How to improve your MMMs with better inputs

New research from The Trade Desk surfaced data-driven marketing mix modeling (MMM) insights that can inform more effective programmatic ad spend. 

MMMs are a go-to tool for measuring media effectiveness. But their outputs are only as good as the inputs that marketing analysts and their modeling vendors feed them. Too often, models rely on incomplete data or oversimplify channel differences, which can skew results and lead to suboptimal media planning. 

To dig deeper into this challenge and find solutions, we conducted research across multiple advertisers and categories over 16 months, focusing on purchase conversions and three core channels: connected TV (CTV), online video (OLV), and display. The goal was to understand how specific inputs influence model accuracy and identify adjustments that can make MMMs more reliable for programmatic campaign planning.

The findings revealed two key insights that advertisers and their measurement partners can use to gain more accurate and actionable discoveries from their MMMs. 

Remember that every channel performs differently 

Many MMM vendors aren’t necessarily media experts, which means their models can fail to account for meaningful differences across channels. A model that only considers spend might treat CTV, display, and OLV as interchangeable, when in reality, each channel performs differently. 

Insight #1: Embed business and media expertise. Supplement raw spend with media quality metrics that represent the true value of media placements, including: 

• CPM: a normalized proxy for production value 

• Engagement and attention metrics: viewability (display), completion rates (video/​CTV), TVQI (CTV)

Adding these signals helps models cut through impressions that never had a chance to influence outcomes and, instead, focus on the actual media exposures most likely to drive business results. In other words, when you model differently, you may uncover insights that wouldn’t have surfaced otherwise. 

Don’t let testing history limit your future 

MMMs are often built with only historical data, including past budgets and formats. But limiting inputs this way can trap a model in a testing loop, constraining its ability to reveal new insights. 

Insight #2: Give models larger ranges of priors. Advertisers should encourage their partners to test broader ranges for key parameters like half-life and frequency. Doing so allows models to surface outcomes beyond what history alone would suggest. 

Data from the provided ranges can serve as starting points for better testing:

It’s important to note that these are modeling ranges, not campaign guidance. Optimal frequency isn’t fixed, but instead depends on factors like creative quality, audience, and campaign type. 

Strong creative may need fewer exposures; broad-reach campaigns may require more. 

To keep models fresh, advertisers should also consider periodically refreshing priors with findings from new measurement. Conversion lift studies are especially valuable here, as they directly link to business outcomes. 

Turn your MMM into a planning tool 

When the MMM model is configured to include media domain knowledge through informed priors, it becomes more actionable. The model can be used to run “what if” simulations by adjusting spend or frequency to guide planning. That means you can explore how shifting budgets or testing different ranges might impact conversions, which can help you craft campaigns more confidently. 

MMMs also give marketers a structured way to test emerging channels. Even if you’re newer to a channel like audio and don’t yet have deep historical data, including it in the model can reveal its potential contribution to the overall media mix. This turns MMMs into both a forward-looking tool and a retrospective tool. 

What advertisers and modelers should remember 

• Embed quality signals. Don’t let models treat channels as interchangeable; include CPM and engagement metrics to differentiate them. 

• Expand your ranges. Broader priors on frequency and half-life open the door to new insights. 

• Refresh with new measurement. Lift studies tied to conversions provide the most meaningful inputs for priors. 

Together, these steps make MMMs more accurate, more actionable, and more valuable for planning programmatic investment. 

Research methodology:

The Trade Desk’s MMM study was conducted by evaluating seven brands across the following categories: pharmaceutical, automotive, quick service restaurants, and tech/​telco. The modeled key performance indicator (KPI) was purchase conversions across an analysis window of January 2024 through May 2025. Explanatory media variables used in the study included: connected TV, online video, and display. Explanatory non-media variables included: intercept, seasonality, and trend.