Fact-check
Worth flagging up front: Google's official tools for measuring true, causal incrementality (Conversion Lift, geo-based experiments) measure a campaign or geo region as a whole, not one channel isolated inside a single, already-running PMax campaign. There's no built-in way in the interface to cleanly "cut out" just Display or just YouTube from an overall PMax campaign and honestly measure its separate incremental value.
A quick primer
Conversion Lift is an randomly splits audiences your audience into a group that sees your ads and a control group that doesn't, to measure the causal effect of the ads on conversions. It's supported for PMax alongside other campaign types, but it measures the effect of the campaign (or its geo variant) as one unit. The geo-based version of the same tool splits audiences by geographic clusters (Geo Matched Areas), comparing results in test and control regions, again at the campaign level, not for one channel within it.
The troubleshooting documentation for the channel performance report states the official position directly, and it explains why isolating a channel has limited value even in principle: PMax factors in marginal ROI when allocating budget across channels, meaning it can send budget to a channel with a lower average ROI if that channel offers a better return on the next dollar spent. So even with a perfect isolation tool, the resulting number by itself wouldn't answer "should we keep this channel." That answer always depends on marginal, not average, return at a given moment.
What to check before you touch anything
- Do you actually need a causal, incremental read on the channel, or would observational data from the channel distribution table be enough. The latter is available directly, with no experiment required.
- Is Conversion Lift available in your account. It isn't available to every account; access is granted through a Google representative.
- Do you understand that even geo-based Conversion Lift measures the effect of the campaign in specific regions, not the effect of one channel inside it. Don't conflate these two different levels of isolation.
- Is there enough conversion volume on the channel you're interested in for a meaningful observational read through the channel distribution table, if a full experiment isn't available.
- Are you evaluating a channel by average ROI as an end in itself. Given marginal cost optimization, that's not a methodologically sound basis for a keep-or-cut decision.
Possible approaches
- If you need a causal read on the campaign as a whole, not a single channel, reach out to your Google representative for access to Conversion Lift or its geo-based variant.
- If you're specifically interested in one channel and there's no built-in tool to isolate it inside a single PMax campaign, use the channel distribution table's observational data as an approximation, while being clear about the method's limits (correlation, not proven causation).
- If a channel is important enough to warrant separate evaluation, consider a temporary structural workaround: a separate campaign with a limited asset set focused mostly on that channel, compared to your main PMax campaign through the official PMax uplift experiment.
- Don't conclude a channel should be dropped based on a lower average ROI alone. First weigh its role in the context of marginal cost optimization and the campaign's overall goal.
- Treat single-channel evaluation as a diagnostic input rather than a final decision-making tool. Base the final call on the campaign's overall result against its goals, not an isolated metric for one channel.