Every expert says the same two things about Google Ads for apps. Start with an install campaign. Google is unpredictable at the start.
I ran both claims as tests on one account’s campaigns, not as a controlled experiment. $73K across 6 campaign types, and a separate $27K across 6 bid strategies with no install campaign at all.
Max Conversion Value beat tROAS everywhere
That held across both platforms, across all markets, every single time.
The reason is signal. tROAS restricts volume before Google has enough data to know what it is optimizing for. You are constraining a system that has not learned yet.
The sequence that worked here is to start with Max Conversion Value if your account has it, let it run 10 to 14 days, and only add tROAS later if you need to hold a return target and have the budget to feed it.
Not every account has it. Google’s page on choosing a bid strategy for App campaigns still says Maximize conversions for in app actions and Maximize conversion value are in beta and open only to a limited number of advertisers. Its page on Maximize strategies for App campaigns describes both without that caveat, and Google announced in August 2025 that it had expanded Maximize conversions bidding to App campaigns for in app actions and return on ad spend. The pages disagree, so look at the bid strategy menu in your own account before you plan around it.
iOS versus Android depends on the bid strategy
iOS ROAS hit 67%. Android was 37%. That looks settled until you check which campaigns those numbers came from.
That ranking only holds for Max Conversion Value. On tD0ROAS and MaxAction, Android outperformed iOS.
The bidding strategy and the operating system interact. What works on one platform does not automatically carry to the other, and a ROAS number at the platform level without the bid strategy attached tells you very little.
Search eats the budget, Display returns more of it
The network split was the part I did not expect.
I worked out revenue share myself by multiplying each network's budget share by its ROAS (Search 25%, Display 43%, YouTube 13%) and taking each result as a share of the total. The published budget shares add up to 99%, so I calculated the revenue shares on that 99%. Per dollar, Display returned 43% against Search's 25%.
Search takes the budget while Display returns more per dollar. If you are not segmenting by network, you are missing half the picture. Even then, an App Campaign will not show you which search terms took that budget, which is why Google Search is so hard to run for app installs.
You do not need an install campaign first
The second test ran 6 bid strategies on $27K over 9 days, with no install campaign, for a subscription app in the US.
Read individually, the campaigns looked chaotic. Watch any single campaign on any single day and you either panic or celebrate, and both reactions are wrong.
Read together, the total across all six was stable every single day. One dips, another picks up.
That is the argument for judging the account, not the campaign. It is also the argument for fewer campaigns. I cut from six to two or three, one target campaign for efficiency and one max campaign for volume. The same structure that works on Meta, where too many campaigns leave each one short of signal.
Why won’t my new Android App campaign spend?
The $27K test ran on Android, and three of the six strategies showed three different ways a new campaign fails. tROAS could not spend its budget at any target I tried, first 70%, then 50%, then 35%. tCPA spent, but it found cheap installs that did not convert. MaxAction scaled to $927 a day with no efficiency guardrails at all. The two that scaled and held were Max Conversion Value with no target and tD0ROAS. None of the six had an install campaign in front of it.
So lowering a tROAS target is not a guaranteed way to get spend, and a campaign that spends freely is not necessarily buying the right users. Before I change a bid, I run through what Google itself recommends. These are Google’s guidelines, not rules the campaign enforces, and a budget multiple does not guarantee delivery.
- Budget against the target. Google’s setup guide recommends a daily budget of at least 10 times your target CPA for an in app action campaign, and at least 50 times your target CPI for an install campaign. That is a budget multiple, not advice to raise the target.
- An action enough people complete. The same guide says to pick an in app action that at least 10 different users complete each day in the campaign. Under tCPA, Google’s bid strategy page advises against selecting more than one action, because the target becomes a blended target across actions that may have different values.
- The bidding option. A new campaign can bid to a target (tCPI, tCPA or tROAS) or use a Maximize strategy with no target, where Google says the budget governs performance instead of a target and budget together, if your account has it. tROAS asks for more: the Google Analytics for Firebase SDK, at least 10 conversions a day or 300 in 30 days, and Google recommends running tCPA first to set a baseline.
- The conversion window. Google says to set the window from how long the action usually takes, using the days to conversion data. A longer window counts more conversions but is more likely to include organic ones that were never incremental. A shorter one misses some that were. Google gives no single right length.
- Which action the campaign optimizes for. An App campaign bids toward the actions you select in its bidding settings. Google’s general rule is that primary actions are used for bidding and secondary actions are for observation only, unless a secondary action is part of a custom goal. Actions Google creates automatically from an attribution partner start as secondary, and when your MMP actions are already primary, Google tells you to set newly imported Google Analytics actions to secondary so they do not disturb existing campaigns. Check which purchase action is primary before you blame the bid.
- Fewer changes. Google says changes before the first 100 conversions may disrupt learning, and advises against moving the budget or the target by more than 20% at a time.
Most of this applies to iOS App campaigns too. If the campaign does spend but the money moves to a network you did not want, that is a different problem, and I cover it in why App campaigns end up on YouTube.
What I would do on a new account
- Start with Max Conversion Value if your account has it, not tROAS, and not an install campaign
- Give it 10 to 14 days before judging anything
- Segment reporting by network from day one
- Read the account total daily, and the individual campaigns weekly
- Add tROAS only when you need a return target and have budget to spend against it
The waiting period has backing from Google itself. Its App campaign guidance says to account for conversion delays before judging early performance. Both tests are one account’s numbers, not a controlled experiment, so rerun the comparison on your own data before you copy the structure.
One caveat on the $27K test, which ran in February 2026: during it, the account’s revenue event sent $1 instead of $0 when a conversion had no value. That inflated the ROAS Google reported across the board, and most on iOS. Read the comparisons between those six strategies as directional, and check that your own events send a real value or none before you trust Google’s ROAS. How far that bug pulled Google Ads away from the MMP on iOS, and how I tell such a gap from a normal one, is in why Google Ads, your MMP and SKAN report different iOS conversions.
The Google Ads Help pages and Google’s August 2025 announcement linked in this post were checked on 29 September 2026. None of the help pages shows a last updated date, and Google changes App campaign bidding often.