Skip to content
Vendor reportCassandra, 2025

Non-brand search returned the highest median ROI of seven Google Ads campaign types, at 5.21x

Channel validation


Campaign TypeNumber of modelsSpend (USD)Median incremental ROI (x)CI low (x)CI high (x)
Search Brand85$87,000,0004.141.957.78
Search Non-Brand60$95,000,0005.212.299.89
Performance Max48$66,000,0004.643.116.41
Shopping16$68,000,0003.091.725.25
Video / YouTube14$3,600,0002.71.54.2
Display15$37,000,0002.740.678.2
Demand Gen15$27,000,0002.221.444.91

Incremental ROI from marketing mix models by campaign type, with 95% CI of the median

Reader Note:
ROI here is incremental revenue per $1 of ad spend, estimated by Cassandra’s marketing mix models. Cassandra sells MMM software, and these figures are usually lower than the ROAS shown in Google Ads. Non-brand search means ads on generic queries (“running shoes”), not your brand name. It is based on 60 models and $95M in spend in 2023–2025. The confidence interval is the widest among search formats, [2.29–9.89], because results depend heavily on keywords and landing pages. The other six types: Performance Max 4.64x, Search Brand 4.14x, Shopping 3.09x, Display 2.74x, YouTube 2.70x, Demand Gen 2.22x.
Evidence:
Modelled / projected
Vendor interest:
High — This comes from a company with a commercial interest in this market, so it is not a neutral source.
Sample:
253 revenue-outcome marketing mix models from 59 unique advertisers, covering $383M in media spend; data spans 2023-2025. Each record is one channel from one model run.

Good for

Planning with a rough expectation, and sanity-checking your own assumptions.

Not for

  • Reporting as what happened. This is what a model expects, not what anyone counted.
  • This comes from a company with a commercial interest in this market, so it is not a neutral source.
  • Nobody counted this directly — it's a calculation. The publisher worked it out itself, or took the ad platform's word for what its ads achieved.
  • This is an aggregated or average number that covers very different cases. Real results in a particular segment may differ.
  • The source says how much it looked at, but not how any of it was chosen.

Coming soon

This feature isn’t available yet.