DMA 7040 Module 2 Cumulative Midterm: Trend Overview and Reach of Marketing Mix Modeling Example

Measuring What Advertising Causes: The Return of Marketing Mix Modeling, Sections 1 and 2

Student Name

MBA in Marketing, Thomas Edison State University

DMA-7040: Digital Marketing Analytics

Mentor Name

Month Day, Year

Measuring What Advertising Causes: The Return of Marketing Mix Modeling, Sections 1 and 2

Section 1: Trend Overview

What the trend is

For most of the past two decades, digital marketers judged channels with attribution models that followed individual users from an ad click or view to a purchase. A sale was credited to the last click, or shared across the clicks before it. Marketing mix modeling (MMM) takes a different route. Instead of tracking people, it uses weekly or daily totals, sales on one side and spending by channel, prices, promotions and seasonality on the other, and estimates statistically how much each input moved sales. The method dates to consumer packaged goods firms in the 1960s and 1970s, but it fell out of favor in digital marketing because user-level data seemed more precise. The current trend is its return in a new form: open-source Bayesian models that run on a laptop, paired with controlled experiments, known as incrementality tests, that check whether the model's estimates hold.

Why it is happening now

Three changes drove the return. First, user-level tracking became less complete. Apple's App Tracking Transparency prompt, introduced with iOS 14.5 in April 2021, required apps to ask permission before tracking users across other companies' apps and sites, and most users declined. Browsers such as Safari and Firefox limit cross-site cookies by default. Attribution models built on these signals now see only part of each journey, and the missing part is not random.

Second, the evidence that attribution measured causation was weak even before the privacy changes. Gordon et al. (2019) checked observational attribution against large randomized Facebook ad experiments and found the observational estimates often missed, usually on the high side. Blake et al. (2015) ran eBay experiments in which paid ads on the eBay brand name added almost no sales, since searchers would have arrived anyway. Attribution had credited those clicks in full.

Third, the tools became accessible. Jin et al. (2017), researchers at Google, described Bayesian methods that build in carryover, the way an ad's effect lingers for weeks, and diminishing returns as spending rises. Meta released its open-source Robyn package, and Google later released Meridian, an open-source Bayesian MMM, making the approach available to firms without a consulting budget. A model that once took months and a specialist firm can now be fit by an in-house analyst.

How the trend works in practice

A modern MMM program has three parts. The model is fit to two to three years of weekly data and estimates each channel's contribution, its carryover and the point at which more spending stops paying. Experiments, such as turning paid social off in a randomly chosen set of regions for six weeks, measure incrementality directly for one channel at a time. The experimental results are then used to calibrate the model, keeping its estimates honest. Chan and Perry (2017) note that MMM alone struggles when channels move together, as they do when a firm raises all spending before a holiday, which is why calibration matters.

Section 2: Reach and Scope

Who is adopting it

The trend began with large advertisers, who always had the weekly data and budgets that MMM needs. Its reach has since widened in two directions. Direct-to-consumer brands that relied heavily on social media advertising, and that lost much of their tracking after 2021, adopted it to decide budgets across Meta, Google, TikTok and newer channels such as connected television. Mid-sized retailers and subscription businesses followed as open-source tools lowered the cost. Advertising platforms themselves now offer incrementality tests, a sign that measurement of causal effect, not only of clicks, has become an expected feature.

Which channels and decisions it covers

MMM's scope is broad by design: it can include television, radio, print, outdoor and direct mail beside digital channels, since it needs only spending totals and outcomes. That makes it the only common method that compares online and offline media on the same scale. It suits budget decisions, how much to spend in each channel next quarter, better than tactical ones, such as which ad creative to run, which still rely on platform tests and user-level data.

Where it does not yet reach

Small businesses with little spending variation and short histories cannot fit a reliable model; for them, simple holdout experiments are more practical. Firms with long, complex sales cycles, such as enterprise software, have too few conversions per week. And many teams still lack the statistical skills to judge a model's output, so the risk of trusting an uncalibrated model is real. Lewis and Rao (2015) found ad effects tiny next to the ordinary noise in sales, so even a careful model leaves wide error bands. The trend's reach is therefore broad but uneven: it is becoming standard for mid-sized and large advertisers and remains aspirational for small ones.

References

Blake, T., Nosko, C., & Tadelis, S. (2015). Consumer heterogeneity and paid search effectiveness: A large-scale field experiment. Econometrica, 83(1), 155-174. https://doi.org/10.3982/ECTA12423

Chan, D., & Perry, M. (2017). Challenges and opportunities in media mix modeling. Google.

Gordon, B. R., Zettelmeyer, F., Bhargava, N., & Chapsky, D. (2019). A comparison of approaches to advertising measurement: Evidence from big field experiments at Facebook. Marketing Science, 38(2), 193-225. https://doi.org/10.1287/mksc.2018.1135

Jin, Y., Wang, Y., Sun, Y., Chan, D., & Koehler, J. (2017). Bayesian methods for media mix modeling with carryover and shape effects. Google.

Lewis, R. A., & Rao, J. M. (2015). The unfavorable economics of measuring the returns to advertising. Quarterly Journal of Economics, 130(4), 1941-1973. https://doi.org/10.1093/qje/qjv023