With 30% tracking loss, last-click gives Paid search 2.6× its true contribution and TV gets nothing, though it really drives 247 conversions a week (82 of them arrive via a search click). MMM finds TV at 219 ± 18 a week — the right ballpark, but it can't tell you which customer, journey or creative did it.
God's-eye view
Extra conversions each channel really causes per week. Only knowable because this is a simulation.
What the journeys say
Credit from 3,000 tracked customer journeys, scaled up for lost tracking. It can't tell a caused sale from one that would have happened anyway.
What the weekly totals say
A regression of 104 weeks of sales on spend. Sees TV and the halo, with a 95% range, but nothing about individual customers.
Dashed line = true value. = MMM 95% range. The × badge compares each estimate with the truth.
Under the hood: each channel has a diminishing-returns curve, TV's effect carries over week to week and can drive brand searches, and email and search sit in front of people who were buying anyway. MTA reads 3,000 simulated converting journeys; MMM fits 104 weeks of simulated spend and sales. Same seed every time, so the same settings always give the same numbers.
MTA follows individual journeys and asks which tracked touch came before each sale. That makes it good for day-to-day decisions inside digital channels, but it hands credit to whoever was last in the room, including for sales that would have happened anyway. Anything it can't see, like TV or a browser that declined consent, scores zero.
MMM steps back and regresses weekly sales on weekly spend. It sees offline channels and halo effects and needs no cookies, but it only speaks at channel level, needs a couple of years of varied spend, and comes with error bars.
- MMM for budget splits across channels, offline media and planning for the year.
- MTA for in-flight tuning within digital: keywords, creatives, audiences.
- Incrementality tests (geo holdouts, conversion lift) as the referee: run them where the two disagree and calibrate both.
Neither view matches the God's-eye view on its own. Together, checked against experiments, they get close. The longer version: MMM vs MTA is the wrong question.
A toy simulation with made-up numbers — real models are far more careful.