Performance Max setup guide

What Performance Max is and isn't

Performance Max (PMax) is a single campaign type that uses machine learning to allocate budget across every Google surface for a single conversion goal. It replaces Smart Display, Smart Shopping, and most Discovery campaigns. The buyer specifies the goal, audience signals, creative assets, and budget; Google decides which surface to serve which creative at what bid.

PMax is not a Search campaign in disguise. It will compete with your branded Search campaigns if you let it, it will serve to placements you can't see in standard reports, and it will reallocate budget across surfaces based on its predicted-conversion math, not your intuition. The operator's job is to set up the campaign so PMax has high-quality inputs and constrained boundaries; the algorithm does the rest.

Before opening the campaign builder: conversion configuration

This is where 80% of PMax outcomes are determined. A campaign launched without rigorous conversion configuration will optimize toward whatever Google thinks "conversions" means, which may not be what you sell.

1. Define one primary conversion goal. One. If you have multiple goals (sale, lead, signup), pick the highest-value one as primary. Mark the rest as secondary or set them to "no value, no bidding." PMax optimizes for what it's told; multi-objective optimization dilutes signal.
2. Set conversion values, not just counts. Use dynamic values from the data layer if you have an ecommerce platform; otherwise, use a thoughtful default (your AOV minus margin loss). Conversion-count optimization buys cheap conversions of any size; value optimization buys the conversions that actually matter.
3. Import offline conversions. If your sale happens off-Google (CRM-closed, store visit, phone call), import the closed-won data via the offline conversion API or a connector. PMax with offline data attached outperforms PMax without by 30-60% in lead-gen verticals.
4. Verify Enhanced Conversions. Hash and pass first-party data (email, phone) to Google so it can recover identity from logged-out users. EC for Web and EC for Leads are both required setup. Audit them in Enhanced conversions.
5. Activate Consent Mode v2. Without consent mode signals, Google can't model conversions from denied-consent traffic, which is 30-60% of EU traffic and growing globally. Consent Mode v2 walks through the install.

Campaign builder: settings that matter

Setting	Recommendation	Why
Bid strategy	Maximize Conversion Value, with tROAS target	Optimizes for the dollar metric you actually care about, not conversion count.
tROAS target	20-30% below your last 90-day blended ROAS	Too aggressive throttles delivery; too soft wastes spend on low-quality conversions.
Location targeting	People in or regularly in your target locations	Default is "or showing interest" — broader than most operators realize.
URL expansion	OFF for most accounts	Sends traffic to URLs PMax decides are relevant. Often includes pages you don't want indexed for the campaign.
Final URL exclusions	Add patterns for /login, /account, /cart, /checkout-success, support pages	Prevents PMax from sending traffic to pages that can't convert.
Customer acquisition setting	Bid higher for new customers (if your business allows)	Forces PMax to find net-new customers rather than re-engaging existing ones.
Brand exclusions	Add your own brand terms	Prevents PMax from cannibalizing your Search Brand campaign.

Asset groups: structure them like SKAGs of intent

Asset groups are PMax's equivalent of ad groups. Each asset group has its own audience signals, creative assets, and listing-group structure for shopping. The biggest setup mistake is one asset group per campaign with everything inside it. Structure asset groups by product line or audience theme — each one becomes a distinct optimization target.

For a multi-product ecommerce brand: one asset group per product collection (e.g., Men's Shoes, Women's Shoes, Accessories). For a lead-gen business: one asset group per offering or vertical. For a single-SKU brand: one asset group per audience theme (prospecting, retargeting, lookalikes).

Audience signals — directional, not directive

Audience signals tell PMax who to start with. They are not targeting; PMax will go beyond them. But signals shape the initial population and accelerate the algorithm's learning. Effective signals:

• Customer Match list of your highest-LTV customers. The strongest possible signal — actual customers who paid full price.
• Customer Match list of high-intent site visitors. Cart-adders, viewers of high-margin SKUs, repeat session visitors.
• Custom Segments built from competitor URLs and high-intent search queries. "People who searched for [competitor product] in the last 7 days."
• In-market audiences relevant to the product. Lower quality than first-party but useful for cold-start.

Avoid layering too many signals. PMax weights them aggregately; if any one signal is poor quality, the aggregate signal is poor quality.

