When MTA says Meta is a top performer and MMM says it is weak, do not treat agreement as the test of whether either system is right. They observe different evidence, make different assumptions, and answer different parts of the same decision.
For a DTC brand with meaningful spend across several channels, the useful question is: what can each system see, what story do the readings tell together, and what should we test next?
Why MTA and MMM disagree
An MMM is trying to pull a channel-specific signal from noisy, correlated time-series data. If Meta moves independently while other channels move differently and the business moves with Meta, the model may isolate a strong Meta effect. If all channels rise and fall together, Meta may still be helping—but the model may not be able to separate its contribution from the rest.
That does not make a weak MMM estimate a verdict. Just because it is hard to tell whether John, Joe, and Jerry who saw a Meta ad are different from Mary, Billy, and Bobby who did not, does not mean the ad did not help. It means the causal difference is difficult to isolate.
Click credit and demand creation are different things
Meta is often a bottom-funnel closer. MTA can see people click a Meta ad and buy, so it gives Meta substantial credit. But those people may have been made ready by upper-funnel media, brand work, search, or a promotion. MMM may give Meta less incremental credit because some of those sales would have happened without the click.
The reverse can also happen. CTV and linear television create almost no click trail, so MTA may show nothing. MMM can detect that spend or impression waves coincide with changes in total sales, including sales that happen offline or later through another channel.
Timing is not as clean as a model needs it to be
MMM uses an assumption about the average time between exposure and purchase, often called decay or adstock. A real customer may see an ad repeatedly for months, consider it, and buy when a Memorial Day promotion finally gives them a reason. People exposed in December, February, and May can all convert in May. Major impression waves around Black Friday, Presidents’ Day, or promotions make a clean channel-specific time pattern harder to find.
A durable identity system can let MTA observe a longer click path: someone clicks, returns, clicks again, and buys. But MTA still cannot reliably see the many non-click impressions that may have shaped that path. MMM sees the aggregate wave; MTA sees part of the person-level journey.
Context changes MMM’s answer
MTA generally observes a click and a purchase. MMM can include promotion periods, seasonality, and other business context, which lets it avoid assigning every sales increase to a paid channel. That can explain why MTA reports a 6x Meta ROAS during a promotion while MMM assigns part of the demand to the promotion or baseline instead.
New-versus-returning customer analysis can make the contrast even more useful. MTA may show existing customers clicking a video ad and purchasing. MMM may find that many would have returned anyway, while also detecting a delayed new-customer halo that click data misses. Neither reading is automatically the whole story.
First, make sure you are comparing like with like
Before treating a disagreement as insight, make the measurement foundation comparable. Start with the spend taxonomy. Both systems should use the same channel and sub-channel definitions: Google Search, PMax, YouTube, or Meta awareness versus conversion campaigns. The goal is not maximum granularity; it is a consistent labeling system.
Then check the revenue scope. An MMM can include phone, Amazon, subscription, and other offline revenue. MTA generally cannot. If phone and Amazon make up 40% of the business, an MTA may still be useful, but it represents a smaller slice of the business than the MMM. Do not call that apples to apples.
Also account for recurring revenue. MMM may associate February spend with March subscription rebills in its time-series view; MTA may see the initial click in February but not the auto-bill. Filter auto-bills if a direct comparison is the aim. If new-versus-returning customers matter, make sure both systems have reliable flags.
After those checks, the reports will still not match. That is normal. Ask what each system can see that the other cannot, then assemble the story.
How to make a budget decision while evidence is incomplete
You can let a tool drive the business, or let the business drive the tool. A budget optimizer can recommend an allocation; an MTA platform can prescribe a KPI. Those are useful inputs, but they can become deterministic. History may say one thing while today’s market, creative, promotion, or competitive environment says another.
A more practical approach starts with the business’s operating norm—often a last-click target—and uses MMM, MTA, and testing to make deliberate exceptions. A channel that seems less incremental may need a higher last-click hurdle. A channel with a proven halo may warrant a lower one.
Imagine Meta normally must deliver a 3.0 last-click ROAS. If one or more MMMs point to lower incrementality, inspect their confidence intervals and the agreement between models. While a lift test is pending, the business might increase Meta’s operating hurdle 15%, to 3.45. That can reduce Meta spend and free budget to trial in a more-incremental channel.
Do it in stages. An upper-funnel reallocation may produce a halo that takes a month to show up. The P&L may not support a major near-term sacrifice, so the right move can be a measured shift, regular recalibration, and executive alignment about what the business is willing to learn.
Use a lift test to close the case
A lift test is the causal check. In a user-level holdout, a platform such as Meta can withhold ads from a randomly selected eligible group while another comparable group sees them. The comparison asks: how many people bought without the ads, and how many more bought after being exposed?
If the held-out group still buys at a meaningful rate, the ads may be recording sales the brand would have received anyway. If few buy without the ads, the channel may be highly incremental. Geo and strategy holdouts can answer similar questions where user-level identity is unreliable or unavailable.
Lift studies take care. They are parallel-universe exercises: the test and control must stand in credibly for what the other would have done, and the learning must be relevant to the future decision. When they are designed well, their results can calibrate MTA expectations and sometimes be used as known constraints in an MMM.
MTA and MMM FAQs
Should MTA and MMM have the same ROAS?
No. They use different evidence and can include different revenue scope, conversion timing, promotions, seasonality, and impressions. Align the spend taxonomy and understand the coverage before interpreting a gap.
Why does MTA give Meta more credit than MMM?
Meta may be closing demand created by upper-funnel channels, promotion activity, or returning customers. MTA can observe the click and purchase; MMM is trying to estimate whether the channel created sales that would not otherwise have happened.
What should a brand do when MTA and MMM conflict?
Use the reports to form a hypothesis, make a measured operating adjustment where appropriate, and run a lift test when the decision is material. Do not select a winner simply because the reports disagree.