The core measurement problem

A customer might discover your brand through a social media ad, research it further by searching your brand name on Google, compare prices by checking your Amazon listing, and finally complete the purchase on whichever channel offered the better price or faster shipping that day. Which channel gets "credit" for that sale? Standard last-click or last-touch attribution would credit whichever channel the purchase actually completed on — but that systematically undercounts every channel that contributed earlier in the journey, and it especially undercounts brand-awareness activity (social ads, content, influencer posts) that rarely gets the literal last click.

This problem is structurally worse for a multi-channel seller than for a single-channel one, because marketplaces generally don't share their own conversion/attribution data with your other channels' analytics tools, and your own site's analytics can't see what happened on the marketplace side of the journey at all. You're working with two separate, incomplete pictures rather than one unified view.

Attribution asks: which touchpoint(s) should get credit for a specific sale? Incrementality asks a more fundamental question: did this marketing activity actually cause a sale that wouldn't have happened otherwise, or would the customer have purchased anyway? A sale can be correctly "attributed" to an ad by a tracking tool while still not being incremental — if the customer was already a loyal repeat buyer who would have purchased regardless of seeing that specific ad, the ad gets attribution credit without having caused anything.

The halo effect and cannibalization — two sides of the same coin

Halo effect: advertising or brand-building activity on one channel (a social campaign, influencer content) drives sales on a different channel that the attribution model never connects back to the source — someone sees a social ad, then searches your brand directly on Amazon and buys there. This makes the social campaign look less effective than it actually is if you're only measuring sales that happened on the same platform as the ad.

Cannibalization: the reverse risk — spend that appears to be driving marketplace sales might actually be pulling customers who would have bought on your (typically higher-margin) own website anyway, net-negative for the business even though the marketplace channel's own numbers look like a win.

Both distortions point to the same underlying lesson: channel-level numbers reviewed in isolation, without any cross-channel view, can lead to systematically wrong conclusions about what's actually working.

Practical approaches, roughly in order of sophistication

Brand-search lift as a rough proxy. Track whether branded search volume (on Google or within a marketplace's own search bar) rises during and after an awareness campaign — sustained lift is reasonable evidence of a halo effect even without perfect attribution.

Geographic or timing holdout tests. Pause a specific marketing activity in a subset of regions (or for a defined time window) while keeping it running elsewhere, and compare sales trends across channels between the held-out and active groups. This is a more rigorous way to estimate true incrementality than attribution modeling alone, since it creates something closer to a controlled comparison.

Coordinated promotional calendars. Avoid running a major push on one channel and a competing promotion on another simultaneously if your goal is to cleanly measure either one's individual impact — simultaneous activity across channels makes it much harder to separate which one actually drove any change.

Unified reporting, even if imperfect. Build a single view that puts owned-site and marketplace revenue, and their respective marketing spend, side by side over the same time periods — even without solving attribution perfectly, seeing both together at least surfaces an obvious cannibalization or halo pattern that reviewing each channel's dashboard separately would miss entirely.

A worked example

A seller runs a paid social campaign aimed at their own website. Own-site revenue during the campaign period is flat, which looks like a failed campaign by pure last-click attribution on the website's own analytics. But branded search volume within their top marketplace rose noticeably during the same window, and marketplace sales for the same core product also rose. Read in isolation, each channel's data suggests something different (website: campaign failed; marketplace: unrelated organic growth) — read together, the more likely explanation is that the campaign drove awareness that converted on the marketplace instead of the website, a genuine halo effect that neither channel's dashboard alone would reveal.

Best practices

  • Build at least a basic side-by-side view of owned and marketplace revenue and spend before drawing conclusions from either channel's dashboard alone.
  • Track branded search volume as a low-effort proxy for cross-channel halo effects.
  • Avoid running simultaneous, uncoordinated promotions across channels when you need to isolate either one's individual impact.
  • Treat "this channel's numbers went up" and "this channel caused incremental growth for the business" as two different claims requiring different evidence.

Checklist

  • Build a combined revenue-and-spend view across owned site and each marketplace, same time periods
  • Track branded search volume as an ongoing proxy metric
  • Before running a major single-channel campaign, check whether a competing promotion is scheduled elsewhere that would confound measurement
  • Consider a geographic or timing holdout test for a significant, ongoing spend commitment you haven't yet validated as incremental

FAQs

Is perfect cross-channel attribution achievable for a typical seller? No — marketplaces generally don't expose the data needed for true unified attribution, so the realistic goal is a reasonable directional understanding (via proxies and holdout tests), not a precise single number.

Should I stop advertising on a channel if I suspect its sales are mostly cannibalizing another channel? Not automatically — first estimate the margin difference between the channels involved; some cannibalization from a lower-margin channel into a higher-margin one is a net negative worth addressing, but cannibalization between two similarly profitable channels may not be worth the effort to fully resolve.

How does this connect to cohort analysis? Cohort analysis (see Cohort Analysis for Ecommerce Retention) is often layered on top of attribution work — tracking a cohort's cumulative revenue across all channels over time, not just its first-touch channel, gives a fuller picture of a customer's total value than single-channel attribution alone.