In game inventory is one of the few places left where attention is genuinely undivided. Someone watching a rewarded video has chosen to watch it, is not scrolling past it, and will be there until it finishes. That advantage is easy to waste. Most underperforming campaigns we audit are not failing because the inventory is poor, they are failing because unity ads targeting was set once at launch, broadly, and never revisited while the algorithm quietly optimised toward the cheapest possible install.
The people who watch an ad to the end and the people who become valuable users overlap far less than a completion rate suggests. Getting from one group to the other takes deliberate structure: the right placement types, a signal that reflects value rather than volume, and enough patience to let the system learn against an event that actually matters. Here is how we build campaigns that hold up after the first fortnight.
Placement type is the first targeting decision
Before any audience setting, decide what kind of moment you want to buy. Rewarded video reaches a player who has opted in for an in game benefit, which means high completion and attention but also a motivation that has nothing to do with your product. Interstitials interrupt at a natural break. Banner and other formats sit at the edge of play. These are not interchangeable slots at different prices, they are different psychological situations.
For direct response, we generally start with rewarded and interstitial inventory and judge them separately from day one, never blended in a single reporting line. Rewarded traffic tends to arrive in higher volume and at a lower nominal cost per install, and tends to retain worse. Interstitial traffic is more expensive and often more valuable. A campaign that reports only the blended figure will drift toward the cheaper source and slowly get worse while the headline number improves.
Unity ads targeting settings that carry weight
Not every control deserves equal attention. These are the ones that consistently change outcomes in the campaigns we manage.
- Geography, split into its own campaigns rather than bundled, because bid dynamics and user value differ enormously between markets and a shared budget will starve the valuable one.
- Device and operating system tier, especially where your app is heavy or your creative relies on a smooth render on older hardware.
- Game category and individual app exclusions, built from performance data rather than assumption, and reviewed at least monthly.
- Optimisation event, which should be the deepest reliable action you have enough volume on, not simply the install.
- Creative variant, treated as a targeting lever in its own right since the ad itself selects who responds.
That last point is underrated. In game audiences are broad, and the strongest filter available is often the first three seconds of the video rather than any checkbox in the interface.
Optimise toward value, not toward installs
If the campaign is told to find installs, it will find the people most likely to tap install, and those are frequently people who install a great deal and keep very little. The fix is to hand back a later event: registration completed, tutorial finished, first purchase, a defined activity threshold in the first week. Whatever you choose must fire reliably and often enough for the system to learn from it, which is the real constraint on small apps.
When volume on the deep event is too thin, we use a two stage approach. Optimise toward an intermediate event that correlates with value and occurs frequently, while reporting on the deep one, then move the optimisation target once daily volume supports it. This is slower than it sounds on paper and worth the discipline. Moving the target too early leaves the campaign learning forever and never reaching stable delivery.
Creative is part of the targeting system
In game viewers are already in an interactive mindset, which is why playable and gameplay led creative outperforms polished brand film so consistently in this environment. Show the thing working in the first seconds. Match the visual register of the surrounding experience closely enough to feel native, without misleading people into thinking they are still playing the game they were in, which produces clicks you pay for and users who resent you.
We run a small rotating set of variants rather than one hero asset, because fatigue arrives faster here than on most channels and because variants reveal audience segments that no targeting control exposes. A version that emphasises competition and a version that emphasises collection will pull different people from the same inventory, and the retention curves usually separate within two weeks.
Reading the data and cutting what does not work
The reporting habit that matters most is cohort based. Group users by install week and by source app, then look at what percentage are still active on day seven and day thirty, and what they contributed. Cost per install tells you what you paid. Only the cohort tells you what you bought. Source level exclusion lists built from that view, applied steadily, are where most of the efficiency in a mature in game campaign comes from.
Effective unity ads targeting ends up looking less like clever audience configuration and more like a routine: separate what should be separate, optimise toward the deepest event you can support, refresh creative before it fades, and keep pruning the sources that deliver watchers rather than users. Done every month, it compounds. Done at launch and forgotten, it decays.
Keep reading: Unity Ads · Advertising