Performance Max is an optimiser, not a strategy. It takes the goals, values, products, pages, creative and constraints you supply, then bids across Google's inventory. The quality of those inputs determines whether the campaign is solving a useful commercial problem or merely producing an attractive number in the interface.
This guide is ordered by consequence. Fix measurement and economics before polishing assets. A beautiful campaign optimising to the wrong conversion is still wrong.
1. Define one commercial job
Write the campaign's job in one sentence: "Acquire first-time UK customers for full-price skincare at a contribution-positive cost" is useful. "Increase sales" is not. The sentence should resolve product scope, geography, customer type and the financial outcome.
If a campaign contains products with incompatible margins, separate regional fulfilment or different acquisition rules, one shared bid target may be inappropriate. Campaign consolidation helps data density, but only when the underlying economics can share an objective.
2. Audit conversion goals before launch
Open Goals > Conversions > Summary. For each action, verify its source, status, counting method, attribution setting, value and whether it is Primary or Secondary. Then open the PMax campaign's goal settings and confirm it optimises only to the outcomes intended for that campaign.
For ecommerce, test a real purchase and reconcile the transaction ID, currency, tax, shipping and order value with the platform. Duplicate purchase events or static values distort value-based bidding. For lead generation, a submitted form is not necessarily a qualified lead. Import later-stage outcomes where possible and use values that reflect expected business value rather than assigning an arbitrary large number to every enquiry.
Observation and inference must stay separate. Google Ads can observe a recorded conversion and its attributed value. It cannot observe your gross margin, cancellation rate or lead quality unless you send that information back.
3. Translate economics into bidding inputs
Start with the maximum acquisition cost the business can afford, not a competitor's ROAS benchmark. For retail, calculate contribution before ad spend: net revenue minus product cost, fulfilment, payment fees, expected returns and variable discounts. For lead generation, estimate value by lead stage using close rate and expected gross profit.
Hypothetical ecommerce example: an order produces £120 net revenue and £48 contribution before ads. If the company needs £12 contribution after ads, the maximum ad cost is £36. That is a 333% revenue ROAS target (£120 / £36). This is a planning calculation, not a recommended universal target.
Launching with an aggressive target ROAS can suppress delivery when the campaign has little relevant history. Launching without a target can spend more freely than the business can tolerate. Choose deliberately, set a risk limit and document when a target will be reviewed.
4. Make the feed operationally reliable
Retail PMax depends heavily on Merchant Center. Check Merchant Center > Needs attention and Google Ads > Campaigns > Products for disapprovals, out-of-stock items, missing identifiers and products falling into unintended listing groups. Prioritise issues by lost commercial opportunity, not raw error count.
Titles should identify the product clearly with attributes customers use to distinguish it. The product feed optimisation guide goes deeper on attribute priorities and feed operations. Images should show the actual item without promotional overlays that breach policy. Price, sale price, availability, shipping and destination data must match the site. Use product type and custom labels for business segmentation, while preserving Google's required attributes for eligibility.
5. Structure around targets and decisions
Use separate campaigns when you need separate budgets, bid targets, countries, goals or customer-acquisition settings. Use asset groups for coherent creative and product or URL themes within the same campaign. Do not create a campaign for every category merely to make the account look organised.
Before splitting, ask whether each part will receive enough conversion value to support a stable decision. A tiny "high margin" campaign that receives one sale a month may be commercially logical but statistically erratic. Sometimes the better answer is a broader campaign plus feed labels and a monthly product profitability review.
6. Build a creative system, not an upload checklist
Give each proposition a claim, proof and action. Produce landscape, square and vertical formats with the subject inside safe areas. Include product demonstrations, customer evidence, concrete differentiators and a clear next step. Avoid ten near-identical headlines that all say "quality products at great prices."
Ad Strength can reveal missing formats or insufficient variety, but it is not a profitability score. Asset labels and combinations are observational. Google's asset-group reporting documentation advises evaluating asset groups alongside campaign-level performance rather than removing a group from average CPA or ROAS alone.
7. Configure search and destination controls
Use search themes to add context that the feed and site do not clearly express. Review the granular Search terms report for irrelevant demand, brand dependence and useful language. Google's PMax search-terms guide confirms that search terms can be reviewed and excluded; historical data is available from March 2023.
Use brand exclusions when the reporting question requires separating branded demand. Google says PMax brand exclusions apply to Search, Shopping and YouTube Search inventory, with a separate option controlling Shopping ads for excluded brands. They do not remove every possible branded exposure on every channel. See the brand-exclusion documentation.
Final URL expansion is on by default and may select another page from the domain and generate matching text. Keep it on when the site is clean and the extra reach is useful; add URL exclusions for careers, policies, support and irrelevant editorial pages. Turn it off when traffic must reach a specific destination. A page feed with expansion on is a signal, not a strict allow-list.
8. Treat audiences and customer rules precisely
Audience signals are starting information, not hard targeting. Add consented first-party lists that represent real value, such as repeat customers or qualified opportunities. Avoid presenting a signal as proof that the campaign only reached that group.
If using new-customer acquisition, define existing customers with the strongest available lists and verify the mode selected. "Bid higher for new customers" and "new customers only" represent different risk. Compare reported new-customer performance with CRM or order data because Google's classification depends on the signals available to it.
9. Check what can actually serve
Open Insights & reports > Channel performance. The Status column can identify missing assets, disapproved assets, product issues and channels that are not eligible or eligible with limitations. Segment by ads using product data or video where available. This is the fastest way to distinguish "Google chose not to spend here" from "the campaign could not serve here."
The official channel report guide documents Search, Display, YouTube, Discover, Maps, Gmail and Search partners, plus channel-level conversion-action and ad-event-type segments. Channel allocation is observable; the counterfactual return from forcing a different allocation is not.
10. Launch with a written control sheet
- Primary conversion actions and values tested
- Campaign job and profitability threshold documented
- Countries, languages and location options checked
- Products or URLs explicitly included and excluded
- Brand, negative keyword and content-suitability rules reviewed
- Final URL expansion and URL exclusions reviewed
- All asset formats previewed and policy-approved
- Budget downside and first review date agreed
- Annotations or change log ready
A reporting rhythm that produces decisions
Weekly, check spend, conversion lag, goal mix, product eligibility, channel diagnostics, search terms and major landing-page changes. Monthly, reconcile attributed revenue or leads with backend outcomes, assess product and customer mix, review creative fatigue and decide whether the campaign still deserves its marginal pound.
When results change, follow an order: confirm tracking, inspect change history, check eligibility and budget, segment conversion actions, review channel and search-term movement, then investigate market or site factors. Do not jump from a ROAS decline to a story about "the algorithm" without observable evidence.
What a well-built campaign still cannot guarantee
PMax reporting is more transparent than it used to be, but attribution is not incrementality. A reported sale may have occurred without the ad. Asset and audience reports are not randomised tests. Channel averages can hide interactions between touchpoints. Strong setup makes the system accountable enough to operate; it does not remove uncertainty.
Use Google Ads experiments when the available design matches the question, and use geographic tests, customer holdouts or broader measurement when the commercial decision requires causal evidence.