Google is rolling out Campaign Mix Experiments (beta), a new testing framework that lets advertisers experiment across multiple campaign types, budgets, and settings within a single, unified experiment.
How it works:
- Advertisers can create up to five experiment arms, each containing a different mix of campaigns.
- Campaigns can appear in multiple arms, with traffic split between them.
- Experiments support Search, Performance Max, Shopping, Demand Gen, Video, and App campaigns (excluding Hotels).
- Traffic splits can be customized (minimum 1%), with results normalized to the lowest split for fair comparison.
What you can test:
- Budget allocation across campaign types
- Account structure, including consolidation vs. fragmentation
- Bidding strategies, targeting, and feature adoption
- Cross-channel performance interactions, not just single-campaign lift
Why we care. Instead of testing Search, Performance Max, Demand Gen, or Video campaigns in isolation, advertisers can now see how different campaign types work together — and which mix actually drives the best business results.
Reporting details. Results appear in the Experiment summary and campaign-level reporting, with advertisers able to choose confidence intervals (95%, 80%, or 70%) and primary success metrics like ROAS, CPA, conversions, or conversion value.
Best practices:
- Keep experiment arms similar, changing only one variable at a time.
- Align total budgets across arms unless budget is the test itself.
- Avoid shared budgets and major in-flight changes.
- Run experiments for at least six to eight weeks to reach statistical reliability.
Between the lines. This is Google acknowledging that modern performance isn’t about winning one campaign — it’s about finding the right mix, especially as automation blurs the lines between channels.
Bottom line. Campaign mix experiments give advertisers a clearer, more realistic way to test how different campaign types and budgets work together — and make smarter decisions about where spend actually delivers incremental value.