How to Build a Business Case for Moving From MTA to Incrementality Testing

Build a business case for moving from MTA to incrementality testing by showing that the decision depends on causal lift, not just recorded customer paths. Use MTA to understand observable customer journeys, MMM to understand the media portfolio, and incrementality testing to establish what caused a business outcome.

That answer becomes useful only after you name the decision: should you spend up, spend down, pause a channel, or change strategy?

If the decision is clear but you are unsure which method should lead, use the MMM or incrementality quiz for a candid starting point.

Should you spend up or down on a channel? If you spend up, will you create more sales—or just pay for sales you were going to get anyway? Are two channels creating useful extra touchpoints, or claiming credit for the same order?

Those are not abstract questions. When a business is running several serious paid channels, platform-reported conversions can easily add up to more orders than the cash register shows. That doesn't necessarily mean the platforms are lying. It means they're observing the same messy customer reality from different angles.

Start with the decision, not the tool

Before choosing MTA, MMM, or a testing program, get specific about the decision and the information missing from it. You may be trying to understand whether a channel is incremental; whether a promotion is closing demand another channel created; or whether two channels are producing a halo effect instead of cannibalizing each other.

Once the question is clear, the methods have different jobs. They are not competing versions of a single truth.

MTA vs. MMM vs. incrementality testing: what each can see

For the common multi-touch attribution versus marketing mix modeling question, start with the decision. MTA is useful when the visible customer journey needs inspection. MMM is useful when several activities compete for budget and the question is broader than a click path. Neither makes a causal claim by itself. When the allocation decision is consequential and testable, incrementality testing is the tie-breaker.

MTA vs. MMM: choose the question first

Choose MTA when the immediate work is to inspect recorded customer paths: which visits, clicks, returns, and orders can the team connect? Choose MMM when the decision is how to allocate a portfolio across channels over time, including activity that does not leave a click trail. MTA assigns credit within the journey it can observe; MMM estimates contribution from patterns in portfolio and business data. They can disagree without either system being broken.

A practical sequence is to use MTA to find a journey or reporting question worth investigating, use MMM when the allocation question is broader, and use a feasible incrementality test when a specific spend change needs a causal answer. The method should follow the decision, not the software already in the stack.

If the first obstacle is whether clicks, sessions, and orders can be reconciled at all, begin with a marketing data foundation. If the record is usable and the question is about observed paths, review what a custom MTA system needs before choosing a rule.

MMM or incrementality? Two different decisions

MMM asks a portfolio question: across channels, promotions, seasonality, and demand, what appears to be contributing to the business outcome, and where should the next budget move be considered? It uses variation over time to put a wider media plan in one view. That makes it useful when several activities are competing for the same budget.

Incrementality testing asks an intervention question: did a defined change cause lift? The change might be pausing a channel in selected markets, changing the level of spend, or withholding media from an eligible audience. The test needs a credible comparison for what would have happened without that change. Its finding applies to the intervention, conditions, and period tested; it does not settle every allocation question across the portfolio.

Neither is a default answer. MMM needs usable history, meaningful movement in spend or activity, stable outcome data, and enough context to account for material promotions and demand shifts. A test needs an intervention the business can afford to make, sufficient outcome volume, and a treatment and comparison that can hold up. If those conditions are absent, forcing either method can produce a precise-looking answer to the wrong question.

A team may need both in sequence. Use MMM to frame the wider allocation story and identify the assumption with the most consequence. Then test that specific intervention when a credible design is possible. The test can calibrate how the team interprets the model; the model can point to the next question worth testing. Sometimes a hard test result comes first, and then informs the portfolio discussion. The order follows the decision, data, and practical downside.

For a candid starting point, take the MMM or incrementality quiz. If a test is the likely next move, use the incrementality testing framework to pressure-test the intervention and comparison. If the question is still about what a visible customer journey can tell you, read what MTA can and cannot answer.

Multi-touch attribution (MTA): the observable journey

MTA is usually the fastest place to begin. It can start producing useful information without a year of history, especially when a brand has durable first-party identity signals and a partner capable of resolving repeat users. It is good at showing how people click to a site, return, and click again.

