Meta today published a white paper titled "Building a Suite of Truth: A Hybrid Approach to Measuring Incrementality," a structured guide pushing advertisers to move beyond single-source attribution toward a layered combination of experiments, marketing mix modeling, and rules-based attribution. The document, released on May 28, 2025, on the Meta for Business blog, draws on 307 studies run by 54 advertisers between March 2022 and November 2024 across North America, EMEA, APAC, and LATAM to anchor its central claim: that last-click attribution systematically misrepresents where conversions actually come from.

The report arrives at a moment when the measurement ecosystem is under sustained pressure from multiple directions - privacy regulation, fragmented consumer journeys, the growth of video formats that do not produce clicks, and intensifying competition between platform-provided and third-party measurement tools. Meta has been reshaping its attribution infrastructure for months, and this paper represents the company's most detailed public articulation yet of where it believes the industry should go.

Building a suite of truth
A hybrid approach to measuring incrementality
668701132_1287008936828259_247937562658765374_n.pdf
6 MB
download-circle

The 31% misattribution problem

The statistical case at the heart of the report is specific. According to Meta, a meta-analysis of calibration tests conducted across its technologies found that at the median, advertisers undervalue Meta by 31% when using rules-based attribution models compared with incrementality measurement. To correct for this, a multiplier of at least 1.45X was required to calibrate rules-based model outputs to the true incremental value Meta drove.

The mechanism behind this distortion is straightforward. Last-click attribution assigns all conversion credit to the final marketing touchpoint, leaving the contribution of every prior channel unaccounted for. Because this model rewards the last link clicked, it structurally over-indexes on bottom-funnel activities - particularly search - and leaves demand-generative channels such as social and video systematically undervalued.

According to the document, 31% of incremental Meta conversions are misallocated to other channels when advertisers rely on non-incremental attribution models. The figure is drawn from the same 307-study dataset spanning nearly three years and four global regions.

The advertising community has debated inflated ROAS figures and misattribution problems at length, with practitioners noting that a reported ROAS of 10 can in practice represent worse performance than a reported ROAS of 2, depending on attribution methodology. Meta's white paper formalizes that concern with numbers from its own experiment data.

A ladder of incrementality

Rather than arguing for a single replacement methodology, the paper introduces what it calls a "ladder of incrementality" - a conceptual hierarchy of measurement techniques ordered by their ability to establish causality rather than mere correlation.

Rules-based attribution sits at the bottom rung. These models measure correlated results but cannot make a causal distinction, which, according to the document, can lead to biased estimates of incrementality. Adding view touchpoints or shifting from last-touch to multi-touch rules can partially address accuracy problems, but the structural limitation remains.

Marketing mix modeling (MMM) and sophisticated multi-touch attribution (MTA) occupy the middle rungs. They offer a deeper assessment of the path to purchase than simple attribution models but still rely on statistical or econometric modeling rather than direct experimentation. The document acknowledges that in recent years the boundary between MTA and MMM has blurred, with MTA vendors incorporating more econometrics and MMM providers evolving toward faster, more granular outputs.

Randomized experiments sit at the top. According to the paper, these provide a direct causal link between marketing efforts and business outcomes and represent the most robust means to measure incrementality. Meta's own Conversion Lift tool is positioned in this category. The paper notes, however, that experiments are not always feasible in every scenario. Each methodology has a role to play when employed strategically within a framework that balances rigor with practical constraints.

The framing closely mirrors what industry measurement bodies and trade groups have been publishing independently. The Institute of Practitioners in Advertising released a comprehensive measurement framework in March 2025 that similarly argued for a combined approach spanning MMM, experimentation, and attribution, with experimentation used as the primary calibration mechanism.

What advertisers currently use

According to Kantar research cited in the Meta document - conducted in May 2025 among 1,935 decision makers across regions, industries, and company sizes - the average advertiser currently has 3.8 measurement solutions in their toolkit. That figure reflects a market that has already moved away from single-source reliance, even if the integration of those tools remains uneven.

The same Kantar research reveals a tension within those portfolios. 55% of advertisers have experienced contradicting results across different solutions, and when faced with inconsistencies, 51% tend to default to the solution that holds the most credibility within their organisation rather than the one most methodologically rigorous. According to the paper, this blind allegiance to established systems - without examining their validity - risks producing suboptimal budget decisions at scale.

