SEA
14 minutes
17.11.2026

How to Structure Performance Max for Online Shops

Top sellers, challengers, slow sellers: segmentation by data and contribution margin

Updated on

Benjamin Uhlmann

Ich bin Benni, Gründer der Agentur pan pan. Gemeinsam mit meinem Team unterstütze ich E-Commerce-Unternehmen dabei, ihren Umsatz dort zu steigern, wo ihre Kunden suchen.

Performance Max campaign for an online shop in Google Ads

In brief

Most online shops need fewer PMax campaigns than they run. Each campaign should get at least 50 to 80 conversions a month, and below 30 no split is worth it. Sort products into top sellers, challengers and slow sellers by data with a script, using one target ROAS derived from your contribution margin. The biggest lever in e-commerce is the feed: a median of 90 percent of PMax spend goes into feed-based ads.

This guide is for online shops and paid search managers. It shows how many campaigns make sense at which conversion volume, how we segment products, how to manage target ROAS and budget, and which of the new features already work in Germany and the EU.

At a glance

  • Under 30 conversions a month, stick to one campaign. In our accounts, Google only hits the target ROAS reliably from around 150 conversions per campaign.
  • No negative verdict without data: products without enough data belong in the challenger campaign, not with the slow sellers.
  • One target ROAS for all campaigns, derived from CM3. Margin splits rarely work in practice.
  • Since 17 August 2026, budget-limited target ROAS campaigns pay a median of 15.8 percent more per click. Don't cap the budget.
  • PMax is far more controllable in 2026: up to 10,000 negative keywords per campaign, plus search term and channel reports.
  • Some new features are not live in Germany and the EU yet, such as ads in AI Mode or title customisation in AI Max for Shopping.

Where does PMax budget go in an online shop?

Performance Max is a Google Ads campaign type in which Google's AI controls bids, placements and ads across all channels: Search, Shopping, YouTube, Display, Discover, Gmail and Maps. In e-commerce, the money still ends up in the same place almost every time. Smarter Ecommerce analysed more than 4,000 retail campaigns: between 74 and 97 percent of spend goes into feed-based ads, with a median of 90 percent (smec, April 2025). That holds even for campaigns running with a full set of image and video assets.

What this means for you: the structure decides which products get how much budget. The feed decides how well they are served. Titles, product types, images and clean availability data matter more than any new headline. If products are being disapproved in Merchant Center, our guide on fixing disapproved products in Google Merchant Center will help.

What you control yourself is limited: the feed, conversion goals and their values, target ROAS, audience signals, exclusions and how products are split across campaigns. This guide is mostly about the last point.

How many Performance Max campaigns does a shop need?

As few as it takes for each campaign to get enough conversions for Smart Bidding. Google itself recommends broad consolidation for retail, with separate campaigns mainly for different ROAS targets or country budgets (Google Ads Help). smec sees stable campaigns from 30 conversions a month, better from 60, and found no evidence that PMax campaigns learn from each other (smec). Every split really does divide the data.

Our own numbers are a little higher. We recommend 50 to 80 conversions per campaign per month, ideally 100. In an internal analysis of PMax campaigns over 30 days, Google only hit the target ROAS reliably from around 150 conversions a month. Below that, results vary a lot.

Conversions per month (all product campaigns)Our recommendation
under 50One campaign with all products. No split, not even by brand.
50 to 150Two campaigns: top sellers and the rest. Only if the rest gets at least 30 conversions on its own, otherwise stay with one.
over 150Three campaigns: top sellers, challengers, slow sellers.
well over 150 per campaignConsider further splits, for example brand, new customers or countries.

The table is our rule of thumb, not a Google requirement. For very expensive products with few purchases it shifts down, for fashion with many variants it shifts up.

Top sellers, challengers, slow sellers: how do we segment products?

We don't split products by category or collection but by what the data says about them. The split is done by our own Google Ads script, which runs daily. It changes nothing in the account. It writes a label for each product into a Google Sheet, which Google Merchant Center or Channable pick up as a custom label. Our campaigns then filter products by these labels.

