Incrementality: the one marketing question your dashboards can't answer

Every metric in your ad account answers a version of "what happened?" Incrementality asks the harder question: "what happened because of us — and what would have happened anyway?" It's the most valuable question in marketing measurement, and the one your platforms are structurally worst at answering, because answering it honestly would usually make their numbers look smaller.

What incrementality actually means

Incrementality is the share of your results that your marketing actually caused — the sales that would not have happened without it. The opposite of incremental is a sale you'd have made regardless: the loyal customer who was going to reorder, who clicked your retargeting ad on the way to checkout and let it take the credit.

Here's the uncomfortable version. Retargeting ads famously post huge returns. But a large portion of the people they "convert" were already heading to buy — the ad just got in front of a decision that was already made. That doesn't mean retargeting is worthless. It means its reported ROAS and its incremental ROAS can be very different numbers, and only one of them should drive budget.

Why ROAS and incrementality aren't the same

ROAS tells you the ratio of revenue to spend. It counts every sale the channel touched. Incrementality tells you how many of those sales the channel added — how many wouldn't exist otherwise. A channel can have a 5x ROAS and a 1.2x incremental return, because most of what it's credited with was going to happen anyway.

This is the gap that quietly wastes budget. Teams scale the channel with the best ROAS, when the channel with the best incrementality — the one genuinely creating new demand — often looks worse on the dashboard precisely because it works upstream, early, before the last click that gets the credit.

How incrementality is really measured

The honest answer: the gold standard is an experiment, not a report. You take a comparable group of people, deliberately withhold the marketing from some of them (a holdout, or control group), show it to the rest, and compare. The difference between the two groups — the lift the exposed group shows that the withheld group doesn't — is the incremental effect. That's a real, causal answer, and it's the only way to be truly certain.

It's also genuinely hard to run well. You need enough volume, a clean control group that isn't contaminated, and the discipline to not market to a segment on purpose — which many teams are unwilling to do. That's why true holdout testing is rarer than the word "incrementality" being thrown around would suggest. When you can run one, it's the strongest evidence there is.

What you can learn without a full experiment

Short of a randomized holdout, there's a spectrum of increasingly strong evidence you can build from your own data — and this is where most practical incrementality work actually lives:

• Observation: compare journeys that included a channel against journeys that didn't. It's associational, not proof — the two groups differ in more ways than just the channel — but it's a real signal and a starting point.
• Correlation with matched audiences: tighten the comparison to more similar groups, so you're comparing like with like as closely as your data allows. Stronger than raw observation, still short of a controlled experiment.
• Cross-channel impact: look at how channels move together — whether a channel tends to appear before conversions on other channels (suggesting it lifts them), or performs on its own. This is how you catch the upstream channel that's quietly making everything else work.

None of these are a substitute for a true holdout, and it's worth being honest about that distinction — anyone claiming certainty without a controlled test is overselling. But layered together, and read against real, deduplicated revenue rather than platform-inflated numbers, they get you a far more trustworthy picture than a raw ROAS column ever will.

The prerequisite nobody mentions

Every method above — holdout included — depends on one thing being true first: you have to be able to see the whole journey, and count each person once. If your data counts the same customer as three people across three devices, your "with channel vs. without channel" comparison is already broken before you start, because you can't reliably tell which group anyone belongs to.

This is why incrementality and identity are the same project. You cannot measure what a channel added if you can't tell who was exposed to it. Resolving the customer to a single identity across every touch is the precondition for any honest incrementality work — not a nice-to-have alongside it.

How to think about it

• Separate the two numbers: reported ROAS is what a channel touched; incremental return is what it added. Budget belongs to the second.
• Be suspicious of your best-looking channel: high reported returns often mean it's harvesting existing demand, not creating it.
• Run a holdout when you can: it's the only true proof, and it's worth the discomfort of withholding.
• When you can't, build layered evidence: observation, matched comparison, and cross-channel patterns — read honestly as signal, not certainty.

At Chapter, incrementality is built on that identity-first foundation: because we resolve each customer to one journey across every touch, the comparisons — with a channel versus without it, matched audiences, cross-channel influence — run against a clean, deduplicated picture instead of a fragmented one. It's not a claim of certainty; it's a claim of honest, layered evidence measured against real revenue. For why your tools disagree in the first place, start with the data-discrepancies guide.