The document also cites GlobalWebIndex data from 2024, covering 955,000 respondents globally excluding China, showing that Gen Z consumers are on average twice as likely to make a purchase without clicking on an ad compared to other age groups. That behavioral pattern makes click-centric measurement frameworks structurally ill-suited for the youngest consumer segment, regardless of how sophisticated the click-based model becomes.

The three core competencies

The paper organizes the practical path to hybrid measurement around three core competencies: calibration, regular experimentation, and measurement-informed ad delivery.

Calibration is defined as the process of adjusting measurement models to ensure they accurately reflect real-world outcomes by comparing predicted results with actual results. The document identifies three distinct calibration applications. Calibration of attribution involves comparing Conversion Lift results to an existing attribution model to identify the multiplier needed to correct for systematic undervaluation. Calibration of MMM can be done qualitatively, by confirming that experiments and MMM results point in the same direction, or quantitatively, by incorporating experiment results directly into model coefficients. Calibration via bid strategy involves tailoring ad bids to the expected incremental lifetime value of specific channels or audience cohorts - for example, distinguishing between branded and non-branded search, or social prospecting and social retargeting.

According to the paper, 53% of advertisers already use calibration as a means to reconcile contradictory results across solutions, based on the same Kantar May 2025 survey.

Regular experimentation is treated as the most direct route to incrementality data. According to a 2020 Harvard Business Review study cited in the document, businesses that conduct 15 or more experiments annually achieve 30% higher ad performance compared to those that run no experiments. The paper outlines specific use cases for different experiment types: single-cell Conversion Lift tests for calibrating cross-publisher budget allocation, multi-cell Conversion Lift tests for validating new campaign strategies, and the newer incremental attribution setting in Ads Manager for in-channel real-time optimisation.

Measurement-informed ad delivery is presented as the next step after a robust measurement framework is in place. This involves feeding measurement outputs directly into ad delivery algorithms. Incremental attribution is the primary mechanism described: the setting uses machine learning models trained on Conversion Lift data to optimise delivery for conversions that would not have occurred without the ad. Meta announced the global rollout of Incremental Attribution in June 2025, describing it as a product that "optimizes for and reports on incremental conversions in real time." Testing data across 37 Conversion Lift studies run between July 2024 and October 2024 showed an average 46% lift in performance for advertisers using the incremental attribution setting.

The paper also describes two additional optimisation tools. Value optimisation allows businesses to share revenue or profit figures with Meta's systems so that ad delivery prioritises conversions with higher economic value, rather than simply maximising conversion volume. The example provided: a business preferring two purchases at $100 each over four purchases at $20 each would configure value optimisation to weight the higher-value outcome in delivery. Custom attribution allows advertisers using multitouch attribution tools - through analytics integrations with partners including Adobe Advertising, Northbeam, Rockerbox, and Triple Whale - to share click-level attribution information and incorporate external measurement insights into Meta's optimisation algorithms.

Three case studies, three different contexts

The paper includes three documented case studies spanning retail, mobile gaming, and food delivery.

H&M, the Swedish multinational fashion retailer founded in 1947 with over 4,000 stores in 79 countries, is the most detailed. The company had relied on a legacy last-click attribution model on a third-party analytics platform and concluded it was not accurately capturing the incremental omni-channel impact of its media. Over several years, H&M built a measurement framework on four pillars: quarterly optimisation using MMM for strategic cross-channel budget allocation, weekly optimisation using an agile digital MMM that replaced the last-click model, in-channel optimisation through a global test-and-learn agenda using Brand Lift and Conversion Lift studies, and model calibration using multiple lift methodologies to validate MMM accuracy.

The documented results: incremental ROAS tripled in key markets from 2023 to 2025, and overall attributed ROAS in the first half of 2025 grew by 21% year over year. According to Caroline Dahle, Global Measurement and Analytics Manager at H&M, "Since including incrementality into our source-of-truth measurement, we've seen increased cross-channel ROIs."

