Installs are the wrong number, and the agent must never bid on them
Cost per install is easy to lower and easy to fool. Cost per trial is closer to money, and trial to paid is the number that decides everything. If a trial costs $6.80 and 37 percent of trials pay, a paying user costs about $18. Whether that is good depends entirely on what a paying user is worth over the next 60 days. Apple's own attribution through SKAdNetwork is aggregated and delayed, so the honest measurement comes from your subscription data, usually RevenueCat, joined to campaign data at the cohort level. An agent that only sees the ads console will optimise the wrong thing.
Apple Search Ads became Apple Ads in 2025 and gained placements beyond search results, on the Today tab and product pages. The rule does not change with the placement: bid on what pays back, measure it from revenue, not from the network's install count.
Connect Apple Ads and RevenueCat. Report cost per trial and trial to paid per campaign, per ad group and per keyword, joined by install cohort. Never report or optimise on cost per install. If a number cannot be measured because of SKAdNetwork delays, say so and wait rather than estimate.
Set the payback window before the first campaign
Sixty days is the usual answer for subscription apps: long enough to include the first renewal, short enough that you are not financing a year of ads on hope. Write it down before the first campaign, because the agent's raise, hold and pause rules all derive from it. A campaign that pays back in 38 days can grow. One at 71 days holds until retention improves. One with no data pauses until it has data.
Our payback window is 60 days. A campaign is healthy when the revenue from an install cohort covers its spend inside 60 days. Classify every campaign every Monday as growing, holding or paused by that rule and nothing else.
Start on your own keywords, with a custom product page per group
The cheapest trials come from the searches you already win organically: your brand, your top terms. Start there at $20 a day. Since July 2025 custom product pages can be linked to keywords and show up in organic search, and since October 2025 you can have up to 70 of them. So the page that receives the ad click for sleep sounds with timer should be a page about the timer, and the same page then serves the organic searcher. Category benchmarks put the conversion lift of a matched custom product page around five to nine percent, which at $6.80 per trial is the difference between growing and holding.
Add competitor terms only once your own terms are saturated, and only if the cost per trial holds.
Start Apple Ads with exact match on our own brand and our five best organic keywords in the US at $20 a day total. Build one custom product page per keyword group with matching screenshot captions, link it to the keywords so it also serves organic search, and use it as the ad destination. Add competitor terms only when our own terms hit their impression ceiling, and only while cost per trial stays within 20 percent of the brand campaign.
Raise, hold, pause: the weekly rule set
Every week: if payback is inside the window and retention holds, raise the cap by a third. If payback is outside the window, hold and fix what is behind it, usually the product page or day 3 retention. If the campaign has run for two weeks without a measurable trial, pause it and say why. Budget moves toward the campaigns that pay, not the ones that spend. And the ceiling you set for the month is a ceiling, not a target: an agent that spends it because it exists is doing the wrong job.
Weekly, per campaign: raise the daily cap by a third when payback is inside 60 days and day 7 retention of the paid cohort is within 10 percent of organic. Hold when payback is outside the window and propose one fix to the product page or onboarding. Pause after 14 days without a measurable trial and tell me why. Move freed budget to the best paying campaign. Never exceed the monthly cap I set; if you think it should rise, make the case with numbers and wait for my yes.
Google App Campaigns and TikTok come second, and only from winners
Google App Campaigns give you less keyword control and reward creative, so they start once you have hooks that earned watch time organically and a custom store listing per campaign, which Google reports as a median 10 percent conversion lift. TikTok Search Ads and Spark Ads start from the same winners: TikTok search grew about 40 percent year over year, and a hook that already carries the search phrase can be bought against that query directly. In both cases the rule is the same window, the same trial to paid measurement, the same weekly classification.
Open Google App Campaigns only with a custom store listing per campaign and our three best performing organic videos as creative. Open TikTok with Spark Ads on the posts that earned the most watch time and a search placement on the phrase inside the hook. Apply the same 60 day payback rule and the same raise, hold, pause classification as Apple Ads.
What you check on Monday
Spend, trials, cost per trial, trial to paid, payback per campaign, and the day 7 retention of paid cohorts against organic. One classification per campaign. One proposed change. If the agent asks for more budget, the case has to be in those numbers, and the answer is allowed to be no.