Is Meta's Incremental Attribution Outperforming Standard Attribution?

Let’s get right into it: Is Meta's Incremental Attribution outperforming standard attribution? The short answer is yes – at least for now. 

In July 2025’s The Meta Report, we published data showing Meta's standard attribution outperforming Incremental Attribution (IA). When we re-ran the analysis on more recent tests, that relationship has flipped: Across incrementality experiments run since July 2025, IA iROAS is beating standard iROAS by a pooled geo-mean of 1.26x, compared to 0.80x in the July 2024-June 2025 period we originally reported on.

What’s a geo-mean and why does it matter? Each test produces a single performance ratio, and the geo-mean averages those ratios using a method built for ratios rather than raw numbers. Each dot is one experiment: Above 1.0x means Incremental Attribution outperformed, below 1.0x means standard attribution outperformed. The geometric mean shows the typical result across experiments without letting one or two extreme outliers skew the overall takeaway.

This analysis walks through what the updated data shows, where the pattern is strongest, and – just as importantly – where we're still mindful about drawing firm conclusions.

A note on how we analyzed performance

Before we dig into the results, an important note to contextualize these findings: We are not comparing what each Meta setting – Incremental Attribution vs. standard attribution – reports.  Comparing Meta's self-reported numbers under two settings would only illustrate which one claims more credit, not which one actually drives more outcomes.

Instead, every data point comes from a controlled Haus incrementality experiment, where a randomized holdout (or, in a head-to-head test, a matched control cell) establishes the causal return. We then compare the incremental return a business actually earns when Meta is optimized toward Incremental Attribution versus toward standard attribution:

IA ÷ standard =
(incremental return when optimizing to IA) / (incremental return when optimizing to standard)

This measures both sides by the experiment, not by the platform. A ratio above 1.0x means running Incremental Attribution delivered more incremental return per dollar than running standard attribution; below 1.0x means standard delivered more.

Those experiments take a few forms. Some are head-to-head A/B tests that run both settings as equal-budget cells at the same time, so spend and audience are held constant and only the attribution setting differs. Others are holdout tests of an IA-optimized program, benchmarked against the same brand's standard-attribution incrementality tests.

As noted above, we summarize with a geometric mean (or “geo-mean”) rather than an average, because these are ratios and the geo-mean is the honest center for ratio data (one outsized result cannot distort it). We show an 80% interval next to each figure so the uncertainty is visible, and with a modest number of tests we treat the direction of the signal as more reliable than any single decimal.

The before-and-after split

Now, the part you really care about: Looking at the full dataset, tests skew in IA's favor, with a pooled geo-mean of 1.11x and a pooled range of 0.96x–1.27x. But that headline number blends two different periods. When we split the data by test window, the shift becomes clear:

• July 2024 – June 2025: Pooled geo-mean of 0.80x (range 0.58x–1.10x) – standard attribution was, on average, outperforming IA.
• July 2025 – June 2026: Pooled geo-mean of 1.26x (range 1.11x–1.43x) — IA is now outperforming standard attribution in the typical test.

That's a notable reversal, and it's the reason we're revisiting the conclusion we published last year.

DTC-only brands drive much of the recent gain

Breaking out the same data by business model shows the recent improvement isn't evenly distributed. We split tests into DTC-only brands and omnichannel brands (those with at least one non-DTC sales channel):

• Before (July 2024 – June 2025): DTC-only tests averaged 0.94x versus standard; omnichannel tests averaged 0.76x.
• Recent (July 2025 – June 2026): DTC-only tests averaged 1.38x versus standard; omnichannel tests averaged 1.02x – essentially at parity.

In other words, DTC-only brands see the larger swing toward IA outperformance, while omnichannel brands have moved from below parity to roughly even with standard attribution.

What might be driving the shift

We considered several dimensions before writing this up, and want to share what we learned:

Brand size doesn’t appear to influence results. We checked whether larger brands were systematically seeing bigger lifts from IA, and they weren't. 

The effect leans modestly toward acquisition-style outcomes over fully blended sales outcomes, but the edge is slim and sits within the noise at our current sample. We treat it as a directional signal worth watching rather than a firm finding.

The clearest pattern is about sales channel mix. As the section above shows, the recent gain is strongest among DTC-only brands (1.38x versus omnichannel's 1.02x). Both business models improved from the earlier period, but DTC-only moved further and the recent window skews far more heavily toward those brands, so part of the aggregate jump reflects the composition of who ran tests, not a uniform improvement across all advertisers.

We also checked whether this is a within-brand "maturity" effect, where a brand's IA gets better as it runs more cycles. The evidence is mixed: Among brands that re-tested, some improved and some declined, and once we control for spend, neither "IA improving over time" nor "standard attribution becoming a noisier signal" separates cleanly from noise at this sample. So we are not claiming IA measurably improves with reps.

Why this matters for brands and enterprises

Attribution comparisons like this one are inherently a moving target – the methodologies, the platforms, and the underlying signal all continue to evolve. Last year's data supported one conclusion; this year's data supports a different one. It's a reminder that a single snapshot in time isn't a permanent verdict.

For advertisers deciding how much weight to put on Incremental Attribution versus standard attribution today, the most useful takeaway may not be "IA is now better" in absolute terms, but rather that the gap has narrowed and reversed direction over the past year, that improvement shows up across business models but is strongest for DTC-only brands, and that it holds regardless of brand size.

We'll continue to track closely and update our published data and research reports as the picture evolves. And as always: Every business is different; test for yourself. Until next time.

FAQ

Does Meta's Incremental Attribution perform better than Standard Attribution?

It seems so, at least as of July 2026. A note that our data doesn't measure better performance directly, but rather the real, experiment-verified incremental return a brand earns when Meta is optimized toward Incremental Attribution versus toward standard attribution. In our July 2025 to June 2026 tests, optimizing toward IA delivered a higher incremental return than standard in the typical test (pooled geo-mean 1.26x), a reversal from the July 2024 to June 2025 period, when standard came out ahead (0.80x).

Does brand size affect whether Incremental Attribution outperforms Standard Attribution?

Based on our testing, no. We didn't find that larger brands systematically saw bigger lifts from IA relative to standard attribution.

Do DTC brands see different results than omnichannel brands?

Yes. In recent tests, DTC-only brands averaged 1.38x (IA over standard), while omnichannel brands averaged 1.02x, close to parity. Both groups improved from the prior period, but DTC-only brands moved further above parity.

Why did the results change between Haus’ 2025 The Meta Report and today?

A mix of factors, and we're not pinning it on one. The gain is broad: Both cohorts improved (DTC-only 0.94x to 1.38x, omnichannel 0.76x to 1.02x), though it's strongest and only clears parity for DTC-only. Within individual brands that re-tested the trend is mixed, so we'd call the reversal real but still developing rather than settled.

Source: Haus incrementality experiments.

‍