Media Mix Modeling and Seasonality: How MMM Handles Black Friday Spikes
Every November, media mix modeling for ecommerce runs into the same problem. Sales jump on Black Friday, ad spend jumps on Black Friday, and a model that can't tell those two things apart will credit your ads for demand the calendar created. Get the seasonality handling right and MMM becomes one of the best tools you have for planning peak budgets. Get it wrong and it tells you to spend more on the days you least need to.
This post explains how a model separates holiday demand from ad impact, what to check in your own model before you trust its Q4 read, and how MMM and click-based attribution fit together through Black Friday (November 27) and Cyber Monday (November 30) this year.
Why Black Friday is hard for any media mix model
Black Friday is hard because spend and demand rise at the same moment, so the model has to decide how much of the spike belongs to each. Adobe forecasts $12.9 billion in US online spending on Black Friday and $15.1 billion on Cyber Monday, with Cyber Week accounting for about 17% of the November to December season. A lot of that would happen if you ran no ads at all.
Brands make this harder in a predictable way: they raise budgets precisely when demand is already rising. To a regression, a channel whose spend doubles every Black Friday looks like it doubles sales every Black Friday. That's a confounding problem, and no amount of extra data fixes it on its own. The model needs to know the holiday is there.
There's a second effect that's easy to miss: pull-forward. Shoppers who would have bought in mid-November wait for the sale, so the weeks before Black Friday dip and the weeks after can soften too. A model that only flags the spike week will read the dip as weak media and the spike as strong media. Flag the surrounding weeks as well, or the model will misread both.
How MMM separates holiday demand from ad impact
A well-built MMM splits sales into a baseline (what you'd sell with no media) and media contributions, and it gives the baseline enough flexibility to rise during known peaks. There are three common ways to do that.
1. A flexible baseline over time
Google Meridian MMM uses a time-varying intercept that adjusts for trend and seasonality automatically. Its documentation notes that because of this, separate seasonality variables aren't required, though you can turn the automatic adjustment off and supply your own. The number of knots controls how flexibly the baseline can bend. Too few, and a Black Friday spike gets pushed onto media. Too many, and the baseline absorbs real ad effects.
2. Holiday and promotion control variables
Holiday dummies flag specific weeks (Black Friday week, Cyber Week, Prime Big Deal Days) so the model can assign them their own baseline bump. Promotion variables do the same for your sitewide sales. Google Meridian MMM's guidance treats these as control variables, whose purpose is to control for confounding. For ecommerce brands, a promo depth variable (average discount that week) is often more informative than a simple on or off flag.
3. Enough history to see the pattern repeat
A model can only learn that late November is always high if it has seen several late Novembers. The Google Meridian MMM documentation recommends at least two years of weekly data for geo-level models and three years for national ones. With only one Black Friday in the data, the model has one example of the spike and no way to tell which part repeats.
What to check before you trust your model's Q4 read
Whether you run Google Meridian MMM yourself, use a vendor model, or work from AdBeacon's integration, ask these questions of the Q4 output:
- Does the baseline rise in late November? If the baseline stays flat while sales triple, the model is crediting media for seasonal demand.
- Does any channel's ROI spike only in Q4? A channel whose modeled ROI jumps from 1.5x to 6x during Cyber Week is more likely absorbing holiday demand than suddenly working four times better.
- Are promotions modeled separately? If your 30% sitewide sale isn't in the model, its effect is landing somewhere, usually on whichever channel you scaled that week.
- Has the model been calibrated with a lift test? Google moved GeoX, its geo testing library for Google Meridian MMM, to general availability in September 2026, and geo test results can feed into the model as priors. A calibrated channel is far less likely to drift during peaks.
AdBeacon's Google Meridian MMM integration sits next to click-only attribution in the same account, free on AdBeacon plans.
MMM vs multi-touch attribution during the holidays
MMM and multi-touch attribution answer different questions, and Black Friday shows why you want both. MMM works on weekly aggregates and estimates how much each channel contributes over time, including demand effects clicks never capture. Click-based attribution works on individual orders and tells you which campaigns, ads, and products drove purchases this morning.
During Cyber Week, that difference matters in practice:
- MMM is too slow for in-flight decisions. A weekly model won't tell you on Black Friday afternoon that one ad set's CPA doubled. Click-based attribution will.
- Click-based attribution can't see baseline demand. It credits every tracked click, including clicks from shoppers who would have bought anyway. MMM can estimate that baseline.
- They check each other. If your model says Meta drove 25% of Q4 revenue and click-based attribution credits Meta with 40% of attributed orders, the gap tells you how much of Meta's click credit went to shoppers who were already coming.
We cover the broader trade-offs between methods in MMM vs attribution vs incrementality. For a closer look at whether running the model yourself makes sense, see Google Meridian MMM: a powerful tool, but is it right for you?
A practical Q4 workflow with MMM and click-based attribution
Here's how the two fit together across the season:
- Mid-October to early November: Use MMM response curves to set Cyber Week budgets by channel. The Scenario Planner for Google Meridian MMM, added in February 2026, lets you test budget splits without code.
- November 26 to 30: Run daily decisions on click-based attribution. Watch ROAS, nCAC, and MER by channel in AdBeacon, with AdBeacon's numbers beside platform-reported ROAS so you can spot inflated platform claims before you scale into them.
- December: Refit the model with the new holiday data, holiday flags, and promo depth. Compare its read on Cyber Week to what click-based attribution showed.
- January: Use both to plan Q1 reallocation. Our post on using MMM insights to shift budgets walks through that step.
AdBeacon's Google Meridian MMM integration sits in the same account as its click-based attribution, so the long view and the daily view are a tab apart instead of two vendors apart. The integration stays free on AdBeacon plans, which removes the usual cost barrier to running MMM next to attribution.
Let the model handle the calendar, let clicks handle the day
Media mix modeling handles Black Friday well when the baseline can flex, holidays and promos are modeled explicitly, and the history covers more than one peak. Pair it with click-based attribution for in-season decisions and you get a long view and a daily view that keep each other honest. To see AdBeacon's Google Meridian MMM integration and click-based attribution on your own store data before Cyber Week, book a demo with AdBeacon.
Frequently Asked Questions
What is media mix modeling in ecommerce?
Media mix modeling is a statistical method that estimates how much each marketing channel contributes to sales using aggregated weekly or daily data, separating media effects from baseline demand, seasonality, and promotions.
How does MMM handle Black Friday and holiday spikes?
A good model lets its baseline rise during known peaks, using a flexible time trend, holiday flags, and promotion variables, so seasonal demand isn't credited to ads.
What's the difference between MMM and multi-touch attribution?
MMM estimates channel contribution from aggregate data over time, including effects clicks can't capture. Multi-touch attribution assigns credit for individual orders based on tracked touchpoints, which suits daily campaign decisions.
How much data does Google Meridian MMM need?
Google recommends at least two years of weekly data for geo-level models and three years for national-level models.
Is AdBeacon's Google Meridian MMM integration free?
Yes. AdBeacon's Google Meridian MMM integration stays free and sits alongside its click-based attribution in the same account.