Performance Max campaigns that actually work: 11 levers we pull every week
The PMax paradox
Performance Max is the campaign type Google wishes everyone would use and most advertisers quietly resent. It promises to unify Search, Shopping, Display, YouTube, Discover and Gmail into one campaign, optimized end-to-end by Google's algorithm. In practice it can be either the best or the worst campaign in the account, and the difference comes down to operator skill. After spending more than fifteen million dollars in PMax across our client portfolio, here are the eleven levers we pull every week to keep ROAS climbing.
1. Feed quality is half the work
For any account with a product feed — e-commerce, travel, real estate — PMax performance is ninety percent a function of feed quality. Titles must contain brand, product, size, color and category in that order. Descriptions must be unique per SKU. GTINs must be valid. Custom labels must segment products by margin, by velocity and by seasonality. Before you touch the campaign itself, spend a week on the feed: it is the single highest-leverage investment you can make.
2. Separate brand from non-brand
By default PMax will cannibalize your brand traffic and take credit for conversions you would have got for free. The fix is to add your brand terms as account-level negative keywords on PMax (now officially supported, finally) and run a separate Search brand campaign. This alone often reveals that PMax's "real" ROAS is 30 to 50 percent lower than reported — painful, but necessary to make good decisions.
3. Use asset groups as audience containers
Asset groups are not just creative collections; they are containers that let you focus the algorithm on a theme. We typically run one asset group per major product category, each with its own creative, audience signals and landing page. The mistake we see is one giant asset group with forty assets and no theme — the algorithm has nothing to anchor to and performance is mediocre.
4. Audience signals matter more than people think
PMax does not require audience signals, but it learns much faster with them. Feed it your first-party customer list, your high-value remarketing audiences and your detailed custom segments. Treat audience signals as a starting hypothesis: the algorithm will explore beyond them, but it will explore intelligently.
5. Optimize creative on a weekly cadence
The fastest-decaying asset in any PMax campaign is the video. Refresh videos every three to four weeks, images every six, and headlines every two. We keep a content calendar with one new video per week per active campaign — usually a fifteen-second cutdown of an existing brand asset, reframed for the specific product category.
6. Bid strategy: tROAS, then tCPA, then Max Conversions
Always start a new PMax with Max Conversions (no target) for two weeks while it gathers data. Once you have at least thirty conversions, switch to tCPA at your historical average. Once you have a hundred conversions, switch to tROAS. Never start with tROAS on day one — you will starve the campaign of impressions and it will never escape the cold-start phase.
7. Mind the conversion goal hierarchy
Account-default goals override campaign goals unless you explicitly tell PMax otherwise. We always uncheck "Use account-default goals" and select only the goals that matter for that specific campaign. If you want PMax to chase Purchases, do not let it count Add-to-Carts.
8. Geo and ad-schedule observations are gold
PMax does not let you exclude geographies or schedules directly, but you can see how each performs in the Insights tab. Use this data to inform separate Search campaigns that do allow exclusion, or to argue for a higher tROAS during off-peak hours. Treat PMax as one input into a larger account strategy, not an island.
9. Use the Insights tab as a research engine
The Insights tab tells you which search themes PMax is finding, which audiences are converting and which assets are performing. Mine it weekly: every emerging search theme is a candidate for a dedicated exact-match Search campaign, every winning asset is a template for the next round of creative production.
10. Cap budgets to force focus
PMax algorithms over-spend if you let them. Set daily budgets at no more than 5x your target CPA and let the algorithm earn budget increases by hitting the target. Loose budgets lead to wasteful exploration; tight budgets force the algorithm to find efficiency.
11. Pair with a structured Search campaign
PMax is at its best when it is the complement to a tightly-structured Search campaign, not the replacement. Run exact-match Search on your top one hundred converting keywords with manual control, and let PMax mop up the long tail and the visual surfaces. Accounts that go "PMax-only" almost always regret it within three months.
How we measure success
The single metric we report to clients on PMax is incremental ROAS, not headline ROAS. We periodically run lift tests by pausing the campaign in a holdout geography and comparing revenue. If headline ROAS is 6.0 but incremental ROAS is 1.8, you do not have a PMax success story — you have a brand-cannibalization story. The eleven levers above exist to push incremental ROAS up, not just the number Google likes to show you.
Want a PMax audit? We do fixed-price audits with a full report and a prioritized backlog. Get in touch.
Bonus: the weekly PMax operating cadence
Tactics matter, but cadence matters more. The accounts that win at PMax are not the ones with the smartest single insight; they are the ones that show up every week with the same disciplined checklist. Our team runs the same Monday ritual on every account: review last week against target, mine the Insights tab for new search themes, queue at least one new creative asset for production, check feed-disapprovals in Merchant Center, validate that conversion tracking still fires, adjust tROAS by no more than ten percent. Small, consistent moves compound; big, occasional moves break the algorithm and reset the learning phase. Treat the campaign like a garden that needs weekly tending, not a machine you set and forget, and the results follow.
One final reminder: PMax rewards patience. The algorithm needs four to six weeks to fully exit the learning phase after any significant change — new bid strategy, new asset group, new feed structure. Resist the temptation to tear things up after a bad week. Define a clear hypothesis before you change anything, give it a fair test window, and document the result. The advertisers who lose at PMax are almost always the ones who panic and pivot weekly. The ones who win pick a strategy, work the eleven levers above, and let compounding do its job.