The case for incrementality: Making better decisions about growth investments

In modern performance marketing, growth leaders are surrounded by an endless stream of metrics. Marketing teams collect vast amounts of data, from platform reporting and Mobile Measurement Partners (MMPs) to internal dashboards that glow with green indicators signaling high performance, or orange and red signaling poor performance. 

Yet, a persistent habit remains: defaulting to last-click attribution or gut confidence when deciding what channels to scale, which campaigns to pause, or where to optimize. This typically paints a picture of steady, predictable ROAS or CAC, implying a simple transactional formula of input and output. 

Except reality rarely cooperates. Many businesses find themselves caught in a frustrating paradox: their marketing reports show record-breaking acquisition efficiency, yet top-line revenue, net profit, and customer lifetime value refuse to scale at the same pace.

The culprit is a quiet reliance on correlation over causation. Traditional attribution tells you who converted after seeing an ad, but it rarely answers the foundational question that should govern every growth decision: Would this conversion have happened anyway without the ad?

View-through and click-based attribution have become increasingly misleading across paid channels. While they offer directional clues, they routinely inflate impact and dodge the only metric that truly matters: what spend is truly incremental. 

To build a resilient, scalable growth engine, businesses must stop guessing and build their foundation on causal experimentation.

The limits of traditional attribution

For over a decade, digital marketing has relied heavily on attribution models such as last-click, first-click, and multi-touch attribution (MTA). But attribution models are designed to assign credit, not establish causation. They can tell you which channels a customer interacted with before converting, and then distribute credit using rules that are, at best, educated guesses. 

What they cannot tell you is what would have occurred in a world where you never ran that campaign at all. That gap between correlation and causation is where significant marketing inefficiencies can emerge.

Attribution can create a compelling sense of precision while obscuring several fundamental problems:

They measure correlation, not causation

If a user clicks a brand search ad immediately before purchasing, last-click attribution may assign 100% of the credit to that ad. But the customer may have already been influenced by organic search, word of mouth, prior brand awareness, or other channels. The ad may have captured existing demand rather than created a new conversion. 

For instance, a geo-based Google Brand Search incrementality test proved minimal channel incremental lift. This gave the causal evidence to reduce brand search spend while overall customer growth continued. Organic acquisition was able to capture demand that had previously been attributed to paid search. 

In effect, we had been paying to intercept demand the business already owned, a form of organic cannibalization that traditional attribution cannot reliably distinguish from genuine incremental acquisition. 

Platform bias

Advertising walled gardens, particularly major social and search platforms, use their own attribution methodologies to measure performance. Each platform can therefore assign credit to itself for conversions that may also have been influenced by other channels. Looking at platform-reported performance in isolation can create an inflated view of overall marketing impact and make cross-channel budget allocation difficult.

Privacy and signal loss

The deprecation of third-party cookies, Apple's App Tracking Transparency (ATT) framework, and broader privacy changes have made deterministic user-level tracking increasingly limited. As the underlying signals become less complete, relying more heavily on attribution models built from incomplete data can introduce additional uncertainty into strategic decisions.

When growth decisions are built primarily on attributed performance, businesses can end up over-investing in channels that capture existing demand, while under-investing in channels that create new demand. The result is a measurement system that can tell you where a conversion was attributed, but not necessarily what caused it.

The causal approach to growth measurement

Incrementality. It closes the gap between what a marketing channel is credited with and what it actually caused. It makes us ask the fundamental question: “What happened after we spent this money?”, instead of “What happened because we spent this money that would not have happened otherwise?”

That distinction changes how growth teams evaluate performance. Attribution follows the customer journey and assigns credit to the channels involved. Incrementality creates a counterfactual: what would have happened if the marketing intervention had not occurred? The difference between those two outcomes provides an estimate of the campaign's incremental impact.

One of the most established ways to answer this question is through a geo-based holdout experiment (matched-market testing):

• Test and control markets: Identify geographic markets with sufficiently similar historical performance and relevant characteristics. The test markets receive the marketing intervention; the control markets provide a baseline against which to measure what would have happened without it.

• The treatment: Run the campaign in the test markets while maintaining normal conditions or reducing/turning off spend in the control markets, depending on the experiment design.

• Measure the difference: Compare outcomes between the two groups over a defined period. The difference provides an estimate of the incremental conversions or revenue generated by the marketing activity.

The power of this approach is that it moves measurement away from who received credit and toward what changed because of the investment.

By using controlled experimentation alongside traditional attribution, companies can develop a more reliable view of marketing impact and make budget decisions based on causal evidence rather than platform-reported performance alone.

How incrementality changes growth decisions

Incrementality changes more than how performance is measured. It changes how growth teams allocate resources, evaluate scale, and build their operating models.

1. It improves budget allocation 

Incrementality gives growth teams a stronger basis for deciding where additional investment is likely to create value. By running tests across channels such as paid search, paid social, influencer marketing, and connected TV, teams can estimate incremental ROAS or CAC (iROAS/iCAC) and compare channels based on their actual contribution to net-new growth. 

