Every few months a new channel arrives with a persuasive story: your players are on their living room screens, or on the home screen of a brand new phone, and nobody is advertising to them there yet. Some of those stories turn out to be real. Many turn out to be expensive experiments with no readable result.

This post is about how we decide whether to test CTV (connected TV) as a UA channel for a mobile game, how to think about on-device discovery and other channels beyond app stores, and how to size and structure a test so that you learn something either way. It will not hand you a magic budget number, because the honest answer depends on your game, and we explain why below.

What CTV means for a mobile game

CTV advertising puts video ads on internet-connected televisions, usually inside streaming apps. For app marketers the idea is simple: reach a person on a big screen, then get them to install a game on their phone.

The mechanics are different from mobile. Most TV devices cannot take a tap, so there is no click to follow. Some formats let a viewer start a download to a phone from the remote, and Unity has worked with a streaming platform on exactly this kind of format since late 2023. Others rely on a QR code or on the viewer searching the store later. In all cases the install happens on a different device from the ad, which is the root of every measurement problem below.

Why CTV is hard to measure

Without clicks, you cannot lean on the click-and-install trail that makes mobile campaigns easy to read. Attribution firms describe the same gaps: impressions need independent verification, and there is no direct link between a view on a television and an install on a phone.

That is why the CTV conversation is really a measurement conversation. If you cannot prove an install came from the ad, you cannot say what return the channel produced. Platform-reported numbers help, but they should not be the only evidence.

How do you know if a new channel deserves a test?

Before any money moves, we ask five questions. They apply to CTV and to any new channel.

  • Does the audience fit the game? A cosy puzzle game and a hardcore strategy title do not share viewers. Ask the partner what they know about game players on their surface, and be sceptical of vague reach claims.
  • Can the format carry the game? A fifteen or thirty second TV spot has to explain a game without taps. Games with a clear, visual hook travel better. Our playable ads guide covers hooks that work on a phone, but TV needs its own creative.
  • Can the result be measured? If the partner cannot explain how installs will be tied back to the campaign, treat that as a red flag.
  • Is the target return realistic for the surface? TV is usually closer to awareness than to direct response. Judge it against the return bar you apply to everything else, not a softer one.
  • Is there creative capacity? A new channel needs new assets. If production would starve your existing campaigns, wait.

If a channel fails two or more of these, we do not test it yet.

How much budget do you need to test CTV?

This is a very common question, and any number quoted without context should be treated with caution. The right size depends on three things.

The size of the effect you need to see. A test must be large enough to separate a real lift from normal day-to-day noise in installs. A small effect needs more exposure to detect than a large one.

Your baseline install volume. If your game already gets plenty of organic and paid installs, a small CTV effect disappears inside the background. A game with low baseline volume can detect lift more easily.

The measurement method. Geo tests and holdout tests need enough regions or audience to form a clean control group, which pushes the minimum up.

So the rule we use is this: size the test from the question you want answered, not from a round number. Decide the lift that would make the channel worth scaling, work backwards to the exposure needed to see it, and only then compare that to what you can afford to lose. If the answer is more than you can afford, do not run a half-size test. A test that is too small proves nothing and wastes the budget. Our guide to UA budget planning shows how to ring-fence a test allowance so it never eats into the campaigns that pay your bills.

How to structure the test

A CTV test is closer to a science experiment than to a normal campaign.

  1. Write the hypothesis first. For example: installs in exposed regions will rise above what we see in matched regions, with payback inside our normal window.
  2. Choose a control. Incrementality tests work like an A/B test: one group sees the ad, a control group does not. The common set-ups are geo tests, where some regions are exposed and others are not, tests where the control group sees an unrelated public service message, and predictive models that build a synthetic control from benchmarks. Each has a cost, and geo tests need the most upfront planning.
  3. Hold everything else steady. Do not change other campaigns, store assets or big creative in the test regions during the window.
  4. Run long enough. Players from TV may install later than players from a mobile ad. Allow for that delay before reading results.
  5. Judge on payback. Measure the return of the players, not just cheap installs, using the same window you apply to every other channel. Our post on what a good ROAS looks like explains how we set that line.

Decide the kill rule in advance too. If lift is not detectable by the end of the window, stop. Do not extend a test just because the first numbers looked interesting.

On-device discovery and other channels beyond app stores

CTV is not the only newer channel. On-device discovery places game promotions on the phone itself, through device makers and carriers, for example during the setup of a new phone or in recommendation areas on the home screen. The pitch is that you reach people at the moment they are choosing what to install, before they have settled into their usual apps.

How to fit it into a wider plan:

  • Treat it as a separate test. Give it its own budget line and its own readout, never blend it into an existing campaign.
  • Check the audience. Players who install because a phone suggested the game may behave differently from those who searched for it. Compare retention and revenue, not just install cost.
  • Ask about fraud and quality controls. Ask how installs are verified and what you can see about the placement.
  • Plan an exit. If the cohorts do not reach your payback target, stop and move on.

The same logic covers any other new surface: a clear hypothesis, a clean control, a fixed window and a kill rule.

Where new channels sit in the mix

New channels earn a place only after the core mix is healthy. In our setup Applovin leads, followed by Google Ads, Mintegral and Unity, and each has proved its return over many months. A new channel is a candidate to join them, not a replacement. Before testing, check that your current channels are not hiding a simpler win, such as untapped geos or tired creative. Our piece on spotting channel saturation shows what to look for.

The short version

Test CTV or any new channel when the audience fits, the format can carry your game, the result can be measured and you have creative capacity. Size the test from the lift you need to detect, use a real control group, keep the rest of your activity steady and set a kill rule before you start. If you cannot meet those conditions, the best move is to wait and keep improving the channels already working.