MMM vs. MTA vs. Incrementality Testing: Which Do You Need?

Imagine you’re deciding whether the next budget increase should go to paid search or television. The paid search platform reports a high return on ad spend (ROAS). The team running television argues that its ads created the demand that search captured. Finance wants to know which investment deserves the next dollar.

A customer who searched for your brand may already have intended to buy. Another may have discovered you through television, talked to a friend, and purchased in a store. Dividing credit across the interactions you can track won’t tell you what caused either sale.

That’s the question this comparison needs to answer: which approach will help you make a better investment decision?

MMM vs. MTA vs. Incrementality Testing at a glance

MMM and incrementality testing work well together. A model can show you where a budget change looks promising. A test can help you find out whether that change delivers.

Why MTA is dead as a budget allocation tool

Multi-touch attribution assigns credit to the customer interactions it records before a purchase. One method might split credit evenly. Another might give more weight to recent interactions. More sophisticated versions use algorithms to decide how much credit each touchpoint deserves.

But changing how you divide credit doesn’t solve the underlying problems.

It misses too much of how people buy

Privacy rules and platform policies limit what advertisers can track. UK rules require consent for advertising tracking technologies. Apple requires permission for tracking across other companies’ apps and websites. You can’t build a reliable investment plan on the assumption that those permissions and identifiers will always be available.

Walled gardens add another problem. Each platform can show you activity within its own environment, but that doesn’t give you a complete view across platforms.

Then there’s everything that doesn’t arrive as a trackable click: linear TV, outdoor advertising, a conversation with a friend, a visit to a store. These experiences can influence a purchase without appearing in the path your attribution tool records.

The risk: you keep moving money toward the channels that are easiest to track, while underfunding the ones that created the demand.

It can reward advertising for a purchase that was going to happen anyway

Think about retargeting. Someone visits your site, browses a product, sees an ad later, and buys. MTA gives the ad credit.

Did the ad persuade them, or were they coming back to buy anyway?

Retargeting reaches people who have already shown interest. That makes them more likely to purchase before the ad ever appears. Giving the ad credit for that existing interest confuses who you targeted with what the advertising accomplished. That’s the problem behind the term selection bias.

Better tracking alone can’t answer what those shoppers would have done without the ad.

Keep using clicks, conversion paths, and platform reports to understand campaign delivery and customer behavior. Just don’t mistake those observations for proof that a channel deserves more budget.

MMM helps you see where the whole budget should go

Marketing mix modeling (MMM) looks at how business results change over time alongside marketing activity, pricing, promotions, distribution, seasonality, and other relevant factors. It uses that data to estimate what each contributes.

That lets you evaluate linear TV alongside paid search and social without needing to identify every person who saw an ad and later bought. It can also reveal effects across retailers that no single platform report would capture.

For Dr. Scholl’s, M-Squared modeled media and trade activity across retail categories. The analysis estimated that trade and promotions received nearly twice the investment of national media, while delivering less than half its incremental ROAS. It also identified effects from Amazon retail media on sales at other retailers.

Those findings pointed to opportunities to move money. Looking at each retailer’s reporting separately could have missed them.

MMM also helps you compare spending plans: what might happen if you increase TV, reduce promotions, or put more money into a channel that’s already heavily funded?

Before acting on those recommendations, check how well the model explains what’s driving sales. If TV and paid search always increase together, the model can struggle to separate their effects. If it leaves out an important promotion or distribution change, it can give marketing credit for growth that came from somewhere else.

A model that predicts sales well hasn’t automatically explained what caused them. At M-Squared, we use incrementality testing to challenge the model’s assumptions and check its recommendations before you move budget.

Incrementality testing helps you check a specific bet

Incrementality testing asks what changes because you change your marketing. A well-designed test compares a test group or market, where advertising changes, with a control group or market, where spending stays unchanged or the campaign is withheld.

You might hold back advertising to find out how many sales it currently adds. Or increase spending in selected markets to see whether more investment produces enough additional business to justify the cost.

Those are different questions. A channel can be worth keeping at today’s budget without being the best place for your next dollar.

TeePublic’s incrementality testing makes that distinction concrete. Testing found incremental lift from Google Performance Max. Meta also drove real lift, but its spending had moved past peak efficiency. Both channels generated sales. The findings supported different decisions about where to invest.

The lesson isn’t that one platform always works and another doesn’t. It’s that a channel’s contribution and its capacity to absorb more budget need separate answers.

Tests need scrutiny too. There may be too few purchases to distinguish an advertising effect from normal ups and downs. Ads intended only for test markets may also reach control markets, blurring the difference between the groups and making the advertising’s impact harder to measure. A short test may finish before customers have time to buy or miss effects that take longer to appear.

An inconclusive result means the test hasn’t answered the question. It doesn’t mean the advertising had no effect. And a result at one spending level doesn’t settle how the channel will perform at twice that budget.

Which approach should you start with?

Start with the decision you’re trying to make.

If you’re deciding whether to scale or cut one channel, look at testing first. Check whether you can create a meaningful comparison, gather enough data, and make the spending change without disrupting the business.

If you’re planning across the whole marketing budget, assess whether you’re ready for MMM. You need dependable business data and enough variation in marketing activity to learn from. The model also needs to reflect how your business actually grows. Revenue alone doesn’t tell you whether you’re ready.

If you already have MMM, test the recommendations with the most money riding on them. Where would a wrong assumption change the plan? That’s a better starting point than running another analysis simply because it’s on the calendar.

You don’t need every method on day one. A simpler business may get more value from sound reporting and a focused test than from a measurement program it isn’t ready to use.

Bring the evidence into the same decision

Having an MMM and running tests won’t help much if the model stays in a planning deck while platform ROAS drives every weekly spending decision.

Triangulation means bringing the evidence together around the decision you need to make. When the findings disagree, investigate what each actually measured before deciding which result to act on.

Start by agreeing with finance on the outcome. Are you trying to acquire new customers profitably, increase repeat purchases, or grow contribution margin? The Marketing Accounting Framework establishes those business definitions so the measurement answers the question both teams care about.

Next, check whether you’re comparing like with like. An MMM might estimate a channel’s average return over the past year. A recent test might measure the return from increasing its budget. Those answers can differ because they describe different investments. Use that distinction to decide what you need to learn next.

Then carry what you’ve learned into everyday reporting. The Multiplier Engine uses evidence from models and tests to calibrate ongoing performance reporting, so teams can make decisions between studies. Those adjustments need revisiting as conditions change. They don’t reconstruct missing customer journeys or prove what each touchpoint caused.