First question: is there enough data for a verdict?

Most labelizers rate every product against fixed click or ROAS thresholds. Producthero, for example, uses a 30-day lookback by default (Producthero). Products without data quickly end up in the worst bucket. Our most important principle is a different one: missing data is not proof of poor performance.

That is why the threshold depends on the price. The script calculates a target CPA for each product as price divided by target ROAS. A verdict is only made after two conversions or once spend reaches 2.5 times the target CPA. An example:

ValueProduct AProduct B
Price€40€600
Target ROAS55
Target CPA€8€120
Verdict from (without 2 conversions)€20 spend€300 spend

An expensive product automatically gets more room before it is downgraded. With a fixed click limit, it would be sorted out after a few days.

Second question: what does the data say?

LabelRuleCampaign
HeroROAS at least 20% above target and at least 3 conversionsTop sellers
SolidROAS at or above target, enough dataTop sellers
WatchNot enough data for a verdict yetChallengers
No trafficfewer than 50 impressions in the periodChallengers
Newadded since the last run, no label yetChallengers
Weakenough data, ROAS below targetSlow sellers

A product with ROAS above target is never classed as weak. Moving up into the top seller campaign takes real conversions, spend alone is not enough. We learned this in a high-ticket client account: there, the thinly backed half of the top seller cluster had a ROAS of 1.02, the well-backed half 2.72.

Keeping the structure stable

  • The script evaluates 90 days, shifted by the conversion lag. In our accounts the lag ranges from 3 to 21 days. If you don't measure it, you downgrade products on immature data.
  • Downgrades only take effect after 21 days in the same state, upgrades immediately. Every move costs the bid strategy learning time.
  • If the price changes by more than 10 percent, the product is re-evaluated. After 180 days in the slow seller bucket it gets a fresh chance.
  • We group variants via item_group_id as long as the price spread within a group is no more than 1.5x. In fashion, individual sizes never get enough data otherwise.

Two or three campaigns?

In data-rich, stable markets two campaigns are enough: top sellers and everything else. Three campaigns pay off when the range fluctuates, for example through many new products or frequent changes in availability. The challenger campaign then catches every product that cannot be judged yet, including those back after a stock gap. Products that are temporarily out of stock keep their label. If they are back in stock within the lookback window, they go straight back to the right campaign.

All campaigns share the same target ROAS. We only set different targets on purpose, for example for a slow seller campaign with a clearly limited budget. Since August 2026 we review that limit account by account, because budget-limited target ROAS campaigns now buy clicks at higher prices (more on that below).

A note on our own advice: in our earlier fashion guide from 2024 we advised against splitting best sellers from poor sellers, because potential was spotted too late. The real problem was rigid rules that penalised products without data. With price-based thresholds and the challenger logic, this segmentation is now our standard.

Brand, margin, new customers: which other splits pay off?

SplitWhen it pays offCatch
Brand and genericWith meaningful brand search volume. Optmyzr found exact overlaps between search keywords and PMax search terms in 91.45% of 503 accounts.Without brand exclusions, PMax flatters its numbers with cheap brand searches. A full exclusion costs visibility in Shopping, though.
MarginWith very large margin differences, for example own brand versus third-party goods.Rarely works well in our accounts. Mixed baskets blur the effect and the data gets thin.
New and existing customersWith a high repeat purchase rate and more than 150 conversions a month.With long purchase cycles, say every eight months, attribution is imprecise. Customer lists must be up to date.
CountryWith separate feeds, currencies or budgets per country.Small countries often lack the volume for their own segmentation.
Season or price bandHardly ever as a separate campaign.Seasonal campaigns restart the learning phase every year.

We usually handle the brand split with a dedicated brand search campaign on exact match plus a brand Shopping campaign in which a script adds all generic search terms as negatives. The Optmyzr overlap figure dates from February 2025 (as cited by Lionelz). It shows that both campaign types bid in the same search space, but on its own it doesn't prove lost revenue.

