning and gives no comparison.

• Run a parallel campaign. Build a second app campaign optimising for purchase funded out of the existing budget, and keep the first-open campaign running at reduced budget to preserve install volume and account-level learning.
• Start the new campaign at 25–30% of budget so the system can gather purchase conversions without a cold-start failure, then ramp in 20% steps as cost per purchase stabilises.
• Expect a 1–2 week learning period and pre-agree not to judge on day three.
• Check the attribution window against the 38% tail. If more than a third of purchases arrive 30+ days after first open, a short window will systematically under-report the new campaign and make it look worse than the old one. Confirm click and engaged-view windows are set correctly and, if repeat purchasing matters, evaluate the D30-to-D90 curve rather than D30 alone.
• Sanity-check the event pipeline: is purchase firing once per order (not per item or per screen), are refunds netted, and does the MMP agree with Google Ads on purchase counts within a reasonable tolerance?
• Report on purchases and revenue from day one, with installs as a secondary diagnostic rather than a headline.

Reframe the metric rather than defend it. Installs were never the goal; they were a proxy chosen when the goal was ambiguous. Under the current setup, cost per purchase is about $22 and D30 ROAS is 3.6 - but spend is being optimised toward users who open the app, not users who buy, so the marginal install is worth less than the average install. That gap is exactly what finance is feeling.

The purchase-optimised campaign will probably buy fewer installs at a higher CPI. That is the mechanism by which revenue grows, not a side effect to apologise for. Set expectations in writing before launch:

• Primary KPI: purchases and revenue per dollar of spend.
• Guardrail: blended cost per purchase must stay at or below roughly $22; if it drifts above that for two consecutive weeks, throttle the new campaign and investigate.
• Secondary: installs, which may fall 20–30% in month one, and cost per install, which will rise.

Also point out the reporting artefact that will make the transition look worse than it is: the old campaign's purchases are credited partly to users acquired earlier, so the new campaign starts with no historical purchase credit. Judge it on incremental purchases against a holdout or a stable pre-period, not on a week-one week-over-week comparison.