But every click is not the same thing as causation. MTA cannot fully tell you whether a person saw an ad without clicking, changed devices, visited family, or was influenced by someone else's exposure. Its strength is observing the journey it can see—not proving every influence that created the sale.

Marketing mix modeling (MMM): the portfolio over time

MMM is useful when the question is bigger: when spend changes, does the business change? It can put spend, seasonality, promotions, and demand into one portfolio-level view. It can make a compelling inference that a channel is contributing to sales even when that contribution does not appear cleanly in click data.

It is still an inference. Correlation is not automatically causation. A model gives you clues and a range of plausible answers; it does not certify the future.

Incrementality testing: the causal tie-breaker

Tests calibrate both of those perspectives. In a user-level holdout, an eligible audience is randomly divided so one group sees media and another does not. In a geo test, a defined market becomes the container: change the media in one set of locations and compare the business outcome to an appropriate control.

Geo tests can be especially useful when cookie-based or cross-device identity is unreliable. If a customer switches phones, buys a new laptop, or tells their spouse about a product and the spouse buys it, the click trail may fall apart. A market-level test can still tell you whether changing the tide moved the boats.

MMM vs. incrementality testing, head to head

An MMM looks at the tide, and a test checks the boats. Here's how they stack up.

Which comes first?

It depends on the decision: if several channels are fighting for budget and you need one plan, start with an MMM. If one specific spend change needs a causal answer, test it.

And you can do both, which is really the point. A well-designed test can calibrate an MMM. And when an MMM says a channel is extremely strong or extremely weak but the read is too unclear to trust, that's often a reason to test before you cut it.

As a rough guide, with one main channel and around $1–2 million in sales, MTA may take you far enough. Once you're around $3–5 million or more, have two to five meaningful channels, or need to understand cross-channel impact, combining an MMM, testing, and MTA gets more valuable. These are guides, not hard rules. Brands with lower volume get evaluated case by case.

No model owns the truth. Use the MMM to form the theory, then test it.

Don't search for one correct attribution model

One of the most common MTA mistakes is asking whether first click, last click, U-shaped attribution, time decay, or an algorithmic model is the correct model. The executive impulse is understandable: give me a North Star, and I'll optimize to it.

But the customer is not a clean machine. They see ads, click some of them, think about promotions, compare competitors, look at the weather, and sometimes return through a completely different path. Each attribution model moves credit around to illuminate a different theory. Use multiple points of view, then test the theories that matter to your budget.

What to do when MTA, MMM, and tests disagree

Multiple MMMs may return different channel scores. A single MMM can return a range. MTA can disagree with both. The question “which one do I trust?” is often just the first question in disguise: which model is right?

Think of measurement like weather forecasting. If several independent models broadly agree, you have more confidence that you've found a real learning. If they conflict, the conclusion may be that you need more evidence—or that the decision is not yet predictable enough to make with confidence. That is a responsible answer.

The mistake is choosing one lens as doctrine. Matched markets preserve the messiness of real cities; synthetic controls can build a closer historical counterfactual. Where practical, validate more than one comparison before a test and use the agreement—or disagreement—to understand how strong the result really is.

How do MTA, MMM and incrementality testing work together?

They're one loop, not three tools fighting over the truth. MTA is the day-to-day view of the journeys you can see. A test answers one budget question with causal evidence. An MMM is the portfolio view, anchored by what the tests found. And each test result changes how you read the other two.

How does the loop run?

1. Form a theory. I use every single point of view to put together my theories. Then I test the theories.
2. Test the one that matters for the next budget decision. Agree on the action before it starts: "If this works, we will do it." Not sure what a geo test is? Start there.
3. Anchor the MMM with the result. Rebuild it on a deliberate cadence: no faster than monthly, often quarterly or twice a year. Here's what a test can calibrate in an MMM.
4. Keep reading MTA and last click. You don't rebuild them. You read them with what the test showed.

Over time, some teams also calibrate their attribution to lift-test results. Others keep that view side by side with the flatter models, since customer behavior and model assumptions change.

Last click isn't nothing. But it undervalues upper-funnel channels, so read it with the test in mind. The same goes for cross-channel attribution: it's observed credit, and the causal answer still comes from a test.