Instead of splitting by margin, we prefer one shared target ROAS derived from contribution margin 3 (CM3) and rate products by conversion value. The section on steering by contribution margin explains how.

How does PMax interact with Standard Shopping, Search and AI Max?

Since October 2024, PMax no longer takes automatic priority over Standard Shopping campaigns. If both advertise the same product, the higher Ad Rank wins (Producthero). Whether the campaigns bid each other up is disputed. Google says no (Ginny Marvin, Google Ads Liaison), while smec and Producthero see self-competition in practice.

Our recommendation: PMax and Standard Shopping may advertise the same products. Google's account managers now also recommend running both campaign types in parallel with the same products. So we deliberately add a Standard Shopping campaign to a very successful PMax campaign. It often brings extra volume, and one plus one makes three. That doesn't always happen. After a few weeks we check whether total revenue and contribution margin really rise, or whether the campaigns just take clicks from each other.

For search queries: a keyword in a Search campaign that matches the query exactly takes priority over PMax (Google Ads Help). For all other matches, meaning broad match, Dynamic Search Ads and AI Max for Search, Ad Rank decides. Google has postponed the automatic migration of Dynamic Search Ads to AI Max until February 2027 (Google Ads Developer Blog).

AI Max for Shopping has been available since July 2026. According to Google Ads Help, automatic product title customisation currently only works with English-language feeds (Google Ads Help). For German and other non-English feeds, this part brings nothing yet.

Target ROAS, budget and learning phase

Start with or without a target ROAS?

For shops with historical data, we start straight away with a target ROAS. Google recommends the average of the last 28 to 30 days as a starting value, leaving out the most recent days because of conversion lag (Google Ads Help). We only start without a target in new markets or new accounts without data, and set the target ROAS once there are enough conversions.

Larger changes trigger a learning phase of 7 to 14 days. We therefore adjust target ROAS and budget in steps of 10 to 20 percent, at most every two weeks. We make a strategic assessment after six weeks at the earliest.

Why you shouldn't cap your budget any more

Until August 2026, a tight daily budget on target ROAS campaigns had a side effect: Google throttled bids to stay within budget. These campaigns ran a median of around 20 percent above their ROAS target. On 17 August 2026, Google ended this throttling. smec measured the consequences (smec, September 2026):

Metric (median)Budget-limited target ROAS campaignsUnlimited campaigns
CPC change+15.8%−13%
Average CPCfrom €0.38 to €0.44down to €0.33
Impression sharefrom 40% to 31%not reported
ROASon target within three weeks instead of around 20% abovenot reported

If you used a tight budget as a hidden ROAS brake, you now pay more per click. Our recommendation: if the campaign hits its target ROAS, remove the cap and steer through target ROAS alone. If the client has a hard monthly cap, we would rather raise the target ROAS than limit the budget. If a campaign misses its target, budget is rarely the lever. Check target ROAS, feed, signals and conversion data first.

How do I steer PMax on contribution margin instead of revenue (POAS)?

ROAS (return on ad spend) knows nothing about returns, cost of goods or shipping. So we turn the logic around: first comes contribution margin 3 per order, which gives the break-even ROAS, which in turn gives the target ROAS. If €38.50 out of €100 revenue is left after cost of goods, returns, fulfilment and payment, the break-even ROAS is around 2.6. Any target ROAS below that buys orders at a loss.

POAS, profit on ad spend, goes one step further. Instead of revenue, you pass the contribution margin of each order to Google as the conversion value, and the bid strategy optimises for profit rather than revenue. Strictly speaking, this is usually not final profit but CM1 or CM2, meaning revenue after cost of goods and, depending on the setup, after shipping, returns and payment. The term has caught on anyway. A POAS of 100 percent means the margin you pass on covers exactly the ad spend.

