The Maze: A client marketing-mix model — a statistical estimate of how channels contribute to revenue — turned branded Google search into the most expensive receipt in the account. Google Brand took £45,000 and was assigned £9,000 of revenue: 0.20 ROI, or £1 back for every £5 spent. Google Generics returned 2.35 and Meta direct response returned 3.20. This does not prove every brand campaign is waste. It proves that a converting click can be very different from a sale caused by advertising.
The platform is not the problem; the job is. In the client example, Google Brand used 2.8% of total spend but produced only 0.2% of modeled revenue. Google Generics used 26.7% and produced 24.3%. One catches people already asking for the brand. The other can intercept category demand before the shopper chooses. Lumping both into `Google` hides an 11.8-times ROI gap and turns a budget decision into a platform opinion.
High intent can impersonate advertising lift. Someone typing a brand name is already unusually likely to buy. Ordinary attribution gives the paid click credit; incrementality asks whether the purchase would have happened without the ad. A large eBay field experiment found no measurable short-term benefit from brand-keyword ads and showed why non-experimental returns can be inflated by pre-existing purchase intent. That mechanism fits the 0.20 result. It is evidence for testing, not permission to copy eBay's answer into every account.
A model can expose waste without pricing the insurance. The client, country, category, model window and uncertainty range are undisclosed. More importantly, the estimate does not isolate competitor interception, branded Shopping, purchases that move to Amazon or another reseller, lost reseller margin, or the value of owning the customer record. Google's own MMM guidance says experiments and models often measure different counterfactuals and should be combined carefully. The right conclusion is not `brand protection is fake`. It is `brand protection needs its own invoice`.
Measure the switch, not the argument. Separate brand text ads from generic search and branded Shopping. Use a holdout, geo split or staged reduction large enough to detect movement. Watch organic and direct sales, competitor impression share, retailer revenue, new-customer mix, contribution margin and customer-data capture — not only Google Ads conversions. Google's Conversion Lift framework makes the logic explicit: incremental value is treatment minus control, not attributed revenue divided by spend. Brand exclusions can keep branded queries out of Search or Performance Max while the test runs.
Why it matters: Branded search should be funded like insurance, not celebrated like growth. A campaign with 0.20 modeled ROI needs to prove that the margin it protects, the customers it keeps direct and the competitors it blocks are worth more than the spend. Some brands will pass that test. Others will discover that organic search was doing the work while paid media collected the trophy. Move the money only after a clean holdout — but stop granting brand campaigns an automatic exemption from causal measurement.