This also helps distinguish between channels that create demand and those that primarily capture it. Rather than simply scaling the channels with the strongest reported performance, teams can increase investment where causal evidence supports additional spend and pull back where attributed conversions show limited incremental impact.

2. Confidence to scale

It’s not only a tool for finding waste. It can also provide the evidence needed to invest more aggressively when conventional metrics create uncertainty. A power-calibrated Google non-brand Conversion Lift Study using a 50% exposed versus 50% control split saw +55% conversion lift and 61% incrementality, which changed the playbook. 

Rather than pulling back because of conservative CAC guardrails, the evidence gave the confidence to relax those constraints and scale a channel that was demonstrably generating net-new customers.

3. It makes scale more disciplined

Incrementality is not static. A channel that is highly incremental at one level of spend may become less incremental as investment increases. As audiences become saturated, the marginal return from additional spend can decline. 

Regular experimentation helps growth teams identify these diminishing returns and understand where additional investment stops producing sufficient incremental value. This creates a more disciplined approach to scaling, one based on observed response rather than fixed CAC thresholds or platform-reported efficiency.

4. Changes how teams operate

An incrementality-first approach does not mean abandoning attribution. Platform attribution remains useful for day-to-day decisions such as bidding, creative, and targeting. But larger budget decisions should increasingly be informed by causal evidence from experiments and market-level analysis.

That requires treating experimentation as an ongoing operating discipline rather than an occasional analytics project. Teams can establish a testing roadmap across their most important channels and investment decisions, using geo-holdouts, spend variation tests, and channel-specific lift studies to continuously improve their understanding of what is actually driving growth.

Over time, results from these experiments can also be used to calibrate everyday attribution models, making operational reporting more useful while keeping causal measurement at the center of major investment decisions.

Choosing the right experiments

With limited time, data, and analytical resources, no organization can test everything. The challenge is not simply deciding whether to experiment, but deciding which questions are worth spending scarce experimentation capacity to answer.

Three considerations can help prioritize:

1. Impact: Will the answer change a meaningful decision?

Prioritize tests where the outcome could materially change how the business allocates resources, sets strategy, or manages growth. A test that could uncover a significant efficiency gain, meaningful incremental growth, or change a major investment decision is more valuable than one designed to move a metric by a marginal amount.

2. Conviction: How strong is the existing signal? 

Not every hypothesis deserves the same level of urgency. If existing data suggests that a channel is being over-credited, a growth opportunity is being missed, or a significant assumption may be wrong, that hypothesis should move higher on the testing roadmap.

Strong signals combined with meaningful business stakes create a compelling reason to test. The purpose of experimentation is not to validate what we already believe, but to resolve uncertainty where the answer could materially change a decision.

3. Speed to learning: How quickly can we act? 

All else being equal, prioritize experiments that can produce a reliable, actionable answer within a timeframe that matches the decision at hand. 

A two-to-four-week test that provides enough evidence to change an immediate budget decision may be more valuable than a theoretically more sophisticated experiment that takes months to complete. The goal is not to make every experiment perfect; it is to generate sufficiently rigorous evidence at the speed the business needs to make better decisions.

The underlying principle is simple: experimentation should exist to change decisions. If the outcome of a test would not change what you do next, it may not be the best use of limited testing capacity.

How AI is changing the economics of incrementality testing

For a long time, one of the biggest barriers to incrementality testing was not a lack of willingness, but the operational friction involved. Designing geo-experiments, identifying comparable control markets, cleaning data, and analyzing results could require significant data science resources and weeks or months of work.

AI and machine learning are beginning to reduce that friction, making experimentation more scalable and easier to integrate into ongoing decision-making.

Faster test design and market matching

Machine learning can analyze historical sales, customer, geographic, and media data to identify comparable markets for geo-based experiments. This can reduce the manual work involved in designing tests and help teams get experiments into the field faster.

More responsive measurement models

Machine learning can also make models such as Marketing Mix Modeling (MMM) more adaptive. As new experiment results and performance data become available, models can incorporate additional evidence and help teams refine their understanding of channel contribution over time.

Earlier detection of anomalies

External factors such as competitor activity, economic shifts, or unusual market conditions can affect experiments. AI can continuously monitor test data for unexpected patterns and flag potential anomalies, allowing analysts to investigate before results are used to make major investment decisions.

The broader opportunity is not for AI to replace experimentation, but to make rigorous experimentation more accessible and repeatable. When AI reduces the analytical and operational cost of running tests, organizations can move from treating incrementality as an occasional measurement exercise to using it as a more continuous input into growth and resource allocation decisions.

Conclusion

Growth investment decisions are too important to rely solely on attribution models and platform-reported metrics. As margins tighten, media costs rise, and privacy changes reshape measurement, businesses need to know not just what happened, but what their investments actually caused.

That question extends well beyond marketing. Whether the decision involves a new product, additional headcount, a strategic shift, or an AI initiative, the underlying principle is the same: would this outcome have happened without the investment?

Incrementality provides a practical way to answer that question. By bringing causal evidence into resource allocation, organizations can distinguish between activity that looks effective in reporting and investment that actually creates additional value.

The goal isn't perfect measurement. It's making better decisions with evidence that gets closer to causality.