The options at a glance, sorted by precision:

MethodWhat it doesLimit
Target ROAS from CM3One target for all campaigns that fits the marginWorks on average, not per product
Margin clusters via custom labelsDifferent targets per product groupFails with mixed baskets, splits the data
Conversions with cart data plus cost_of_goods_sold in the feedGross profit per order in Google Ads reportingMainly for reporting, shipping and returns are missing
Profit optimisation (beta)Bids on gross profitLimited access only, without shipping and returns
POAS: CM1 or CM2 as conversion value (server-side)Bids on contribution margin instead of revenue, break-even at 100% POASNeeds clean server-side tracking and cost data per order

For most shops the first step matters most: a target ROAS that comes from CM3, not from what the account happens to deliver. Our segmentation script rates every product against exactly this target. Moving to POAS then pays off for shops with stable tracking and big differences in margin or return rate.

What can you control in PMax today? (As of November 2026)

The pure black box days are over. These features matter for online shops:

FeatureWhat it doesNote
Negative keywordsUp to 10,000 per campaign, also via lists (Google Ads Help)Only apply to Search and Shopping inventory, not to YouTube, Display or Discover
Brand exclusionsBrand lists per campaign, including misspellings and variantsCheck the effect on brand Shopping before using them
Search terms reportSearch terms in the same depth as in Search campaigns, also via APIAvailable for all PMax campaigns since August 2025
Channel performance reportCost, conversions and value by channel and format (Google)Rolled out globally in November 2025, data from June 2025
Audience exclusionsExclude Customer Match and remarketing lists at campaign level (Google)Since 2026, check the rollout in your account
LanguageSince late September 2026 no longer applied on the Search Network, only on YouTube, Display, Discover and Gmail (Google)Keep ads and landing pages clearly in the target language
Image extractionOn by default for new campaigns from 4 December 2026 (Google)Opt-out possible per campaign, existing campaigns stay as they are

We use search terms for more than exclusions: good terms go into Search campaigns, others feed ideas for SEO and the product range. We check the channel report monthly. If a lot of budget moves into Display or YouTube without conversions, something is usually wrong with the feed or the signals.

Asset groups and audience signals for shops

Signals are hints for the algorithm, not hard targeting. Which signals work best is disputed. The common recommendation names first-party data such as customer lists and website visitors. In our tests, interests were usually the strongest signals. That's why we test this in every account instead of following the recommendation blindly.

With a Shopping focus, varying the asset messaging rarely makes a difference. Our standard: many identical assets, spread across asset groups with different audience signals. Image and video formats don't need to be complete, because Google builds many formats dynamically from the feed in e-commerce. Under Assets and Performance you can see which assets Google has added with AI. From 4 December 2026, image extraction from your website is on by default for new campaigns. If you don't want that, switch it off per campaign.

After setup, asset groups should run for at least six weeks without major changes. Only then can you see which group carries the results and which just takes budget.

What is different in Germany and the EU?

  • CSS partner: In our experience, a comparison shopping service (CSS) pays off from around €300 Shopping budget a month. It saves 10 to 20 percent on click costs, with low fixed costs. We recommend Channable as a CSS partner. PMax uses the setting from Merchant Center. For background, see our guide to organic Shopping in the EU and UK.
  • Consent Mode v2: Without correctly passing ad_user_data and ad_personalization, Google lacks data for bidding and audiences. Server-side tracking and Enhanced Conversions close part of the gap.
  • Not available yet: Ads in AI Mode are not running in Germany so far. Generating assets from a URL is limited to the US according to Google Ads Help. Title customisation in AI Max for Shopping only works with English feeds. Profit optimisation is a closed beta.

Many English-language guides describe features that are not live in the EU yet. Check in your own account what is really available before you build your structure around it. And ask your Google account manager: they can unlock many new features and betas for your account before general availability.

Common mistakes that cost PMax budget

Final URL expansion runs without URL exclusions

Google may then also advertise pages that are not in the feed, such as blog or guide pages. They rarely convert. Exclude them with URL rules.

