The revenue shown in a marketing channel’s report is not necessarily the revenue that channel created. A platform may claim credit for a purchase made by a customer who interacted with an ad. But if that customer would have purchased anyway, the spending may have created no additional economic value.
This distinction is easily lost in budget meetings. The channel reporting strong returns is protected; the channel absent from the final step of the purchase journey is cut. A company can therefore reduce the investment that generates demand while increasing the spending that captures existing demand. Its reports may improve while its capacity to grow weakens.
In The Lie of the Last Click and ROAS Lies, we examined the limitations of these metrics. The question here is how to allocate budgets with those limitations in mind.
The Attribution–Incrementality Matrix offers a starting point. We assess each investment through two questions: How much revenue does the platform claim? How much revenue would we actually lose if the spending stopped?
The first answer may appear in a report. The second requires a comparable control condition. Appearing alongside a sale is not the same as causing it. The economic justification for the budget lies in that difference.
1. Four quadrants: A budget map, not yet a verdict
Attribution is the revenue a platform claims under its attribution rules. Incrementality is the revenue that would not have occurred without the spending.
Platform attribution
Low incrementality
High incrementality
High
Shadow: May be capturing existing demand. A priority candidate for a spending reduction test.
Engine: Creates demand and receives credit for it. Scale once profitability is verified.
Low
Waste: Discontinue once adequate measurement confirms the finding.
Seed: Creates demand without receiving sufficient credit. Verify and protect its contribution.
Branded search and broad retargeting are common candidates for the Shadow quadrant. Video, creator partnerships and PR may be Seed candidates. These placements are hypotheses, not test results.
A video investment does not qualify as Seed simply because it receives little last-click credit. Its contribution to demand must be demonstrated. Equally, branded search capturing existing customers does not, by itself, prove that it is unnecessary.
Placement should be considered at the campaign, customer segment and spending level before applying it to an entire channel. Different parts of the same channel can occupy different quadrants.
2. What eBay showed: Channel averages can hide where the budget goes
In 2012, eBay stopped advertising on search terms containing its own brand name across Yahoo and Microsoft’s search network. Almost all the lost paid-click traffic shifted to free organic results. Substitution between the paid and organic routes was nearly complete. The study by Blake, Nosko and Tadelis was published in Econometrica in 2015.
For non-brand keywords, advertising influenced new and infrequent users. Yet most spending went towards frequent users whose purchasing behaviour was unaffected by the ads.
Through our framework, this illustrates how the same channel can contain both Engine and Shadow spending. The conclusion is not simply to stop search advertising. The relevant question is whose behaviour the budget changes.
The management implication is clear: Evidence that a channel works does not establish that all spending within that channel works. A mechanism that influences new customers does not automatically justify the budget directed at existing customers.
3. Turning a hypothesis into causal evidence
Three methods can help test a channel’s placement:
Controlled shutdown: Stop spending for a selected campaign or audience, then compare total sales against a comparable control group where advertising continues. A simple before-and-after comparison can mistake seasonality for an advertising effect.
Geographic holdout: Divide regions with similar historical behaviour into treatment and control conditions, using random assignment where possible. Change spending in one group while maintaining it in the other. Regional differences and advertising exposure spilling between groups must be addressed in the design.
Platform lift tests: Measure the difference in conversions between groups eligible to receive advertising and groups held back from it. This is an experimental approach, distinct from a standard attribution report. At Sellf, we do not consider platform-run measurement sufficient on its own without validation against independent sales data and a review of the methodology.
For example, switching off retargeting may reduce the sales reported in the advertising dashboard. What matters for the decision is how much the company’s total sales change relative to the control group.
Before testing begins, the measurement period, sales cycle, decision threshold and acceptable uncertainty are defined. Finding no difference in a short or underpowered test does not prove there is no effect. Delayed effects require particular attention for Seed candidates.
4. Allocating budgets by incremental contribution
Identifying incremental sales is the beginning of the economic assessment. Our decision rule is:
Net incremental contribution = Incremental net revenue × contribution margin rate − relevant spending
Contribution margin is what remains after product costs, commissions, delivery and other variable expenses. Advertising spending is excluded from this margin calculation and deducted at the end of the equation. Returns and discounts must also be treated consistently.
If the result is negative, the channel is not generating profit within the period and scope assessed. A positive result still does not justify unlimited scaling: The return on the existing budget does not guarantee the return on the next increase. Additional spending can be tested separately.
This is why Sellf’s ROI- and RGI-focused approach follows the sequence of hypothesis, testing and scaling. Decision-Grade Data determines which budgets to protect and which to release.
Separating the reporting team from the measurement team is part of that approach. The assessment made by the team deploying the budget is tested through independent financial or analytical review.
Three questions provide a useful starting point for the management table:
- What percentage of the budget may sit in the Shadow quadrant?
- Which channels have undergone a shutdown or holdout test in the past 12 months?
- Are the reporting and measurement teams the same?
We view marketing budgets as capital allocation decisions. When capital remains expensive, reducing spending that testing has confirmed as Shadow can release cash without sacrificing sales.
At your next budget meeting, which investment will you test for the contribution it actually creates?
