BFCM is your window. You've got full thrust, demand is high, and every dollar you fire into a channel has the potential to cover more ground than at any other point in the year. But that thrust only gets you where you want to go if the math under your guidance system can keep up.
Last week I told you the math under most MMMs has structural limits today. This week, I want to walk through three specific places where that shows up, and what our research found.
A quick reminder of what your MMM is doing all at once, from the same equations:
• Measurement what happened. Attribution.
• Forecasting what will happen. Trajectory.
• Optimization what to do next. Guidance.
Three jobs. One set of equations. Peak is exactly where the math gets tested on all three at once, and it's also the moment a marketer most needs clarity.
Here's where the math gets tested at peak, in three places:
1. Measurement. Spend and demand move together at peak.
You spend more when you know demand is higher. But most MMMs aren't built to untangle the question that follows: was that revenue driven by your ad, or by the fact that everyone shops on Black Friday? Mathematically, the model can't tell. There are infinitely many ways to split your revenue between "things your ads drove" and "things that were going to happen anyway," and they all fit the data equally well. The model tends to pick one based on its own built-in assumptions, and the assumption the model brings to that split is what tilts the answer.
We call this baseline leakage, and we measured it across ten datasets where we already knew the right answer. Standard models overattribute 2-3x more revenue to your paid channels than a model built to reflect how marketing actually works. At BFCM, that means Meta, TikTok, or Amazon are quietly getting credit for sales that were going to happen anyway, and that ripples directly into next year's planning.
2. Forecasting. Efficiency at peak shifts week to week.
The same dollar of TikTok spend on November 20th is not the same dollar on November 26th. Context, like promotions, holidays, and competition, changes how well your spend converts. The math in most MMMs simplifies by treating efficiency as roughly stable and absorbing the variation into "seasonality." That works in steady-state. In high-variation contexts, the simplification matters more.
That means the forecast for next week is built on the assumption that this week's efficiency holds, when it really doesn't. When context is built directly into how the model reads channel performance, forecast accuracy improves dramatically. In one test, campaign-level forecast error dropped by more than 90%. The model's predictions moved from significantly off to close to what actually happened.
3. Optimization. Standard response curves flatten. Peak demand can climb.
The response curves that power most MMMs (Hill, Weibull, S-curves) are all designed to flatten out eventually. It's simply how they're built. But during BFCM, when demand is unusually high and your channels are capturing it near-linearly, a curve built to flatten will signal saturation before the market actually saturates. The recommended spend gets capped earlier than the channel needs.
Growth is left on the table.
Put the three together and you find the headline from our research:
At BFCM, standard models recommended overspending by up to 81% versus the true optimum. Prescient's model stayed within roughly 1%.
As you build your BFCM plan, three questions worth asking about the math powering your MMM:
• How does the math split revenue between media and underlying demand at peak?
• How do the equations adjust efficiency for the specific context of the week, not just the season?
• Where does the response curve flatten, and what would that mean if a channel is still scaling past that point?
Every MMM is built around a particular set of math. Most of it was designed for steady-state. Peak is a different set of conditions, and that's the gap our research set out to close.
The Halo
Exclusive insights, every week.
Subscribe to The Halo for sharper marketing thinking.