Cross-channel attribution is one of the hardest problems in digital advertising. Every platform claims credit for the same sale, your total attributed revenue across all channels is 2-3x your actual revenue, and nobody can agree on what “incremental” means. Here is our practical approach.
The Problem with Platform Attribution
Amazon uses a 14-day click and view attribution window. Meta uses 7-day click, 1-day view by default. Google uses 30-day click. When a customer sees your Meta ad, searches on Google, and buys on Amazon — all three platforms count the sale. Your spreadsheet shows $300K in attributed revenue but your bank account shows $120K in actual revenue.
Our Framework: Contribution Modeling
Instead of trying to find the “true” attribution, we use a contribution model that asks: what happens when we turn each channel up or down? This is essentially a simplified version of media mix modeling that any brand can implement.
The process: establish a baseline period (4 weeks of stable spend). Then systematically test each channel by increasing or decreasing spend by 20-30% for 2-week periods while holding other channels constant. Measure the impact on total revenue (not just attributed). This gives you marginal ROAS — the true incremental value of each dollar spent on each channel.
What We Typically Find
Across our client base, here are common patterns: Amazon Sponsored Products typically has the highest direct ROAS but limited scale. Meta prospecting has the lowest direct ROAS but the highest impact on overall brand search volume and Amazon organic sales. Google branded search has inflated ROAS (these customers would likely have found you anyway). DSP retargeting shows high platform-attributed ROAS but often just accelerates purchases that would have happened organically within days.
Practical Takeaway
Allocate budget based on marginal ROAS (what each additional dollar actually produces), not platform-reported ROAS. For most e-commerce brands, this means investing more in upper-funnel channels (Meta, DSP prospecting) and less in lower-funnel (branded search, retargeting) than platform metrics alone would suggest.