The optimisation score is treated as a benchmark

We have found no link between the optimisation score and campaign success. Many recommendations raise spend, not contribution margin.

Too many campaigns

Five campaigns with 20 conversions each perform worse than one with 100. Merge anything below 30 conversions a month.

Soft conversions in the conversion goals

Page views or newsletter sign-ups as primary conversion goals make the campaign optimise for early steps. For bidding and segmentation, only primary conversions count. We never use All conv. value, it is misleading in almost every account.

The budget is capped

Since August 2026 this costs measurably higher click prices on target ROAS campaigns. See the section on budget and learning phase.

The feed is built once and never touched again

With 90 percent of spend going into feed-based ads, this is the most expensive mistake. Titles, product types and images need regular care. If you don't know where your feed stands, a product feed audit is the quickest way to find out.

Frequently asked questions about Performance Max in e-commerce

What is Performance Max?

A Google Ads campaign type that covers all Google channels with one campaign. Google's AI handles bids, placements and ad combinations, you provide the feed, conversion goals, assets and signals.

Where does PMax show ads?

On Search, Shopping, YouTube, Display, Discover, Gmail and Maps. In e-commerce, most of the budget goes into feed-based Shopping ads. The channel performance report shows exactly which channels.

Do you still need Standard Shopping alongside Performance Max?

Not necessarily, but it often pays off. Google's account managers also recommend running PMax and Standard Shopping in parallel with the same products. We add a Standard Shopping campaign to a very successful PMax campaign and check whether total revenue and contribution margin really rise. It often works, but not always.

How many conversions does a PMax campaign need?

At least 30 a month, better 50 to 80. In our accounts, Google only hits the target ROAS reliably from around 150 conversions a month.

How many asset groups make sense?

As many as you have different audience signals to test. Technically up to 100 per campaign are possible, but for shops a few groups with identical assets and different signals are usually enough.

Does PMax work without your own videos?

Yes. In e-commerce, Google builds video and image formats dynamically from the feed. A clean feed matters more than your own videos.

How long is the learning phase?

7 to 14 days after major changes. Assess the structure after six weeks at the earliest.

Where can I find the search terms of a PMax campaign?

In the campaign's search terms report, available since August 2025 in the same depth as for Search campaigns and also via the API.

Can I exclude keywords in PMax?

Yes, up to 10,000 negative keywords per campaign. They only apply to Search and Shopping inventory.

Conclusion

A good PMax structure for shops is lean rather than granular: few campaigns with enough conversions, products sorted by data rather than category, and a target ROAS that comes from your contribution margin.

Your next step: pull the conversions per PMax campaign for the last 90 days. If a campaign is below 30 a month, merge it with another one. Then check whether budgets are capped even though the target ROAS is being met. If you'd like us to look at your account with you, take a look at our Google Ads audit.

Quellen

Latest Articles

Knowledge Base

Interviews, tips, guides, news and best practices on the topics of search engine optimisation, SEA and digital strategy.
Strategy
12 minutes

Agentic Commerce: Who It Affects – and Who It Doesn't

Agentic commerce hits purchase patterns, not industries: AI agents take over rational, recurring purchases while emotional ones stay human. The matrix shows where your range stands and which product data matters now.
To the article
Strategy
10 minutes

Contribution Margin 3 in E-Commerce

How to calculate contribution margin 3 per order, why new and existing customers need separate numbers, and how marginal CM3 steers your ad budget.
To the article
7 budget killers in SEO projects and how to avoid them
SEO
8 minutes

7 Budget Killers in SEO Projects – And How to Avoid Them

From audits no one implements to tool overkill and perfectionism in the wrong places: The 7 most common budget killers in SEO projects – and how to protect your budget.
To the article

Get in touch

We'll get back to you with an answer within 24 hours — we promise.

By entering your data, you agree to the Privacy Policy I agree with Pan Pan.

Great, we have received your message!

We are looking forward to working with you to achieve your success!
Oops, something went wrong!

Please try again or contact us directly via email.