How I approach Performance Max cleanup for ecommerce brands

A messy Performance Max account rarely has one dramatic setting that fixes everything. The visible symptoms, including volatile ROAS, uneven product spend, weak search relevance, or confusing reporting, usually sit on top of several connected issues. Conversion inputs may be unreliable. Campaigns may have overlapping jobs. Branded demand may be carrying the average. Product data may limit matching before bidding gets a chance to work.

I approach cleanup as a remediation sequence rather than a wholesale rebuild. I first protect measurement and account safety, then clarify campaign roles, then improve the inputs and controls around those roles. Only after that do I make aggressive budget or target changes. This order gives each change a better chance to produce interpretable evidence.

The objective is not to make PMax look sophisticated. It is to give the campaign a clear business purpose, dependable conversion signals, suitable products and assets, and enough surrounding context that I can explain whether the next dollar belongs there.

Triage the account before rebuilding anything

I begin with a map of the current system: conversion actions, PMax campaigns, product coverage, brand Search, Standard Shopping, budgets, targets, geographies, customer settings, feed sources, and major recent changes. I note immediate risks such as duplicate purchase conversions, account-level Merchant Center warnings, broken landing pages, or campaigns spending outside the intended market. Those take priority over structural preferences.

Then I identify what is merely untidy and what prevents a sound decision. Duplicate asset groups may be inefficient to maintain, but a doubled purchase value can invalidate the entire optimization system. A weak audience signal may matter less than hundreds of disapproved best sellers. Triage keeps the cleanup focused on consequences instead of interface completeness.

Validate conversion inputs first

PMax bidding reacts to the primary conversions and values it receives. I verify that purchases count once, use the correct currency, include the intended revenue components, and pass stable transaction IDs. I check whether add-to-cart, page views, imported analytics events, offline conversions, or other actions are accidentally marked as primary. I also review enhanced conversions, consent behavior, attribution settings, and recent tag changes.

I compare Google Ads with the ecommerce platform and analytics directionally across multiple periods. Exact agreement is not required because attribution and processing differ, but unexplained shifts are a warning. If platform revenue jumps while commerce revenue does not, I investigate measurement before celebrating the campaign. If consent coverage changed, I record that context before comparing pre- and post-change efficiency.

Treat new-customer settings as inputs, not truth

New-customer acquisition goals can be useful when customer lists, definitions, and values are dependable. They can also overstate acquisition when known customers are not recognized or when the assigned value is arbitrary. I check how the business defines new, what the account can identify, and whether the extra value reflects real economics. I do not use the setting as a substitute for broader customer and incrementality analysis.

Give every campaign one written role

For each PMax campaign, I write a short purpose: core catalog efficiency, new-customer acquisition, high-margin category growth, seasonal promotion, inventory clearance, or another real business job. I compare products, geography, audiences, URLs, conversion goals, budgets, and targets across campaigns. When two campaigns have the same inputs and objective, I question whether the overlap is providing control or just dividing data.

Consolidate when the work is truly the same

Consolidation can increase useful volume and reduce maintenance when products share economics, targets, and purpose. I do not consolidate categories that need different budgets merely to create a cleaner diagram. Likewise, I do not split campaigns because a custom label exists. Structure should preserve decisions the business needs to make and remove divisions that cannot be maintained or evaluated.

Separate when economics or intent require control

A high-margin product line, a short promotion, a market with different shipping economics, or an acquisition-specific initiative may justify separation. I define the reason before creating the campaign and decide how success will be read at account level. Separate campaigns are not automatically independent; product overlap, brand demand, and attribution can still connect their outcomes.

Diagnose brand leakage and query quality

PMax can convert branded searches efficiently, which can make generic performance appear stronger than it is. I review brand Search coverage, search-term insights, landing pages, product names, returning-customer patterns, and changes in brand demand. If the evidence suggests meaningful overlap, I establish a properly funded brand Search baseline before testing exclusions.

My full brand-separation framework covers baseline creation, exclusion options, test design, and account-level interpretation. The important point during cleanup is that campaign ROAS alone cannot tell me whether PMax is creating demand or collecting it.

