Incrementality testing has become the default fix for a real problem: platform attribution disagrees with itself constantly, with Meta and Google routinely claiming credit for the same conversion. But treating a single lift study as the final word on whether a channel deserves budget has pushed growth-stage brands into a specific, repeatable mistake — cutting an upper-funnel channel that fails a standalone test, then watching total revenue drop because that channel was doing work no isolated test could see.

The fix isn’t to abandon incrementality. It’s to stop asking it to answer a question it was never built to answer, and to anchor budget decisions in a metric that actually reflects how the business performs as a whole.

Why a single lift study can’t settle the allocation question

Incrementality measures the causal impact of one channel or campaign. That’s useful, but it isn’t the same thing as understanding how marketing contributes to the business overall.

Take a customer who sees a Meta ad on Monday, doesn’t click, then searches for the brand on Wednesday and converts through a paid brand search ad. Meta logs a view-through. Google logs a last-click conversion. Run a lift study on either channel in isolation and it may show only a modest incremental contribution — but both ads did real work, just different work. The Meta impression built consideration; the branded search closed the sale. Cut either one and the journey breaks.

This is exactly how brands talk themselves into bad reallocations. A team sees Meta’s lift study come in low, concludes the channel is taking credit for conversions that would have happened anyway, and shifts the budget elsewhere. Six weeks later, brand search volume drops, blended efficiency drops with it, and nobody can explain why. One lift study on one channel only tells you what happened inside that test — it can’t tell you whether the channel deserves the budget.

The metric most teams are missing: MER

Marketing Efficiency Ratio — total revenue divided by total ad spend — is the one commonly available metric that doesn’t care which channel gets the credit. It treats marketing as a single investment producing a single revenue stream, which is exactly what marketing is at the business level, and it’s the question CFOs and founders are actually asking when they review performance.

MER by itself isn’t sufficient. It can’t tell you how to allocate within a budget, and it can be inflated by seasonality or organic demand growth unrelated to paid spend. But it answers the question that should anchor every other measurement decision: is blended marketing investment producing acceptable returns at the business level? Once that anchor is in place, every other layer of measurement gets easier to interpret correctly.

Build a three-layer measurement stack, not one metric

A working measurement stack has three layers, and each one answers a different question:

• MER answers: Is total marketing spend producing the returns this business needs?
• Incrementality answers: If I add or cut spend on this channel, what happens to MER?
• Attribution answers: What touchpoints did customers actually engage with, and what does that tell me about each channel’s role in the journey?

Most measurement mistakes come from using one layer to answer a question that belongs to another. Cutting Meta because brand search closed the sale is reading attribution as causation. Trusting Meta’s self-reported ROAS at face value makes the same error in the other direction. And treating an isolated lift study as a final verdict ignores what that channel might be contributing to MER indirectly, through its effect on other channels — a pattern that shows up constantly when brands try to size upper-funnel budget allocation using bottom-funnel logic.

Four ways to actually run incrementality testing, ranked by accessibility

Running proper incrementality tests used to be expensive enough to put off entirely. That’s changed meaningfully through 2025, and the options now span a real range of cost and rigor.

Platform-native lift studies

Meta Conversion Lift and Google Conversion Lift run inside the existing ad platforms at no additional cost. Per Google’s official documentation, Conversion Lift now reports directional results for studies with budgets above $5,000 and 1,000 conversions, thanks to a shift to Bayesian statistical methodology that supports lower budgets and fewer conversions than the older frequentist approach required — a change Google’s own 2025 product recap confirms.

Meta’s Brand Lift studies sit at the other end of the spectrum: per Meta’s minimum requirements documentation, Brand Lift in the United States now requires a $120,000 minimum budget over the study duration, up from $30,000 — putting it out of reach for many brands. Meta’s Conversion Lift studies carry much lower thresholds and remain a viable starting point; the two products measure genuinely different things at very different costs. Platform-native tests also have a hard ceiling: they only measure incrementality inside the platform running the test, so they can’t capture cross-channel effects. Treat the result as one input, not a verdict.

