How should a DTC brand measure incrementality when every channel claims the same sale?
When email, paid social, and affiliates all claim the same order, the reporting is not wrong, it is answering the wrong question. Incrementality asks what would have happened without the channel, and only experiments answer that honestly.
The credit pile-up is a feature of the measurement, not a bug
Open any multi-touch dashboard and add up the revenue each channel claims. The total will exceed actual revenue, sometimes by multiples. This is not a data error. Each channel's reporting answers 'how many converting customers touched this channel,' and a customer who touched three channels gets counted three times. The pile-up is the honest output of touch-based measurement.
The problem is that brands then use these numbers comparatively: email drove X, paid social drove Y, therefore shift budget from one to the other. That comparison is invalid the moment the same sale appears in both columns. Touch-based credit tells you about customer journeys. It does not tell you what caused the sale, and budget decisions need causation.
This is the question incrementality answers: of the sales a channel touched, how many would not have happened without it. Everything else is journey description wearing a causation costume.
Start with the experiments you can actually run
Full geo-experiments and matched-market tests are the gold standard and also the thing most DTC brands will never run properly. Start smaller. The most accessible incrementality test is the holdout: suppress a channel for a matched audience segment and measure the difference in total revenue, not channel-attributed revenue.
Email holdouts are the easiest starting point. Hold back ten percent of the list from a campaign series and compare total purchase behavior between the held-out group and the mailed group over a full attribution window. The difference, divided by the cost, is the closest thing to true email ROI most brands will ever see. It is routinely lower than the platform's reported number, which is exactly why the test is worth running.
Paid social holdouts are harder because you cannot hold out individuals cleanly, but geo-holdouts work: pause spend in a set of regions and compare sales trends against control regions. The methodology matters less than the discipline of measuring total business impact instead of platform-reported conversions. Any experiment that compares totals beats any model that allocates touches.
Read the results without flinching
Incrementality tests almost always deflate someone's favorite channel. Branded search, retargeting, and email to engaged segments routinely show low incrementality: they harvest demand that already existed. Prospecting, influencers, and top-of-funnel video routinely show higher incrementality than their last-click numbers suggest. The pattern is consistent enough to plan around.
The correct response is not to kill the low-incrementality channels. Retargeting with low incrementality may still be cheap insurance, and email to existing customers has retention value beyond the test window. The response is to reprice them: budget low-incrementality channels on efficiency and high-incrementality channels on growth, instead of ranking everything by the same blended ROAS.
Run the tests on a cadence, not as a one-off. Incrementality drifts as the brand grows, as creative fatigues, and as the media mix changes. A holdout result from Q1 is stale by Q3. The brands that measure well treat incrementality testing as a recurring operating cost, like creative production, not as a special project.
What to do when you cannot run experiments
Some channels resist holdouts: affiliates with contractual minimums, marketplaces you do not control, PR you cannot pause. For these, use triangulation. Compare the channel's claimed credit against three independent signals: the timing of spend changes versus total revenue changes, the channel's share of new-customer acquisition specifically, and survey-based self-reported attribution from post-purchase surveys.
No single leg of the triangulation is reliable alone. Post-purchase surveys are biased toward memorable channels. New-customer share ignores assist value. Spend-timing analysis confounds with seasonality. But when all three point the same direction, you have a working answer, and a working answer beats a precise fiction.
The meta-lesson is about humility in measurement. Every attribution number your brand reports should carry an implicit confidence interval, and incrementality is how you calibrate it. Channels do not claim the same sale because they are lying. They claim it because they each contributed something and the measurement cannot split the credit. Admitting that, and testing for the causal slice, is what separates brands that scale efficiently from brands that scale expensively.