Tectonic shifts in advertising are forcing marketers to sharpen their strategies. We’re faster and more efficient (thanks, AI). But more importantly, we’re making smarter choices - we’re measuring what works and what doesn’t to understand the best use of our marketing dollars.
Conversion lift studies on Meta are a great way to answer these questions. These studies measure the incremental impact of your campaigns, or the degree to which your ads influence a given outcome.
You’ve probably already asked ChatGPT a thing or two about CLS, or read some articles on Meta Help Center. So, we won’t bother you with similar content in a different packaging. Instead, we sat down with our in-house experts to gather measurement insights from the ground floor. This is a deep dive into lift measurement, built for (and by) performance marketers.
What makes the conversion lift study essential?
A conversion lift study compares the behavior of people who saw your ad (treatment group) with those who didn’t (holdout group). This allows you to isolate the incremental impact of your campaign, or the actions that wouldn’t have happened without your media strategy.
“Lift measurement helps determine the extent to which your media strategy is driving results from people who wouldn’t have acted otherwise.” - Abby Schommer, Director of US Client Partnerships, Hunch
This type of clarity is critical when you’re:
- Making budgeting decisions across channels with different baseline intent; for example, lower-funnel Google Search Campaigns (higher intent) vs. upper-funnel Meta Campaigns (lower intent)
- Testing a structural change, especially when that change is likely to reach an even slightly different audience with different baseline intent, for example, moving from Manual campaigns to Advantage+ Sales campaigns
- Working with long or multifaceted purchase cycles that fall outside typical attribution logic
CLS helps you move beyond assumptions and optimize for what’s actually working.
Until recently, running a Conversion Lift Study was only possible with the help of a Meta rep.
Now, advertisers can do it on their own, but that raises new questions which I’ll address in the following paragraphs.
Who can run a self-serve Conversion Lift Study on Meta?
To make sure CLS is accurate and useful, you need to meet three conditions:
- You need to have a lower-funnel goal, such as purchases or sales
- You have to be technically ready, meaning
- Meta’s Conversions API (CAPI) is fully implemented
- Your Event Match Quality (EMQ) score is above 5
- You’re generating enough signal, with at least 50–100 weekly conversions per test cell
Single-Cell vs. Multi-Cell testing - when and why?
Conversion lift studies can measure the incremental impact of one strategy (via a single-cell CLS) or of multiple strategies at the same time (via a mult-cell CLS).
Single-cell conversion lift studies are often used to get an objective understanding of a strategy's incrementality. These single-cell studies answer questions like: What is the incremental sales value driven by all of my Advantage+ catalog ads?
Multi-cell conversion lift studies are often used to get a relative understanding of a strategy’s incrementality - as compared to at least one other strategy. These multi-cell studies answer questions like: Do my Advantage+ catalog ads with dynamic overlays drive more incremental sales value than my Advantage+ catalog ads without dynamic overlays?
See multi-cell lift testing in action with our Academy Case Study: how Academy Sports tested Hunch-powered creatives vs no overlay creatives
Conversion lift study vs A/B testing - what’s the difference?
Alongside multi-cell CLS, A/B testing is another way to get a relative understanding of a strategy’s performance. However, A/B testing doesn’t measure incremental impact because a holdout is not applied to any of the cells in an A/B test. As a result, one can only rely on the attributed results of each cell of an A/B test to make performance assessments.
While conversion lift measurement is the best way to understand how much ‘Strategy A’ influenced a given outcome vs. ‘Strategy B’ - A/B testing is still a viable option in the event both ‘Strategy A’ and ‘Strategy B’ are reaching very similar audiences with the same level of baseline intent (e.g. you’re testing a minor creative change - new button color, CTA, etc)
Budget constraints and low conversion volume may also make A/B the fallback option, but you should aim to meet Meta’s CLS requirements with time.
How to design a reliable conversion lift test on Meta - Step by step
The results of a conversion lift study are only as good as its design. Intentional planning upfront means your test results will be far more actionable after the study is over.
Use the following steps to make sure your conversion lift results are reliable, every time:
Step 1:
First and foremost, define a single business question that isolates the strategy you’re looking to measure.
- An example question would be: Do Advantage+ catalog ads with dynamic overlays drive more incremental sales value than Advantage+ catalog ads without dynamic overlays?
