Marketing Mix Modeling for E-commerce: A DTC Brand's Guide

Running a DTC brand means running 8-12 paid channels simultaneously while your VP of Marketing asks why CAC went up 22% this quarter. The answer is usually buried somewhere in the interaction effects between channels, but your Shopify dashboard cannot tell you that.

Marketing mix modeling was built for CPG companies with $50M media budgets. But the methodology works just as well for e-commerce brands spending $50K-$500K per month, if you adapt it correctly.

DTC-Specific Challenges

E-commerce MMM has three challenges that CPG does not. First, you have many channels relative to your data. A DTC brand with 18 months of data running Google, Meta, TikTok, Pinterest, email, SMS, influencer, and affiliates has 8 variables to estimate from ~78 data points. That is tight. You need to be smart about which channels to model separately vs. group together.

Second, DTC budgets shift fast. You might double your TikTok spend in a month because a creative went viral. That variance is actually good for the model (it needs spend variation to learn), but it means your results can change quarter to quarter.

Third, e-commerce has strong promotion effects. Flash sales, discount codes, and seasonal events drive huge revenue spikes that the model needs to separate from marketing effects. If you do not include promotion flags, the model will attribute that Black Friday spike to whatever channel you were spending on that week.

Which KPI Should You Model?

Revenue is the default, but it is not always the best choice. If your average order value varies a lot (like a fashion brand where orders range from $40 to $400), modeling orders instead of revenue reduces noise and gives cleaner estimates.

For subscription DTC brands, model new customer acquisitions. Your LTV math handles the revenue side. What you need MMM to answer is: which channel acquires customers at what cost? Revenue-based MMM will be distorted by retention and upsells.

My recommendation: start with revenue for simplicity, then run a second model on new customer count if acquisition efficiency is your primary concern.

Minimum Data Requirements for E-commerce MMM

Real Example: A $200K/Month DTC Brand

A skincare brand spending $200K/month across Google ($70K), Meta ($80K), TikTok ($30K), and influencer ($20K). Platform reporting showed total attributed revenue of $1.1M, but actual revenue was $750K. Classic over-counting from overlapping attribution windows.

The MMM results: Google was the most efficient at 4.2x ROAS (heavily branded search capturing existing demand). Meta was at 2.1x ROAS (lower than Meta reported, which was 3.8x). TikTok was at 1.4x ROAS but with the steepest growth curve, meaning it had room to scale. Influencer was at 2.8x ROAS, the biggest surprise.

The reallocation: they shifted $15K from Google (saturated) to TikTok ($10K) and influencer ($5K). Revenue increased 8% the next quarter with the same total budget. That is $60K in incremental quarterly revenue from a budget reallocation. See what this kind of analysis looks like in our demo report.

Common E-commerce MMM Mistakes

Modeling email and SMS as paid channels. These are retention channels with near-zero incremental cost per send. Including them inflates their contribution and steals credit from acquisition channels. Model them separately or exclude them and focus on acquisition spend.

Ignoring promotions. If you run a 20% off sale and revenue doubles that week, the model needs to know about the promotion. Otherwise it attributes the spike to ad spend, which overstates channel ROAS.

Not grouping small channels. If you spend $3K/month on Pinterest and $2K on Snapchat, group them as "other social." The model cannot estimate reliable coefficients for channels with tiny budgets.