Retail media has become one of the fastest-growing parts of the media mix, and as the spend grows, brands need to show whether that investment is paying off. Modern marketing mix modeling (MMM) is well placed to answer that, because it measures every channel in the context of everything else driving the business. However, the quality of the answer depends on modeling each channel in a way that reflects how it influences customers.
When someone opens DoorDash, they’re already deciding what to order for dinner. An ad on the platform reaches them minutes away from a transaction, which is a very different job from a TV spot building awareness over weeks. That difference creates an opportunity to model the channel more precisely and give brands a clearer view of its value.
So we partnered with DoorDash to test it. Ipsos MMA designed the study and built the models across six national and regional quick-service restaurant advertisers. You can check out the full white paper here. Below are the main findings.
The input you choose changes the answer
We tested five ways to represent DoorDash advertising in the model: campaign orders, campaign sales, campaign spend, impressions, and clicks. Campaign Orders performed best across all six advertisers, which follows from how the platform works. DoorDash campaigns are designed to generate orders, so orders are the most direct measure of what the campaigns are actually accomplishing. Impressions and clicks, the metrics most often used for digital media, consistently understated the channel’s contribution.
DoorDash ads work fast
Across advertisers, most of DoorDash’s impact landed in the week the ads ran, with a small tail into the following week or two. Many MMM’s are specified for media like TV, where effects build and linger, and applying those assumptions to DoorDash misattributes how it works.
We also found little sign that returns drop off sharply as spend increases, at least within the ranges these brands were investing, though this varied by brand. Assuming steep drop-offs without evidence would likely also understate what the channel is contributing.
Experiments make the model more reliable
Many models start with some assumptions (statistical priors) about how each channel is likely to perform. We tested three versions: loose assumptions, neutral ones, and assumptions anchored in the results of controlled DoorDash experiments. The experiment-anchored models produced steadier results that held up across repeated runs and matched real-world test outcomes more closely.
Retail media benefits from this approach more than most channels, because spend on platforms like DoorDash tends to rise when demand rises. From historical data alone, it’s hard to tell whether the ads drove the sales or simply rode along with demand, and a well-run experiment gives the model a way to tell the difference.
And of course, the tests should be properly designed, with randomization and real control groups, and the model’s assumptions should be updated as new results come in, at least once a year. When run that way, MMM and experimentation strengthen each other over time.
Why this goes beyond DoorDash
The same principle applies to every channel in a model: each one should be measured with the metric and assumptions that match how it influences customers. MMM is built to accommodate that, and as retail media takes a bigger share of budgets, the payoff from getting these choices right keeps growing.
Building a model this way takes people who understand each channel well enough to question those defaults and have the test results to support a different choice. That’s the standard we hold our own models to, and this study is how we made the case for DoorDash.
We also want to be clear about the limits. This was a DoorDash-funded study covering six QSR brands, and the recommendations are starting points to refine with each advertiser’s own data rather than universal rules. We expect the framework to evolve as more categories and platforms are studied.
Questions worth asking about your own model
If you invest in DoorDash or other retail media, these are good questions to put to whoever builds your MMM:
- Is DoorDash measured on its own, or lumped in with other delivery apps?
- Does the model separate orders your ads drove from orders that would have come in anyway?
- Is DoorDash measured by orders, or by impressions and clicks?
- Does the model reflect how quickly DoorDash ads work, or does it reuse settings built for TV?
- Are your controlled experiments feeding into the model?
Read the full paper
The complete white paper includes the detailed modeling framework, and it’s a useful one to share with your analytics team. Download it here.
If you’d like to pressure-test how your current model handles DoorDash or any other retail media channel, get in touch with our team.
