How to Test Whether Performance Max Brings New Customers

Incrementality testing Performance Max for net-new customers comes down to three tools that answer three different questions, and our position is that the one teams reach for first is the weakest. The new customer acquisition goal tells Smart Bidding whom to chase, then reports what it caught. An uplift experiment measures extra conversions for the whole account. A geo holdout, read against your own order table, is the only one of the three that can show first-time buyers who would have stayed away without the campaign.

Google's own documentation supports the point about the goal. Its measurement page on Google Ads Help says: "Incrementality testing of lifecycle goal settings isn't supported within Google Ads, however, you can use the Conversion Lift based on geography and Conversion Lift based on Users features to test if your media mix is delivering incremental new customers to your business." Our explainer on what incrementality means covers why attributed and incremental results drift apart.

What each method can and cannot prove

As of October 2026 we found no dedicated new customer acquisition experiment in Google Ads. The Performance Max incrementality experiments Google documents are the uplift experiment, Conversion Lift studies and holdout geo experiments.

Does the new customer acquisition goal measure incrementality?

The Google Ads PMax "new customer acquisition" goal reports new customers without measuring incrementality, and the gap starts with how "new" gets decided. On that measurement page, a buyer counts as new when they haven't purchased within the last 540 days under Google autodetection, when they're missing from the existing customer list you supplied, or when the new customer reporting tag flags them. Autodetection is the default, and Google describes it this way: "This is the easiest method to distinguish new customers from existing customers but is less comprehensive because users can remove cookies and opt out."

Reporting widens the gap further: when conversions are counted as "every", repeat purchases a new customer makes inside the conversion window are also classified as new in the conversion counts. The "New Customers" column is the exception, showing a distinct count of new customers irrespective of the repeat purchase rate. Google's suggested check is a pre and post comparison of the new customer ratio in your own analytics, which shows whether the customer mix shifted and leaves open whether the campaign caused any of those purchases.

Which mode should run while you test?

Google Ads new customer acquisition goals give Performance Max two main modes, and they treat existing customers and value-based bidding differently. The settings table on Google Ads Help lists the requirements:

• New Customer Value bids higher for new customers than existing ones, so returning customers still receive bids. It requires value-based bidding (Target ROAS or Maximize conversion value) and at least one Purchase conversion goal.
• New Customer Only bids only for new customers. It accepts value-based bidding plus Maximize conversions and Target CPA, and a Purchase conversion goal is recommended without being required.

A third setting, High Value New Customer, covers Performance Max and Search and is labeled a beta on Google's detection page. Google's advice is plain: "We recommend New Customer Value for all advertisers with purchase conversion goals." We'd also keep it on through a test, since switching modes midway changes who can see the ads. One caution on the numbers: lifecycle goals adjust conversion value for new customers, so judge ROAS on the "Original conv. value" column, which the measurement page says displays the unadjusted value.

What a Performance Max uplift experiment proves

Google Ads Help defines the uplift experiment in one sentence: "You can use Performance Uplift experiments to measure the uplift or incremental benefit of using Performance Max with your existing Search, Video, and Display campaigns." The traffic split is 50% by default and is applied at the eligible auction level using cookie-based or user-based splits, according to the experiments FAQ on Google Ads Help. Google recommends at least 4-6 weeks, longer when conversion delay is long, and discards the first 7 days as ramp-up. On budget we found no minimum in the FAQ, which says that the higher Performance Max spend is relative to total account spend, the better the chance of statistically significant results.

Two limits matter for a new-customer question. Retailers on a Merchant Center feed are excluded, since the FAQ states: "Uplift experiments don't support Performance Max with GMC feed." Results also arrive as aggregated conversions, conversion value, CPA, ROAS and spend for each group, and we found no new-versus-returning breakdown in the results Google documents. Because the split happens by cookie or user inside the auction, there is no region or customer list to match against your own orders.

How do you set up a geo holdout for Performance Max?

A Google Performance Max geo experiment uses a holdout setup to measure incrementality: ads keep serving in control regions, go dark in test regions, and the gap in orders is the lift. For a campaign that is already live, the go-dark steps in the geo experiment guide on Google Ads Help are:

1. Remove country-level targeting and target only the control regions (DMA regions in the US, city IDs or ZIP codes elsewhere). Multi-cell designs add explicit exclusions for the dark regions.
2. Set Location options to Presence to avoid leakage.
3. Keep the campaign out of shared budgets. If its budget is capped, cut the daily budget to the control regions' historical share of spend; Google warns that the system otherwise redistributes unspent funds to the control regions and inflates the baseline.
4. Hold pre-test data covering at least 3 times the test period, and drop any learning period from the analysis.

Google runs its own version, Conversion Lift based on geography, which supports Performance Max. Google Ads Help calls it a beta limited to holdback and go-dark studies and says Conversion Lift isn't available for all accounts, so access goes through a Google account representative. User-based Conversion Lift supports Performance Max too, with a minimum campaign budget of $5,000 USD and at least 1,000 observed conversions, per Google Ads Help.

Reading the result against your own order data

Count first-time orders by region from your order system, using your own definition of new. Google's autodetection treats anyone without a purchase in the last 540 days as new, so unless a list you uploaded says otherwise, a customer coming back after two years is new to Google and returning to you.

A hypothetical example, with every figure assumed for illustration:

The holdout regions ran at one third of control before the test, so 1,860 control orders imply 620 in the holdout, and they delivered 540. The 80 missing first-time orders, about 13% of the expected 620, are the campaign's incremental new customers in those regions. If the budget withheld from those regions over the six weeks was $21,000, the incremental cost per new customer is $262.50. Suppose the platform-reported cost per new customer before the test was $70: the same $21,000 would have been credited with 300 new customers, so roughly 27% of them were incremental. The $262.50 belongs in your CAC model, and our ROAS calculator is a further resource for the return side of the math.

Before trusting the 80, check how far the holdout-to-control ratio drifted across rolling six-week windows in the 18 pre-test weeks. If it moved by 13% or more with no test running, the shortfall sits inside normal variation and the test needs more regions or more time.

Practices that hold for every version of the test

Google Ads Performance Max incrementality testing best practices mostly protect the comparison from your own edits:

• Run one test at a time and freeze budgets and targets. Google's FAQ says experiments run at once can interfere with each other and puts the learning phase at 7-14 days; the geo guide adds that any budget change greater than 20% starts a new learning period.
• Give Google your customer list anyway. Google uses the new versus existing customer parameter in the conversion tag in place of autodetection when you send it, and notes that a customer list with fewer than 1,000 members may affect the goal.
• Fix the pass mark before launch: the incremental cost per new customer you would accept, written down.

Our recommendation is to treat the three as a sequence. Keep New Customer Value on as a steering input, use the uplift experiment when the question is whether Performance Max adds anything on top of Search, and reserve the phrase "new customers" for what a holdout and your order table agree on. Once the incremental number is known, decisions about scaling paid ads without killing ROAS get easier to defend. If you want help designing the test, talk to our paid media team.

Sources

• Google Ads Help: Measure your lifecycle goals campaigns
• Google Ads Help: About customer lifecycle goals
• Google Ads Help: How customer lifecycle goals detect customer segments
• Google Ads Help: Set up Uplift experiments for Performance Max
• Google Ads Help: Experiments FAQs
• Google Ads Help: Implement campaigns for geo experiments
• Google Ads Help: Set up Conversion Lift based on geography
• Google Ads Help: Set up Conversion Lift based on users