Incrementality testing is an experiment that measures how many sales or other conversions a marketing channel actually caused. The audience is split at random into a test group that can see the ads and a holdout group that cannot, and the difference in conversions between the two is the lift: the sales that would not have happened without the ads. For an ecommerce brand, it is the most direct check on whether a channel's reported return holds up, because platform attribution counts conversions the ads touched, not conversions they caused. This answer is from Brand Media Works, an AI consulting and marketing data engineering firm for ecommerce brands.
What does incrementality mean in marketing?
Incrementality is the part of a result that happened because of the marketing. Meta describes a lift test as an experiment that compares "groups of people who did and did not have the opportunity to see your Meta advertising to understand its causal impact". Google describes Conversion Lift the same way: the audience is split into a treatment group that sees the ads and a control group that does not, and the difference in conversions is the lift.
Two details make it a real experiment. The groups are assigned at random, and people stay in their group for the whole test. Meta notes that a lift test is not the same as an A/B test, because an A/B test compares two versions of an ad and has no randomized holdout group that is kept from seeing the advertising at all.
What does a brand need in place before it can run one?
Three things: enough conversions to clear the platform's minimums, clean data to read the result against, and a way to run the analysis.
- Platform minimums. Google Ads requires at least 1,000 observed conversions and a minimum campaign budget of $5,000 for a Conversion Lift study, and says the tool is not available for all accounts (read 29 October 2026). Meta starts showing holdout results once a test has observed at least 100 conversion events.
- Data by day and by region. A geography-based test compares regions, so the brand needs orders, revenue and ad spend for each region and each day, from the store and every ad platform, under one definition of a conversion. Without that, a test can only be read inside the ad platform that is being tested.
- A way to analyze it. Meta's GeoLift is a free, MIT-licensed R package that selects test markets, runs power calculations and measures the lift. Dedicated vendors run the same kind of test as a managed service.
The data foundation is often cheaper than the software. Google's BigQuery Data Transfer Service loads Google Ads data into BigQuery with no transfer orchestration charge, and bills the Facebook Ads connector by compute used, with a planning upper bound of about $1.20 per hour of transfer run time in its Iowa region. BigQuery's first 1 TiB of queries each month is free, then $6.25 per TiB, and the first 10 GiB of storage is free each month (read 18 October 2026). Daily orders and spend by region are small tables, so for many brands the query and storage cost stays inside those free tiers.
Vendor prices are harder to compare. On 29 October 2026, Haus, Lifesight and Prescient AI published plan features but no prices. Prescient says its platform is built for brands spending $100,000 or more a month on advertising, on annual contracts. Northbeam lists its Professional attribution plan at $3,500 a month, with incrementality as an optional add-on at an unpublished price.
A custom setup tends to beat a vendor when a brand tests one or two channels a quarter, already keeps its sales and ad data in a warehouse, and has an analyst who can run GeoLift. A vendor tends to win when a brand wants to test many channels at once, including TV, podcasts or influencers, and has no one in-house to design and read the tests. Marketing data engineering describes how Brand Media Works builds the warehouse a test reads from, and data engineering consulting cost sets out what consultancies publish for that work.
What is the formula for calculating incrementality?
The core formula is the difference between the two groups, once they are the same size:
- Incremental conversions = test group conversions minus control group conversions (scaled to the test group's size).
- Relative lift = incremental conversions divided by control group conversions.
- Incremental cost per action (iCPA) = total spend divided by incremental conversions, as Google defines it.
- Incremental return on ad spend (iROAS) = incremental conversion value divided by total spend.
Meta gives a worked example: if a test observes 5,000 events in the test group and 2,000 in the control group, the conversion lift is 3,000 events. That is a relative lift of 150% over the control group, and 60% of the test group's conversions were caused by the ads (our arithmetic). The scaling matters because the groups are rarely the same size: in Meta's single-cell holdout tests, the control group is 10% of the size of the test group, and Meta scales its results up before comparing.
How do you measure incremental sales?
There are three common ways, and they differ in what they need from the brand:
- Platform conversion lift. Meta and Google split users into test and holdout groups inside their own platforms. It is the quickest to set up, but each platform measures only its own ads, and Google allows a campaign in only one lift study at a time.
- Geography-based tests. Ads run in some regions and are held back in others, and sales in the two sets of regions are compared. Google offers geography-based Conversion Lift, which supports offline data, and GeoLift builds a synthetic control from the regions that did not see the ads. This method works across channels, but it needs sales data by region.
- Before-and-after tests. Spend is switched off or raised for a period, and sales are compared with a forecast. These are the weakest of the three, because seasonality and promotions move sales at the same time.
Vendors publish their own examples. Lifesight reports that a US eyewear brand's geo holdout on its largest social channel estimated an incremental ROAS of 3.48, a single brand's result under its own test conditions.
How long should an incrementality test run?
Google allows Conversion Lift studies as short as 7 days but typically recommends more than 14, and at least 14 when customers take time to buy (read 29 October 2026). Meta recommends waiting until a test has completed and its results are final before acting on them, even though results start to show earlier. On reading the result, Meta treats a confidence percentage above 90% as reliable evidence that its ads caused the lift for that event.
For an ecommerce brand, the practical rule is to run the test across at least one full purchase cycle, avoid launching it during a major sale, and decide in advance which number will change the budget.
Frequently asked questions
What is incremental testing?
In marketing, it is another name for incrementality testing: an experiment with a randomized holdout group that measures the conversions a campaign caused. In software engineering the same phrase means something different, testing a program by adding and checking components one at a time.
What does incremental impact mean in marketing?
The change in sales, sign-ups or other results that would not have happened without the marketing. Meta calls it conversion lift and Google reports it as incremental conversions, incremental conversion value and incremental ROAS.
Is a lift test the same as an A/B test?
No. An A/B test compares two versions of an ad or campaign. A lift test keeps a randomized holdout group from seeing the advertising at all, which is what lets it measure the effect of the advertising itself.