blog · performance marketing · how-to · 12 min read
Setup, structure, and the common mistakes we see in fashion and D2C accounts - the 2026 playbook for feed, asset groups, budgets, and the new reporting.
Calyx Core · july 2026
quick answer
Performance Max for e-commerce works when three inputs are right: a product feed with search-worthy titles and complete attributes, asset groups structured around intent and margin rather than your whole catalog, and conversion values the system can trust. Setup takes a day, learning takes two to three weeks, and ongoing management means weekly search term and channel reviews - now finally possible with PMax's 2026 reporting. Most failures trace back to feed quality or a lazy one-group structure, not to the algorithm.
00
Performance Max is a single campaign that serves your products across every Google surface: Search, Shopping, YouTube, Display, Discover, Gmail, Maps, and since this year, placements in and around AI Overviews and AI Mode. You supply the feed, creative assets, audience signals, and a goal; the system decides where and when to show what.
That division of labor is the whole game. Google runs the auction; you control the inputs. This guide is about making those inputs excellent. For where PMax sits in a full account, see our complete Google Ads for e-commerce guide.
Much of what follows draws on managing PMax for Indian ethnic fashion brands - sarees, lehengas, and occasion wear - a category that stress-tests the campaign type: high average order values, extreme seasonality, hundreds of near-identical variants, and buyers who research for weeks before a wedding purchase. If PMax can be tamed there, your category is easier.
01
02
Variant-heavy categories need feed decisions most guides skip:
03
The default mistake is one campaign, one asset group, all products. That hands the algorithm your entire catalog with zero information about what matters. Our working structure for a fashion store doing meaningful volume:
Rules that make this work:
04
Set target ROAS from margin, per campaign. Premium silk at 60% margin and daily-wear at 35% cannot share one target. That is precisely why they are separate campaigns: the premium campaign can profitably accept a lower ROAS than the volume campaign.
Scale before the season, not during it. The learning phase is the tax you pay on every big change. For wedding and festive season, raise seasonal budgets 6 to 8 weeks ahead in steps of 15 to 20%, so the system enters peak weeks already trained. Doubling budget in the first week of the season resets learning exactly when clicks are most expensive.
Mind the returns problem. Fashion's dirty secret is RTO and returns, especially on COD. PMax optimizes toward reported conversion value, which includes orders that later come back. Feed conversion adjustments or margin-adjusted values if you can, and watch a blended MER (total revenue after returns divided by spend) monthly so platform ROAS never becomes your only truth. If you still run Dynamic Search Ads, migrate deliberately - see our DSA to AI Max migration guide.
05
PMax used to be a black box; in 2026 that excuse is gone. The weekly loop:
One asset group for the whole catalog. No themes, no signals, no accountability. The most common setup and the worst.
Optimizing while the feed rots. SKU-code titles, missing occasion words, flat-lay-only images. The feed is the campaign.
Brand traffic left inside. Inflated ROAS, misallocated budget, and a report that flatters everyone except your bank account.
Judging during learning. Every target change or budget leap restarts the clock. Change in small steps, judge in two-week windows.
Equal-value conversions in a high-AOV category. Value bidding without values is just conversion counting with extra steps.
Building seasonal campaigns in season. The wedding-wear campaign born in November pays learning-phase prices at peak CPCs.
faq
Expect two to three weeks of learning before performance stabilizes, assuming roughly one or more conversions a day. Judge the campaign on rolling two-week windows after that, and remember that big budget or target changes partially restart learning.
As many as you have genuinely distinct themes with distinct creative, and no more. For most stores that is two to five per campaign: category-level bestseller groups plus a seasonal or premium group. Ten groups sharing one set of generic assets is structure theatre.
With 30 or more monthly conversions it is worth testing a hybrid: Standard Shopping on hero products where you want query-level control, Performance Max for reach. PMax takes priority on overlapping products, so structure listing groups deliberately.
Usually brand traffic inside the campaign, or returns and RTO deflating real revenue after the platform counts the sale. Exclude brand terms and track a blended MER after returns alongside platform ROAS.
Yes, with adaptations: accurate order values, a separate campaign with its own ROAS target, video creative, and patience for longer consideration cycles. Bridal buyers click weeks before they buy, so use appropriate conversion windows and judge on full-cycle data.
Performance Max punishes neglect and rewards preparation in equal measure. The stores winning with it in 2026 share the same profile: feeds written in their buyers' language, asset groups with a reason to exist, values the algorithm can trust, seasonal scaling done before the season, and a weekly half hour inside the reports Google finally opened up. None of it is clever. All of it is work the algorithm cannot do for you.
calyx core · pmax audit
We don't just point out problems, we engineer the systems to solve them. We audit feed, structure, values, and the new reports, and hand you a prioritized fix list.