Incremental Attribution

Your Meta ROAS drops the morning after you flip one campaign setting, and nothing else changed. Fewer conversions, a worse-looking number, the same ads running.

The setting told Meta to stop counting sales it thinks would have happened without the ad. That setting is incremental attribution. Standard attribution credits every conversion inside your window; incremental attribution runs a counterfactual model and keeps only the ones it estimates the ad actually caused.

The number Meta leads with

The setting drives a 24% increase in incremental conversions against standard attribution, and reached a multi-billion-dollar annual run-rate seven months after launch — Meta’s own figures, from its January 2026 performance update.

That is the case for turning it on: the same spend, credited to the sales the ad caused rather than the sales it merely witnessed.

Haus tested the same setting and got 43%

An independent measurement firm reads it differently. Meta’s incremental attribution beat standard settings in only 43% of tests — the win rate Haus measured across its client experiments, a rate it called “far from a home run,” on what it flagged as a limited sample.

A larger follow-up narrowed the gap. Across 640 experiments run from July 2025 through June 2026, Haus reported the setting delivered a higher incremental return than standard in the typical test, an improvement on the early 43% figure.

So the disagreement is not that one side is lying. Meta measures the feature from inside its own ecosystem and reports a clean 24% lift. An outside firm running geo-based holdouts on live accounts started at a coin-flip and climbed. The two numbers describe the same setting from opposite sides of the platform boundary.

It only sees conversions that touch Meta

The model reads conversions that hit Meta’s own tracking, and nothing else. A shopper who sees a Meta ad, then searches your brand on Google and buys there, finishes the purchase off Meta’s tracked path — so the model cannot record that sale as incremental, a blind spot WorkMagic documents.

This bites hardest on brands running both surfaces. Our Place runs 502 Meta ads, 94% of them template-driven DCO, alongside a 42-country Google presence. Any demand the Meta machine creates that resolves on Google is invisible to a Meta-only incrementality model.

There is a delivery cost underneath the measurement one. Fewer counted conversions mean less signal for Meta’s optimization engine, and a high-velocity account feels that first. Our Place launched 156 new ads in a single week; a learning phase that churns that fast now has fewer events to learn from.

The setting measures a real thing, and measures a partial slice of it. The gap between Meta’s number and Haus’s is the part still moving.