This distinction matters whenever a person discovers a product in one place and completes the transaction somewhere else. Short-form video may create demand. Branded search may capture it. A marketplace may close the transaction.
Last-touch reporting can reward the final measurable interaction while the demand-generating channel appears less effective than it really was. A July 2026 preprint describes one version of this failure as an assisted own goal: advertising causes an incremental purchase, but the conversion is booked on a downstream platform. The paper is a new methodological proposal—not settled industry consensus—but the decision problem is practical.
First check: did the business result fall?
Which measurable interaction received credit?
What would not have happened without the activity?
Attribution is useful for fast operational monitoring. It gives teams granular campaign, creative and user-path signals. But attributed conversions do not establish a counterfactual. They do not tell you what would have happened if the campaign had not run.
Incrementality attempts to estimate that missing outcome. A randomized or quasi-experimental design compares observed results with a credible estimate of the outcome without the intervention.
A simple assisted-own-goal example
Short-form video
Branded search
App store
Depending on the rules, the app store, branded search or an MMP touchpoint may receive the conversion. The video campaign may receive partial credit or none. If the team pauses video because its attributed ROAS is below target, branded search efficiency may deteriorate later because less demand is being created upstream.
More attribution detail does not solve causality
Multi-touch attribution can improve path visibility, but it still assigns weights to observed interactions. Those paths are affected by targeting, prior intent, platform optimization and missing data. Users exposed to a campaign were usually not selected at random.
- High-intent users are easier to target.
- They are also more likely to convert without the ad.
- A campaign can therefore look effective while capturing demand that already existed.
A decision hierarchy before cutting spend
- 01
Validate the measurement chain
Check links, SDK and consent changes, conversion windows, missing revenue events and web-to-app breaks.
- 02
Separate attribution from business outcomes
Compare platform and MMP results with backend purchases, subscriptions, margin or retained revenue.
- 03
Test whether the channel creates demand
Measure uplift with the strongest feasible split: users, geos or repeated campaign-on/off periods.
Five questions before cutting a channel
- Did the business outcome fall, or only the attributed outcome?
- Did another channel, marketplace or branded search capture more of the same demand?
- Did tracking, consent, SDK or conversion routing change?
- What experiment or natural variation supports the causal conclusion?
- What result would cause the team to reverse the budget decision?
If these questions are unresolved, the correct state is not “the channel failed.” It is decision pending: measurement gap.
The rule before cutting spend
An attributed decline proves that credit changed. A backend decline proves that the business result changed. Neither one, by itself, proves that this channel caused the loss. Stabilize the measurement chain, check where the conversions moved and test demand creation before removing budget from the channel.
Should you cut this channel?
Separate a business loss from a change in attribution credit.
If the team is deciding what measurement capability it needs next, run the measurement readiness review. When the immediate trigger is a CPI spike, rank the CPM, CTR and CTI contributions before debating channel causality.
Sources and limits
- Media Measurement and the Assisted Own Goal, submitted July 10, 2026. Recent preprint; not universal empirical proof.
- AppsFlyer Incrementality for UA Guide, checked August 1, 2026. Vendor documentation for a geo-experimentation workflow.