Ad platforms find your next customer by learning from your current ones. Every click, visit, add to cart, phone call, store visit, and purchase is a lesson. Most of it never arrives.
Marketing mix modeling doesn't change that. MMM and incrementality testing can tell you where your budget is working, even when user-level tracking has gaps. They don't change what Meta, Google, TikTok, Snap, and Amazon learn from while they spend that budget. Measurement and signal solve two different problems, and fixing one leaves the other where it was.
What do marketing mix modeling and incrementality testing measure?
Marketing mix modeling is a statistical method that estimates each channel's contribution to sales from aggregate data: spend, revenue, and outside factors like seasonality and pricing, tracked over time. It doesn't follow individual users, which is why it holds up when cookies and click IDs go missing. Incrementality testing compares a group exposed to ads with a group that isn't, so the difference shows what the ads actually caused.
Both answer the same question: did this spend produce sales that wouldn't have happened anyway? That question matters for budget. It's a different question from the one ad platforms answer every time they enter an auction.
Measurement and optimization run on different data
Ad platforms decide in real time. Each auction, Meta's Advantage+ or Google's Smart Bidding chooses who sees an ad based on what the platform has learned from the conversion events it received. If a purchase never arrives, or arrives without a click ID or customer identifiers, it doesn't teach the platform anything. We cover this in more depth in what ad platform AI decides when it receives your conversion event.
MMM works after the fact, on totals. It reads weeks or months of spend and sales and tells you how to split the next budget. Nothing in that output goes back into a platform's bidding. A better model changes how much you give TikTok. It doesn't change which shoppers TikTok picks.
Measurement decides where the money goes. Signal decides what the platform does with it.
How signal loss still shows up in your measurement
MMM doesn't need user-level tracking, but it measures each channel as it actually performed. If a channel ran on incomplete signal, the platform optimized from a partial picture of your buyers, and its results reflect that. The model then measures the weaker version of the channel. It may recommend cutting spend there, when the cause sits in the data the platform received, not in the channel.
Incrementality tests have the same blind spot. A holdout test measures the campaign as it was optimized, so a campaign learning from missing purchases tests worse than it could. Platform-run lift studies, such as Meta's Conversion Lift, also count the conversions the platform receives. Purchases that never reach the platform don't show up as lift.
Daily, ad-level attribution views sit closest to the problem, because they're built from click and conversion data in the first place. Gaps in that data become gaps in the report. Our attribution modeling guide covers how missing events, delayed signals, and poor match rates distort credit assignment.
Measurement can show that a channel is underperforming. It can't tell you whether the cause is creative, audience, or signal.
Check signal before you act on what measurement tells you
This isn't a choice between measurement and signal. Brands spending across several platforms usually need both. The order matters when the two disagree, though.
Before cutting a channel because a model or a test says it's underperforming, check what that platform has been receiving:
- Orders versus reported purchases. Compare backend orders with the purchases the platform reports for the same period. A large gap means it's learning from fewer buyers than you actually have.
- Match rate. Check how many events the platform can connect to a real person. Match rate by platform shows how Meta, TikTok, Snap, and Google differ.
- Duplicates. Check whether any order is counted more than once.
If the signal is incomplete, fix it first, then let the measurement run again. The result will describe the channel, not the gap.
Journify works at this layer. It captures conversion events across web, mobile, offline, CRM, ecommerce, and backend systems, validates and enriches them, and delivers them through official Conversion APIs before any measurement happens downstream.
FAQ
Does marketing mix modeling need conversion tracking?
Not at the user level. MMM works from aggregate spend and sales data, so it needs accurate totals, not click IDs or pixels.
Is incrementality testing affected by signal loss?
Yes, in two ways. Any test measures the campaign as optimized, and platform-run lift studies only count conversions the platform receives.
Can MMM replace a Conversion API?
No. MMM informs budget decisions. A Conversion API delivers the events ad platforms optimize from. They do different jobs.
Should I fix signal or invest in measurement first?
Check signal first when a measurement result would lead you to cut a channel. Otherwise you may be cutting a channel for a data problem.
Better measurement makes budget decisions sharper. It doesn't fix what ad platforms learn from, and that's where the spend actually gets optimized.
Book a call to see what your platforms are learning from today.