Product Madness, founded in 2007 and part of Aristocrat Leisure Limited, provides a mobile gaming case study. Facing data availability constraints from privacy regulation changes, the company partnered with Meta's Marketing Science team to run 10 single-cell Conversion Lift studies each quarter across iOS and Android campaigns for five game titles. Campaigns were optimised for ROAS and purchase conversions, using both value optimisation and conversion optimisation delivery. The reported outcome: 3X improvement in reporting accuracy for Android campaigns, and 5X improvement for iOS campaigns. According to Stefana Pesko, Head of Socials at Product Madness, "This work has helped us to expand our measurement toolkit for improved accuracy and confidence in our results on Meta."

Huel, a UK-based lifestyle food brand founded in 2014, used Conversion Lift studies to confirm that its lower-funnel-only Meta strategy had reached saturation with audiences in mature markets. The company then shifted to testing upper-funnel reach campaigns and used Brand Lift studies to assess the headroom for awareness. The reported results from April 2024: a 6X increase in marginal return, 11.8 million more incremental people reached, and a 10% average increase in brand favorability.

Laura Geller, a New York-based makeup brand targeting women over 40, tested the incremental attribution setting in Ads Manager using a three-cell Conversion Lift experiment comparing standard click-based optimisation against incremental attribution variants. The top-performing cell, optimised using incremental attribution, showed 3.3X higher incremental ROAS than the business-as-usual cell with a 99.6% winning score, and 71% lower incremental cost per acquisition. According to Scott Kramer, Growth Marketing at AS Beauty, "We've since scaled nearly all of our campaigns to using incremental attribution as the primary optimization lever."

Why this matters for the marketing community

The paper's significance for marketers extends beyond its technical content. It arrives at the same time as a broader industry-wide reassessment of measurement infrastructure. Google cut incrementality testing budget requirements to $5,000 in May 2025, dramatically lowering the financial threshold for causal measurement. The IAB and IAB Europe published measurement guidelines for commerce media in November 2025 that similarly positioned incrementality experiments as the primary foundation, not optional enhancements. The French marketing trade group Alliance Digitale followed in January 2026 with a detailed white paper recommending hybrid combinations of attribution and contribution models.

Meta is not alone in arguing the case for MMM calibrated by experiments. Meta's own open-source Robyn MMM toolhas been available for several years, and LiveRamp announced expanded Meta measurement capabilities for retail media networks in October 2025, connecting Meta campaign data with retail sales data inside clean room environments. Meta also presented data clean room measurement case studies at Australia's MeasureUp conference in October 2025, showing quantifiable incremental revenue outcomes from first-party data integration.

What distinguishes this paper is that it connects measurement methodology to ad delivery optimisation in a single framework. The argument is not merely that advertisers should measure better, but that improved measurement inputs can flow directly into Meta's machine learning systems - changing who sees ads, how bids are weighted, and what counts as a valuable conversion. That feedback loop between measurement and delivery is the structural proposition the document is built around.

According to the paper's conclusion, "measurement is a revenue driver, not a cost center." The framing is deliberate. For years, measurement has been treated as a compliance or analytics function. Positioning it as a direct contributor to revenue performance is a shift in how the discipline is being sold internally within advertiser organisations, and it represents the case Meta is making for investment in the methodology described throughout the document.

Timeline

Summary

Who: Meta, publishing as the author of the "Building a Suite of Truth" white paper, with case studies drawn from H&M, Product Madness, Huel, and Laura Geller. Supporting research data provided by Kantar (1,935 decision makers, May 2025) and GlobalWebIndex (955,000 respondents, 2024).

What: A detailed measurement framework arguing that last-click attribution misallocates 31% of incremental conversions and that advertisers should shift to a hybrid system combining experiments, marketing mix modeling, and rules-based attribution, calibrated against incrementality data and connected to ad delivery algorithms.

When: Published on May 28, 2025, on the Meta for Business blog. The underlying dataset spans March 2022 to November 2024.

Where: The framework is designed for global application, with case studies drawing from H&M's operations across 79 countries, Product Madness campaigns across iOS and Android globally, Huel's international market expansion, and Laura Geller's North American campaigns.

Why: Meta argues that current measurement practices systematically undervalue demand-generative channels and lead advertisers to make suboptimal budget decisions. The broader context includes privacy regulation reducing user-level data availability, the growth of video and social formats that do not generate clicks, and intensifying industry pressure to establish causal rather than correlational measurement as the standard.

Share this article
The link has been copied!