Conversion lift vs. brand lift study: what each measures

A brand lift study splits your audience into people who can see your ads and a control group who can't. Then it shows a short survey to people in both groups, asking things like:

• Do you remember seeing an ad for this brand recently? (ad recall)
• Have you heard of this brand? (awareness)
• Would you consider buying from this brand? (consideration)

The difference in answers between the two groups is the lift. Google's Brand Lift, for example, shows its surveys on YouTube and reports the difference in positive answers between people who saw the ads and people who didn't. Meta and other platforms offer similar studies.

Brand lift is useful for ads meant to build awareness, like video or launch campaigns, where sales may come weeks later. Its limit is that it measures what people say, not what they do. Remembering an ad doesn't mean buying.

A conversion lift study uses the same split, but instead of asking questions it counts purchases or other actions in each group. The difference is the number of sales the ads caused. This is incrementality, measured directly.

On ad platforms, the platform randomly picks a holdout group of users who won't see your ads, then compares purchases using its own tracking, such as its pixel or data you send from your server. Researchers used studies like these at Facebook to show that non-experimental methods often get the effect of ads wrong (Gordon and others, 2019).

A platform's conversion lift study can only count conversions the platform receives. For a brand selling on its own site with good tracking, that's most sales. For a brand that sells on Amazon, it's a big gap: Amazon doesn't send its purchases to Meta, so someone who sees a Meta ad and then buys on Amazon is invisible to a Meta conversion lift study.

That doesn't make the study wrong. It means the study measures part of the effect. If a lot of your ads' effect shows up on Amazon, the study will understate it. The halo effect calculator estimates how much might be missing.

The way around this is to run the test by place instead of by person. A geo test pauses ads in some regions and compares sales from any source that records where orders ship, including Amazon.

• Look at the range, not just the number. Both kinds of study rely on samples: survey answers or purchases. A lift with a wide range that includes zero means the study couldn't tell.
• Brand lift is a leading signal. More people remembering your ad is good news, but it isn't sales. Check it against a sales measure before moving budget.
• Conversion lift counts what was tracked. If tracking misses some sales, the study understates the ads' effect.

• Launching a product or a brand campaign? A brand lift study shows whether people are noticing it, long before sales are measurable.
• Deciding whether ads pay for themselves? You need sales, so run a conversion lift study or a geo test. Then compare the incremental ROAS with your break-even ROAS.
• Selling mostly on Amazon or in stores? A platform conversion lift study will miss those sales. A geo test can count them.

Spillmark runs geo holdout tests that measure the Amazon sales Meta ads cause, the part a Meta conversion lift study can't see. It counts orders, not survey answers, and shows every result with its uncertainty range. The method is on how lift is calculated.