How to measure Demand Gen creative impact with asset uplift tests

Demand Gen campaigns have high visibility across YouTube, Discover, and Gmail. However, they pose a key challenge: the “attribution illusion.” You’ll often question whether reported conversions in the platform are truly incremental or if these users would’ve converted through search either way.

That’s why in November, Google launched asset uplift experiments, giving you the ability to measure the impact of Demand Gen creative through an A/B split test. This means you can replace assumptions with a clearer view of what’s actually driving incremental results.

Relying too heavily on creative instinct or default reporting can lead you down an inefficient path and divert valuable creative resources toward poor-performing assets. Using Google’s A/B testing capabilities helps you isolate the impact of individual assets and avoid that outcome.

Why attribution doesn’t equal incrementality

If a user views a Demand Gen ad on YouTube and doesn’t click but then searches for the brand and converts, Google may attribute partial or full credit to the Demand Gen campaign and creative. This attribution more so reflects correlation rather than causation.

Accurate measurement and the scientific method show the need to understand the scenario in which the creative isn’t shown. By withholding the test assets from a segment of the target audience, it’s possible to establish a baseline. 

The difference in conversion rates or any primary KPI between the treatment group — those who were exposed to the ad — and the control group — those who weren’t exposed — shows the actual incremental lift the creative is driving.

Dig deeper: Why incrementality is the only metric that proves marketing’s real impact

What you need before testing creative uplift

One common mistake is launching experiments without enough data to reach statistical significance. To avoid inconclusive or invalid results, make sure your campaign meets these prerequisites before setting up the test.

Conversion volume 

Google recommends having at least 50 conversions across treatment and control arms during the experiment to measure lift accurately. If your primary conversion doesn’t receive this volume, consider optimizing the test around high-intent micro-conversion actions, such as “Add to Cart.”

Budget minimums

Experiments should run with continuous, uninterrupted spending. If your Demand Gen campaign is limited by budget and stops early each day, the control group data will be skewed. 

The campaign must have a sufficient budget to run for at least four weeks, or until a statistically significant result is achieved.

Creative isolation

Test only one new variable at a time. To determine if a specific video asset drives uplift, keep all other campaign elements, such as audience, bidding, and standard image assets, unchanged.

Dig deeper: Why Demand Gen is the most underrated campaign type in Google Ads

How to run an asset uplift test in Google Ads

Setting up a creative uplift test is now more streamlined within Google Ads. To build a valid experiment, follow these steps.

1. Define a clear hypothesis

Every valid scientific test begins with a clear hypothesis. Avoid running tests without a defined objective. For example:

• Bad hypothesis: “Let’s see if our new video works.”
• Good hypothesis: “Adding user-generated content (UGC) to our Demand Gen asset group will drive a 10% incremental lift in ‘purchase’ conversions compared to standard static image carousels.”

Navigate to the Experiments interface

Log in to your Google Ads account and navigate to the left menu. Select Campaigns > Experiments. Click the plus (+) button to create a new experiment, choose Asset tests provided by you, and make it a Demand Gen campaign experiment.

Configure a 50/50 split

Google will prompt you to define your split. To set up statistically sound results, use a 50/50 cookie-based split. 

This ensures both control and treatment groups have equal historical data and algorithmic weighting, and prevents users from ending up in both arms of the test. Assign your existing campaign as the control, and the duplicated campaign with new assets as the treatment.

Lock your variables

Once the experiment begins, you must practice extreme discipline. Don’t change audiences or targeting, and avoid drastic bid and budget changes. 

Any adjustment made to either campaign during the testing window will introduce noise and could invalidate the statistical significance of your results.

Set the duration

Run the experiment for at least four weeks. 

• Week 1 serves as a learning period while the algorithm adjusts to the audience split, new creative, and bid model learning (especially if leveraging smart bidding). 
• Weeks 2 to 4 provide actionable performance data. 

For longer conversion cycles, such as B2B SaaS, consider extending the test to six or eight weeks.

Dig deeper: What it takes to make demand gen work for B2B and ecommerce

What your experiment results actually mean

When the experiment concludes, review results in the Experiments dashboard, where a report showing the performance of each arm and its confidence interval across metrics is available. Interpret the outcomes as follows to validate your hypothesis made earlier.

Outcome 1: Positive lift (statistically significant)

If the treatment group shows a positive lift with 95% confidence, your creative asset has been proven to drive incremental conversions. 

From there, you can calculate incremental cost per acquisition (iCPA) by dividing the treatment group’s total ad spend by the incremental conversions above the control arm. 

Use this iCPA as your benchmark for scaling the campaign going forward.

Outcome 2: Negative lift

Occasionally, a new creative asset may suppress performance. It may be too disruptive, or the video may have a high skip rate, causing the algorithm to reduce delivery to high-intent users. Pause the treatment asset immediately. This allows you to let data guide your budget decisions vs. preference.

Outcome 3: Inconclusive result

If the difference between groups is negligible and the system cannot confidently attribute conversions to the ad after four weeks and adequate conversion volume, consider extending the test for two more weeks to collect additional data. 

If results are still inconclusive, it could be that creatives are too similar. Test a significantly different creative asset, as small changes rarely produce a statistically significant lift in Demand Gen.

Prove creative impact with incrementality testing

Creative is a key remaining lever and differentiator you can pull to drive performance. Producing high-quality video or UGC is just the first step in this world, where creative bandwidth and impact must be proven as a driver of results. 

Demand Gen is a powerful tool for visual storytelling, but justifying its budget to stakeholders requires rigorous, scientific evidence of its impact. Asset uplift experiments enable just that. Begin your first holdout test, establish a baseline, and let data guide your creative decisions and roadmap.

Dig deeper: The Google Ads Demand Gen playbook

Read more at Read More