When a brand ran MTA and incrementality side by side, how far apart were they?

When a brand ran MTA and incrementality side by side, how far apart were they?

The definitive case study for this channel: a single advertiser measured the same channels two ways at once — multi-touch attribution and randomized incrementality — and published the gap. The question: do these methods agree, and if not, which do you trust?

The setup. Multi-touch attribution (MTA) stitches user-level touchpoints and distributes credit algorithmically across the path. Incrementality runs randomized holdouts per channel. Several documented head-to-head comparisons (including platform meta-analyses and brand-side studies) ran both simultaneously on identical spend.

What they found. The methods diverged sharply and systematically. MTA over-credited lower-funnel, high-frequency channels — retargeting, branded search, email — relative to their measured incremental lift, often by 2x or more. It under-credited upper-funnel prospecting and video, whose incrementality experiments showed real lift the user-level path under-weighted (partly because view-level exposure is poorly tracked). Where they disagreed, the holdout was repeatedly vindicated by subsequent budget shifts.

The nuance. MTA is a correlation engine over observed paths — it can only redistribute credit among touches it sees, and it sees lower-funnel clicks best. Incrementality is the causal benchmark. The divergence isn't a tie to be split; it's MTA's structural blind spots made visible.

Bottom line for practitioners: when MTA and incrementality disagree, weight the experiment — MTA's errors are predictable (over-credits the bottom funnel it tracks well, under-credits the top funnel it tracks poorly). Use MTA for granular, real-time pacing and diagnosis; use incrementality to set the channel-level budget. Neither alone is the answer, and pretending one model is universally correct is how the 2x errors get funded.