The campaign hit its CPI target every single day. Installs were cheap, volume was steady, and the dashboard looked like a win. Then the revenue never showed up. By the time anyone looked closely, weeks of budget had gone to users who installed and never came back.
The algorithm was not broken. It did exactly what it was told. The signal underneath it was the problem. This is one of the most common patterns we see across attribution audits: a smart-bidding campaign performing beautifully against the wrong objective, with total confidence.
Smart bidding is an agent acting on the signal you give it
Meta Advantage+ and Google App Campaigns are not dashboards you read. They are optimisation agents that act.
Signal-based (smart) bidding: automated bidding that optimises toward whatever conversion signal you send it, so an algorithm fed only install events will buy the cheapest installs, while one fed revenue or key-event signals will buy users who generate value.
- The agent pursues the objective you set, and the objective is defined by the signal you send back to the platform.
- Feed it "install," and it will get very good at finding the cheapest possible installs. That is the goal you gave it.
- The failure is structural, not algorithmic. And structural failures do not announce themselves. The campaign looks healthy right up until the revenue gap becomes obvious.
It is the digital equivalent of an automated system confidently shipping winter coats to a warm city because the calendar said November and nothing in the data contradicted it. The system was not wrong about its instructions. Its instructions were missing the signal that mattered.
Install-only signals teach the algorithm to buy cheap, not valuable
Garbage signal in, wasted spend out, delivered with complete confidence.
- An install-optimised campaign converges on the people most likely to install cheaply. Those are rarely the people most likely to pay.
- CPI looks excellent while revenue quietly fails to follow, because the two metrics are measuring different things and only one of them is on the bidding objective.
- The gap compounds. Every day the algorithm learns a little more about how to find low-value installers, so the longer a mis-signalled campaign runs, the better it gets at the wrong thing.
This is why chasing the lowest CPI without protecting user quality so often backfires. The number improves while the business does not.
The fix: send revenue and key-event signals, not just installs
The correction is to feed the agent a signal that correlates with value, and to do it fast enough that the algorithm can learn inside its optimisation window.
- Map the handful of in-app events that actually predict value (first purchase, deposit, trial start, subscription) and send those back as postbacks.
- Do not flood the algorithm. Sending too many events dilutes the signal; three to five high-signal events per platform is the sweet spot. The postback setup guide for Meta, Google, and TikTok walks through the mechanics, and Linkrunner's revenue tracking API handles the value events.
- The signal arrives fast enough to matter. Across roughly 670k installs that generated a revenue event within 30 days, spread over 55 Linkrunner projects in the 90 days to 26 June 2026, the median gap between install and first revenue was around 24 hours, with the slowest tenth stretching to about 18 days. Half of payers reveal themselves within a day, comfortably inside the algorithm's learning window.
Teams that point the algorithm at value see it in the numbers. Playo cut Google Ads CPI by 34 per cent over three months, and Matiks cut Meta CPI by 46 per cent, both by optimising toward the right signal rather than raw install volume.
A framework for choosing the right signal to send
Run this before you set any smart-bidding objective.
- Identify the predictive event. Which single in-app action best predicts that a user will be worth acquiring? Use your own cohort data to find which events actually predict LTV, not the event that is easiest to fire.
- Confirm it fires reliably and early. A perfect predictor that arrives on day 30 is useless to a campaign optimising daily. Favour the earliest event that still correlates with value.
- Send it cleanly. Configure the postback, map the event correctly, and validate that the platform is receiving it. A mis-mapped value event is worse than none, because it teaches the wrong lesson.
- Review what the agent did with it. Check whether cost-per-value-event and downstream ROAS improved, and correct the signal if they did not.
What this looks like by vertical
The right signal depends on what "value" means for your app.
- Subscription and dating apps: optimise toward trial start or first payment, not install. For these apps especially, revenue attribution matters far more than install attribution.
- eCommerce: first purchase, or a high-intent event like add-to-cart when purchase volume is too thin to train on early.
- Gaming: tutorial completion or first in-app purchase as early proxies for a user who will stick and spend.
Keeping a human in the loop
Automation handles the optimisation. It cannot tell you that the objective itself is wrong.
- The algorithm will never flag that you pointed it at the wrong signal. It will simply pursue it efficiently. That judgment call is yours.
- A clean signal foundation is the starting line, not the finish line. Once the right events are flowing, the work shifts to deciding which signals to test next and reading what the agent does with them.
Platforms like Linkrunner make the value-signal layer dependable: revenue and key-event postbacks configured once, validated, and sent automatically, so the agent is always learning from the signal you actually chose.
FAQ
Why does install-optimised smart bidding bring low-value users?
Because it optimises for exactly what you measured: cheap installs. The cheapest installers are rarely the highest-value users, so CPI improves while revenue does not.
What event should I send to Meta or Google for bidding?
The earliest in-app event that reliably predicts value for your app, such as first purchase, deposit, or trial start. Pick it from your own cohort data, not from whichever event is easiest to fire.
How many postback events should I send per platform?
Three to five high-signal events. Sending more dilutes the algorithm's ability to learn, and sending only installs points it at the wrong objective entirely.
How fast does a revenue signal need to reach the ad algorithm?
Fast enough to land inside the learning window. In Linkrunner data the median first revenue event arrives around 24 hours after install, which is well within range for daily-optimising campaigns.
Where to take this next
Smart bidding is only as smart as the signal you feed it. If your campaigns are hitting CPI targets but revenue is not following, the algorithm is not failing, it is succeeding at the wrong objective. Audit what signal each campaign is actually optimising toward, then switch the ones running on installs to a value event.
To set up reliable revenue-signal postbacks without hand-maintaining them, request a demo from Linkrunner, or start by listing your top three campaigns and the exact event each one is currently optimising for.