It's been a while. Let me explain why.
I paused The Prescient Perspective more than a year ago. We had a choice between publishing another marketing newsletter or going heads-down on the research that would make a newsletter worth opening.
We picked the research.
We now have two research papers we couldn't have written a year ago. But these papers are honestly the culmination of some serious hard work developing our proprietary MMM from the ground up. So, essentially, these learnings took us five PhDs, six years, $25M in R&D, and $7B in ad spend worth of training data.
That's what allowed me to step back, confident I had insights worth the click.
Every Tuesday, I'm going to send you one idea about how marketing measurement, forecasting, and optimization actually work. No gated PDFs. No benchmark reports. Just the view from here, with the math behind it.
The view from here this week:
You are on a rocket.
If you're running a DTC or omnichannel brand, that's what the last five years have felt like. You light the engines in Q4, you hit Black Friday at max thrust, you pray the trajectory holds, and you check the telemetry in the war room at 2am.
If you've compared the reports from your platforms, you've probably seen they don't add up. Meta says it drove 4.2x. Amazon says 3.8x. TikTok says 6x. You add it up and the numbers come to 147% of your actual revenue. Something in how it's being measured just doesn't fit.
The MMM is supposed to be the guidance system, but it's three jobs for one system:
• Measurement the telemetry. Where is the rocket right now? Which channel actually drove that sale?
• Forecasting the trajectory. Where is it going? What happens next week, next quarter, at peak?
• Optimization the guidance itself. Where do we fire next? Which channel gets the next dollar?
All three depend on the same equations being right. Attribution, forecast, and budget recommendation are not three separate problems. They are one problem with three outputs, which means the math matters more than ever.
But many current solutions just weren't built to keep up with the modern marketing environment.
In 1962, NASA lost Mariner 1 to a hyphen. A single missing overbar in one equation in the guidance software sent a $135M spacecraft into the Atlantic Ocean 293 seconds after launch. The hardware was fine. The data was fine. The math was wrong, and wrong math at the wrong moment is the whole mission.
Our research team (Cody Greco and Philip Hofmeister) spent the last year studying, with synthetic ground truth across ten independent datasets, the mathematical assumptions most MMMs share today. The main one: your channels are additive. Meta's contribution plus TikTok's contribution plus Amazon's contribution equals your revenue, each measured independently.
The reality of modern DTC is different. Your channels are inextricably linked. Your YouTube spend changes your Amazon search volume. Your CTV campaigns change what your retargeting costs.
That's what makes modern marketing so powerful. It's also what makes it so complicated to measure and predict.
Most of the tools contain math designed for stable conditions. When that math gets tested at peak, the consequences show up exactly when you can least afford them.
During peak marketing events, both your spend and organic demand increase. Your channels interact more than usual. Here's what that looks like in practice, based on our research across ten independent datasets:
• Measurement: The math underneath an MMM determines how credit is split between your paid channels and underlying demand. Across ten datasets with known ground truth, we saw that attribution vary 2-3x depending on the math the model used to make the split.
• Forecasting: When we built channel interaction context directly into the equations, forecast error dropped by more than 90% in one of our tests. Predictions moved from significantly off to close to what actually happened.
• Budget recommendations: At BFCM, the gap between budget recommendations and the true optimum varied by up to 81% depending on the math underneath the model. The mechanistic approach we tested stayed within roughly 1%.
• Lift testing: A lift test adjusts the scale of an answer, but it doesn't change the math that generates it. If the math doesn't represent how your channels interact, no rescaling is going to fix it.
That last point is the #MathMatters thesis in one sentence. For a lift test to deliver what it promises, the math of the model underneath has to be up to it.
That's why we built Prescient differently and designed our model to match the structure of modern marketing systems. What was easier to track didn't matter. Nailing the math did.
Over the next seven weeks, I'll break down each of these problems and what fixing them actually changes for your brand, but you can start by checking out the paper for yourself:
The Halo
Exclusive insights, every week.
Subscribe to The Halo for sharper marketing thinking.