Some channels never get a click at all, like TV and radio. Here’s how to measure CTV, TV and podcast ads without one.

Should we switch from MTA to incrementality testing?

Usually, no. You add tests. You don't retire MTA.

The business case is the decision in front of you. Do your platform totals add up to more sales than the business made? Is a big spend change coming? Then a test is how you avoid moving that budget on the wrong read. And the more channels compete for real budget, the more it pays to combine all three.

If you're building that business case now, talk through which test should come first.

What is "unified marketing measurement"?

It's the industry's label for combining the three. The label matters less than the plumbing. Every reading should share one order record and one set of spend definitions. Otherwise MTA and an MMM aren't even arguing about the same thing. (More on why MTA and MMM disagree, and on what each can see.)

No model owns the truth. The loop is how you get closer to it.

The brand campaign that MTA could not see

A published case from Javvy Coffee illustrates the point. The company compared separate top- and bottom-of-funnel Meta campaigns using two-cell holdout experiments. Haus reports that top-of-funnel creative was more than 13 times as incremental as the lower-funnel effort, despite click-based MTA materially under-reporting its value. The business then reallocated more budget to top-of-funnel creative and reported improved CAC and MER.

That is not an argument against MTA. The direct-response ads were still producing observable click-led sales. The problem was treating that view as complete. Brand work is designed to build memory and preference, make a later promotion more effective, and sometimes create demand before a person is ready to click. A carefully designed test can show the delayed, non-click effect that a click-based system cannot.

Source: Haus's Javvy Coffee case study. Treat vendor-published results as a useful example of study design and validate the underlying evidence before generalizing its exact results.

A practical starting point for DTC brands

For a smaller brand with one primary paid channel and roughly $1–2 million in annual sales, a sound MTA foundation may take you a long way. As a business reaches roughly $3–5 million or more, adds two to five material channels, or needs to understand cross-channel effects, the case for triangulation becomes stronger.

Start with the decision. Build the most reliable observable journey you can. Use MMM to understand the portfolio. Use tests to calibrate the models. Then decide whether to spend up, spend down, pause a channel, or change strategy—with a clear view of what you know, what you infer, and what you still need to learn.

A measurement stack by business stage

Build in order. Each layer depends on the one before it, so adding another tool rarely fixes a gap lower down.

1. A trusted record. Orders and revenue from the system of record, with order IDs you can reconcile. Client-side event coverage of about 75% is normal, so reconcile against the record rather than the pixel. See the marketing data foundation.
2. Reconciled reporting. One set of definitions the team agrees on before anyone argues about channel credit.
3. Observed journeys. MTA for inspecting the click paths you can see, with its limits stated.
4. Experiments. Incrementality tests, geo-level where possible, once more than one meaningful channel competes for budget or platform conversions exceed actual orders.
5. An MMM. A portfolio view for budget decisions, calibrated with the tests above.

The measurement stack video walks through the same layers. If you would rather have help choosing the next layer, see consultant, platform, or agency.

Marketing measurement FAQs

Should a DTC brand use MTA or MMM?

Use both when the business has enough channel complexity to benefit from them. MTA is often the practical first foundation because it helps teams inspect observable click paths quickly. MMM becomes more useful when leaders need to make portfolio-level budget decisions that click data cannot answer on its own.

Is incrementality testing better than MMM?

Testing is better for the specific causal question a well-designed holdout can answer. MMM is better for a broader, ongoing view of allocation. They work best together: use experiments to calibrate the model, then use the model to identify the next question worth testing.

When does a DTC brand need incrementality testing?

Consider it once more than one meaningful channel is competing for budget, when platform conversions exceed actual orders, or when a large spending decision depends on whether a channel is adding sales rather than claiming them.

When a geo test is the next step, the choice between a real comparison market and a constructed one matters; see matched-market testing versus synthetic control before setting the design.

What is unified marketing measurement?

It's a label for using MTA, MMM and incrementality tests together. MTA reads the journeys you can see. Tests answer specific budget questions with causal evidence. An MMM gives the portfolio view, anchored by those tests. What matters most is one order record and one set of spend definitions behind all three.