- In this instance, use of dynamic overlays is what you’re looking to measure
Step 2:
Define your test structure - whether it be a single-cell or multi-cell Conversion Lift Study - based on the business question you've already defined.
- For the above example, you would select a 2-cell design - one cell using Advantage+ catalog ads with dynamic overlays, and one cell using Advantage+ catalog ads without dynamic overlays
Step 3:
When running a multi-cell study, standardize everything across your test cells apart from the strategy you’re looking to measure. That strategy should be the only differentiating factor between cells in your study.
- For the above example, you would ensure that the campaigns in both test cells are brand new and are set to launch on the same exact date (after the CLS begins). You’d also ensure that the catalogs attached to each campaign are also brand new, and that the targeting, optimization, and any other campaign setting is exactly the same across both cells.
Step 4:
Limit contamination to your holdout group(s) by minimizing campaign spend outside of your test cells (especially if those outside campaigns are targeting the same audience as your test cell campaigns).
- For the above example - assuming both test campaigns were targeting the US - you would ideally pause additional US-targeted catalog ads outside the scope of the test for the duration of the study
Step 5:
Once launched, limit changes to the campaigns in your test cells for the entire duration of the study.
How can Hunch help you out with CLS?
Hunch is a creative performance platform, so the most straightforward answer would be - we’ll help you get enough creatives to test. But running a conversion lift test on Meta isn’t just a technical setup & creative workload - it’s a strategic decision.
At Hunch, we act as your thought partner to make sure you’re asking the right question, structuring the test correctly, and interpreting the results with clarity. Our team is trained on Meta’s Marketing Science best practices and has guided many brands through high-impact CLS design and execution.
We’ll work with you to decide if a CLS makes sense for your goals, and if it does, we’ll help you run it smoothly!
FAQ - Answered in Partnership with Meta's Marketing Science Team
#1 How long should I run CLS?
Most CLS tests should run between 2 to 4 weeks. Extending the test won’t necessarily improve statistical significance and could have the opposite effect if additional ‘noise’ (data fluctuation) is introduced during the extended period.
What matters more than test duration is whether your setup generates sufficient conversions and isolates the test variable effectively.
#2 Should I stop all marketing activity outside of my conversion lift test? What are my options if this isn’t possible?
Ideally, you would pause any external campaigns targeting the same audience as your test campaigns, meaning they can contaminate the holdout groups of your test cells. But we know that’s not always possible.
In cases where external campaigns must remain active, you can include those campaigns as a part of your test in what’s known as a ‘sandbox cell’ - essentially a campaign holding cell with a very small (e.g. 2%) holdout. This ensures that users in your test holdout groups aren’t eligible to see any media from your sandboxed campaigns, but without applying a large holdout on the sandboxed campaigns themselves.
Note: To run the ‘sandbox’ method, you’ll need to have your Meta account team configure the test for you. For accounts without a Meta Rep, you cannot run the sandbox method via self-managed CLS, so you should instead assume some level of contamination to every cell’s holdout group if external campaigns remain active.
#3 Do all campaigns in my conversion lift study need to be ‘new’?
For multi-cell conversion lift studies, it is best practice to launch new campaigns across all cells of the test.
For a ‘business as usual (BAU)’ cell, this means you might need to pause existing campaigns and duplicate/relaunch them alongside any other newly launched test cell campaigns.
For catalog-based campaigns, it’s also best practice to have newly launched catalogs across all cells of the test. For the BAU cell, again, this means you might need to pause use of an existing catalog and create a duplicate version of that catalog to be used in the newly launched BAU campaign(alongside any other newly launched test cell catalogs and campaigns).
#4 Is it necessary to have a ‘warm-up’ period for new catalogs included in the test cells?
It is best practice to have newly launched campaigns and catalogs across all cells of a multi-cell conversion lift test. However, if this isn’t possible and there’s a long-standing catalog that must be used in the BAU cell of a test, it’s best to accrue spend against any new test catalogs outside of the testing environment for at least two weeks prior to the test.
This ‘warm-up’ period will allow any new catalogs to generate some history before being included in the testing environment alongside other catalogs with an existing history of learning.
After this period, it’s still advised to pause all active campaigns and launch brand new ones leveraging the existing and ‘warmed up’ catalogs across the cells of your test..