Query review goes beyond individual terms. I group recurring themes such as compatibility, jobs or research intent, competitor searches, informational questions, irrelevant materials, and low-value product types. Search-term insights are incomplete, so I combine them with Search data, product patterns, landing-page traffic, and business context. The goal is to identify actionable themes, not claim visibility that the platform does not provide.

Repair Merchant Center and feed inputs

PMax cannot advertise a product that is missing or disapproved, and it cannot reliably match a vaguely described one. I review Merchant Center account health, active product coverage, identifiers, titles, categories, variants, prices, availability, shipping, images, and landing-page consistency. I prioritize account safety and commercially important inventory before low-impact warnings.

I use the same feed health checklist for PMax cleanup because campaign recommendations are only as trustworthy as the catalog they evaluate. A feed rule that silently removes a category can look like a bidding problem in the ad account.

Create product groups the business can act on

I use product type, category, margin, inventory, season, bestseller status, promotion, or another maintainable label when it supports a distinct decision. I check whether low-volume products are being split into groups too small to learn and whether high-priority products are buried in catalog-wide averages. Product grouping should improve budget control and analysis without turning normal merchandising changes into constant campaign maintenance.

• Separate products when margin, inventory, promotion, or strategic priority requires a different budget or target.
• Keep product overlap deliberate and document which campaign should serve which inventory.
• Use labels sourced from stable catalog logic, not manual values that will decay after launch.
• Evaluate both spend concentration and the eligible products that receive no meaningful traffic.

Review assets, URLs, and audience inputs

I inspect text, image, logo, and video assets for accuracy, product fit, promotional currency, and policy safety. A platform strength score can point to missing formats, but it cannot determine whether the creative makes a persuasive promise. I also check whether asset groups align with the landing pages and product sets they are meant to support.

Final URL expansion, page feeds, excluded URLs, language, location options, and destination settings can change where traffic lands. I test the actual pages, including mobile behavior and variant selection. If the campaign can route shoppers to irrelevant articles, sold-out collections, or unsupported regions, I narrow the controls rather than hoping automated selection will resolve the mismatch.

Use audience signals as direction, not a fence

Customer lists, site visitors, search themes, and relevant segments can provide useful starting information, but PMax can reach beyond them. I check list freshness, consent, size, and the logic behind each signal. Adding many loosely related audiences to complete a setup checklist does not create a strategy and can make future review harder.

Sequence changes so the result remains readable

My usual order is measurement, urgent Merchant Center issues, campaign roles and overlap, brand handling, product grouping, assets and URLs, then bidding and budget tests. I may adjust that order for account safety or a time-sensitive promotion, but I record the reason. The change log includes dates, exact settings, expected effect, evaluation window, and rollback conditions.

Consider a hypothetical catalog where one PMax campaign spends mostly on branded best sellers while a new category receives little traffic. I would not immediately split every category. I would first validate purchases, fund brand Search, inspect feed coverage, and define the core campaign's role. Then I could test a priority-category campaign with controlled overlap and a budget tied to its economics. The sequence prevents the new structure from inheriting the original ambiguity.

Monitor the account after cleanup

Cleanup is complete only when the operating rhythm changes. I monitor conversion volume and value, total cost, brand and non-brand context, product eligibility, category contribution, asset or URL drift, budget pacing, and major query themes. I watch the account and business together because restored inventory, promotions, or site changes can alter mix even when campaign settings remain stable.

I also review what receives no spend. A strong average can hide eligible products that never enter meaningful auctions. Some deserve no additional budget; others may expose weak titles, low competitiveness, incorrect grouping, or a target that prevents exploration. The point is to understand the choice rather than assume automation has evaluated every product equally.

When I recommend a larger rebuild

A larger restructure may be justified when conversion goals conflict, product overlap is pervasive, geography or economics require separate control, feed taxonomy cannot support decisions, or historical campaign roles are impossible to recover. Even then, I phase the migration when possible. Preserving a usable comparison and protecting revenue matter more than launching an idealized account diagram overnight.

Performance Max is only one part of my ecommerce Google Ads management. If you need diagnosis before ongoing work, book a discovery call to discuss measurement, campaign structure, Shopping feeds, and wasted spend.