Performance Max Rebuilt With Brand and Feed Under Control

Most Performance Max campaigns we take over are reporting a return in which a third to a half of the revenue is brand search wearing a different label. Before we change anything, we measure that split, and the number usually decides the shape of the rest of the engagement. Getting brand out of the number: Campaign-level brand exclusion lists and account-level negative keywords both work in Performance Max now. We apply them, and reported ROAS falls. That is the point. The campaign stops being paid for orders it did not cause. Where you need harder evidence than a before-and-after, we run a brand-term holdout across matched German regions for three to four weeks and compare total non-brand revenue rather than campaign-attributed revenue. Asset groups built on margin, not on your category tree: A retailer's product hierarchy exists for merchandising. Copying it into Performance Max produces asset groups whose only shared property is that a buyer once put those items on the same page. We split on the economics instead. In fashion, full-price core, carryover staples and end-of-season markdown are three different target problems and should never share a budget. In wellness, a subscription-first SKU is a fourth, because its acceptable acquisition cost is set by the second order and not the first. Each asset group gets its own target, its own listing group, and its own argument. Search themes, used sparingly: Search themes are signals, not keywords, and twenty-five per asset group is a licence to dilute. We add a small number, drawn from queries we can already see converting in your search campaigns, and we remove any that pull Performance Max onto terms the search side covers more cheaply and with more control. A feed-only control group: For retail accounts we keep one feed-only asset group running alongside the asset-rich ones. It is the cleanest available read on what the creative is actually contributing. In roughly a third of the catalogues we work on, the honest answer for the long tail is that it contributes very little and the budget belongs back in standard Shopping. Creative cadence: Performance Max will assemble combinations your brand team would not sign off, and asset-level reporting is too thin to litigate individual headlines. We judge at asset group level, replace the weakest concept on a fixed monthly cadence rather than reacting to noise, and switch off automatically created assets and auto-generated video. A slideshow of product photos with stock music is not a brand asset, it is a placeholder that shipped. What we refuse: Performance Max as the whole account. It absorbs whatever demand exists and reports it back as growth, which is comfortable and tells you nothing. We keep search, Shopping and Performance Max separable so the account can still answer questions a year from now. We also refuse to set a target ROAS from a blended figure. If the number came out of a spreadsheet mixing brand, returning customers and new customers, we ask for the split first, and if it does not exist we build it before touching bidding. Not included: Merchant Center and feed engineering, which is separate work done by different people. Video and photography. Landing page changes. Attribution modelling beyond Google Ads and GA4. Who this is not for: Accounts that need placement-level transparency and inventory control. Performance Max does not offer it, and that conversation is DV360 and probably a different agency. Retailers whose feed is disapproved or thin, where the campaign will fail for reasons that have nothing to do with its settings. And anyone whose question is simply how to get Performance Max ROAS up. We can do that in an afternoon by letting it eat more brand, and it will not make the business a euro.