Geo holdout testing

If sales are distributed across enough markets to run a real holdout, geo testing produces cleaner results than user-level lift studies. Pause spend in matched markets while continuing it in others, then measure the revenue gap. Test and control markets need to be matched on baseline performance, seasonality, and customer demographics, with several weeks of pre-test baseline data confirming they behave similarly under normal conditions.

Spend-down testing

This is the most direct way to measure a channel’s effect on MER specifically. Cut a channel’s budget by 50 to 75% for a defined window and measure total business impact, not channel-level metrics. Cut Meta by half and total revenue drops 40%? That channel was contributing more than its lift study suggested. Cut it by half and revenue holds? The channel was likely harvesting demand other channels were already creating. Spend-down testing is less statistically rigorous than a properly structured geo holdout, but it’s the only method that explicitly measures a channel’s contribution to MER rather than to its own attributed revenue.

Full causal inference models

Synthetic controls, difference-in-differences analysis, and test-calibrated media mix modeling sit at the top of the stack. Google’s open-source Meridian MMM, released in 2025, brought Bayesian causal inference modeling to advertisers without requiring a proprietary vendor relationship, but it still demands real data science capability to implement well. Most brands don’t need to operate at this layer — the first three methods answer the budget questions that matter day-to-day.

A testing cadence that actually builds signal

For a brand spending $100,000 to $1 million monthly across paid channels, a practical cadence looks like this:

• Weekly MER review at the blended level, broken down by new versus returning customer where possible.
• Quarterly incrementality test on the largest channel by spend, structured as a geo holdout where feasible.
• Annual full-channel holdout on each major channel to refresh baseline contribution assumptions.
• Continuous platform-native lift studies on new campaigns and significant creative refreshes.
• Spend-down tests whenever MER moves materially without an obvious explanation.

Brands that build this rhythm end up with a defensible view of channel sensitivity no platform dashboard can give them, and pairing it with a steady MER read sharpens every allocation conversation that follows.

Reading results without overcorrecting

The hardest part of incrementality testing is interpretation. A low lift study on Meta does not automatically mean Meta should be cut — it means the channel isn’t creating standalone incremental volume during the test window, which is a different claim from whether it’s moving MER through its effect on brand search, direct traffic, or returning customers.

Read the lift study alongside MER, not instead of it. If lift comes in low and MER holds steady after you reduce spend, the channel may genuinely be replaceable. If lift comes in low but MER drops when you pull back, the channel is doing work the test simply couldn’t measure.

Most brands already have the pieces — they’re just not stacked

Most growth-stage brands have all three layers available and aren’t using them together. They watch platform ROAS, occasionally glance at a lift study, and treat MER as a number that lives in a finance report rather than a measurement decision. The incrementality conversation has spent two years arguing about whether attribution is broken. That’s the wrong argument: attribution describes the journey, incrementality measures sensitivity, and MER is the number the business actually runs on. Brands that build all three into one decision system will allocate paid media budget with far more confidence than teams still debating which platform’s number to trust.

Frequently asked questions

What is Marketing Efficiency Ratio (MER)?

MER is total revenue divided by total ad spend. It measures blended marketing performance at the business level without assigning credit to any single channel, which makes it resistant to the attribution disputes that plague platform-level reporting.

Why isn’t incrementality testing enough on its own?

A single lift study only measures what happened inside that specific test, usually within one platform. It can’t capture how a channel influences demand for other channels, so acting on it in isolation can lead to cutting budget from a channel that was actually supporting overall revenue.

What’s the cheapest way to start incrementality testing?

Platform-native lift studies, like Google Conversion Lift, are free to run inside the ad platform and now work at lower budget and conversion thresholds than in previous years, making them the most accessible starting point.

How often should brands run incrementality tests?

A practical baseline is a quarterly test on the largest channel by spend, an annual full-channel holdout, continuous platform-native tests on new campaigns, and spend-down tests whenever MER shifts without a clear explanation.