Creative requirements by surface

Surface	Creative needed
YouTube	Vertical video 9:16, horizontal 16:9, square 1:1. Min 10s, ideal 15-30s. If you don't upload video, Google auto-generates from your image and text assets — usually low quality.
Display	Images at 1200×628, 1200×1200, 1200×400 (logo). 15+ images per asset group recommended.
Search	Headlines (up to 15, max 30 chars each), descriptions (up to 5, max 90 chars).
Discover/Gmail	Mobile-optimized images, marketing-style headlines.
Shopping	Product feed in Merchant Center, no separate creative needed in PMax.

The single highest-leverage step most operators skip: upload at least one quality vertical video. Without it, PMax serves auto-generated YouTube creative that performs poorly and absorbs significant impressions.

Reporting and optimization

PMax reporting has improved substantially since 2022. The reports that matter:

• Asset group performance. Identifies which asset groups are pulling weight and which are burning budget on the wrong audience or product set.
• Asset performance. Per-asset combination effectiveness. Pause "Low"-rated assets and replace with new variations.
• Search terms insights. Aggregated themes of search terms triggered. Less detailed than Search reporting but enough to spot brand cannibalization or irrelevant queries.
• Channel performance (via API/scripts). Split out Search vs Display vs YouTube vs Shopping impression share. The native UI hides this; the API and third-party tools (Optmyzr, Adalysis) expose it.
• Audience insights. Which audiences PMax is converting. Use these to refine the audience signals on subsequent asset groups.

The brand cannibalization problem

PMax left to its own devices will serve to brand searches and claim "new conversions" that were already going to convert via your Brand search campaign. The fix is two-step: (a) add your brand terms to the campaign's brand exclusions list (a 2023 PMax feature), and (b) compare PMax's branded-impression share to your Brand search campaign's impression share — if PMax is showing material brand traffic, the exclusions aren't working and you need to escalate via your Google rep.

Performance Max vs standard Shopping

For most ecommerce brands, PMax replaced Smart Shopping in 2022-2023 and standard Shopping is now used selectively. Standard Shopping retains use cases: precise SKU-level bidding, exclusion-heavy strategies (sale-only shopping), and brands that want full search-query transparency. PMax wins on scale, surface diversity, and automation; standard Shopping wins on control.

How long should PMax learn before I judge it?

14-21 days minimum. Less than that and you're judging noise. Make zero changes in the first 7 days unless the campaign is obviously broken (zero spend, zero impressions, broken landing pages). Adjust budget but not structure between days 7-14. After day 21, asset-level and asset-group-level optimization is appropriate.

Should I run PMax alongside Search?

Yes, with brand exclusions on PMax. PMax for prospecting and broad-funnel; Search for branded and high-intent nonbrand. Without brand exclusions, PMax cannibalizes Search Brand and inflates its own attributed conversions.

What's the minimum budget for PMax?

Google recommends 10-15x your target CPA per week for learning. For a $50 target CPA, that's $500-$750/week. Below that, PMax doesn't get enough signal to optimize. Smaller budgets are better off in standard Search.

How do I prevent PMax from spending on retargeting?

Exclude your customer list and recent-purchaser list from the campaign via audience exclusions (added as a PMax feature in 2023). Enable the customer-acquisition setting to bid lower for existing customers. Run separate retargeting in a Demand Gen or remarketing campaign with explicit targeting.

Why does PMax love sending traffic to Display?

Display impressions are cheap and Display conversions are over-credited by Google's last-click attribution. PMax's algorithm sees high attributed-conversion rates on Display and allocates more budget there. Counter this with rigorous offline conversion data, incrementality testing, and channel-level impression-share monitoring via API.

Can I see exactly where my PMax ads are serving?

Partially. Asset group performance and placement reports show display and YouTube placements at a high level. Search term insights show themes, not individual terms. For full transparency, run controlled tests with placement exclusions enabled and disabled and measure the delta.

Operating checklist

1. Define the conversion event and value model before campaign creation.
2. Onboard first-party data via Customer Match and offline conversion imports.
3. Set the bid strategy and target based on actual margin, not aspiration.
4. Validate creative requirements per ad type before launching.
5. Monitor first 7 days closely; resist the urge to edit before 14 days.
6. Pull search-term and asset-group reports weekly.
7. Document the campaign architecture in a runbook for the next operator.