Re: [attribution] Methodological concerns regarding Attribution Level 1 and causal measurement

Hi Rick,
Thanks for the feedback. I would like to understand precisely what
technical gaps you see in using the attribution API to measure causal
signals like incrementality. From my perspective, the API is well-suited to
run an incrementality study on a single publisher at a time, where
diversion is done using a first party identifier. The flow is something
like this:

- Divert users using 1P identifier
- For users in the ablation group, continue to log an "impression" /
opportunity when the creative would have been shown
- Log the experiment group in addition to other fields like campaign ID at
conversion time
- Breakout the conversion counts by experiment group

The gap as I see it is that we lack a good tool for consistent cross-site
diversion. Unfortunately supporting that is rife with challenges and
difficult constraints (some of which we tried to solve in
https://wicg.github.io/turtledove/).

I am curious: do you agree with my characterization? I wonder if we are
missing something with respect to running incrementality on even a single
publisher.

Best,
Charlie.

On Tue, May 26, 2026 at 6:09 AM Rick Bruner <rickbruner@gmail.com> wrote:

> To the Attribution Working Group,
>
> Thank you for the opportunity to comment on the Attribution Level 1
> specification.
>
> By way of background, I have worked in advertising measurement and
> analytics for more than two decades, including research and analytics
> leadership roles at DoubleClick and Google, and I currently focus on
> incrementality measurement and experimental design.
>
> At the outset, I want to acknowledge the legitimate and important goals
> behind this work, particularly around privacy preservation, browser
> interoperability, and reducing reliance on opaque third-party tracking
> mechanisms. Standardized browser-level APIs for advertising-related event
> reporting may offer meaningful improvements in transparency, consistency,
> and user privacy relative to prior generations of ad-tech infrastructure.
>
> Nothing in this comment should be interpreted as opposition to
> privacy-preserving browser infrastructure itself.
>
> My concern is instead with the underlying measurement paradigm being
> standardized.
>
> The specification appears to formalize and modernize a fundamentally
> observational attribution framework that remains unable to reliably measure
> advertising incrementality or causal advertising effectiveness.
>
> Traditional attribution modeling generally does not estimate a
> counterfactual. It observes advertising exposures and subsequent conversion
> activity, then allocates credit across conversion pathways using
> deterministic or statistical rules. This is not equivalent to measuring
> whether advertising caused an incremental business outcome.
>
> The core methodological problem is longstanding and well understood in
> both academic and practitioner literature: consumers exposed to advertising
> are not randomly selected. Users already close to purchase naturally
> generate more searches, clicks, site visits, retail activity, and
> conversion events. Attribution systems systematically confuse this
> underlying purchase propensity with advertising persuasion.
>
> This is not merely a weak form of experimental inference. In most
> implementations, attribution modeling is not a causal methodology at all.
>
> This concern is heightened by the specification’s repeated
> characterization of attribution as a means of identifying “effective”
> advertising and improving advertising performance. The document states that
> attribution enables advertisers to learn “what ads perform best” and to
> spend more on “effective advertising” and less on “ineffective
> advertising.” However, observational attribution systems generally do not
> establish whether advertising caused incremental outcomes. At best, they
> identify statistical associations between exposures and subsequent actions.
>
> Given the extensive longstanding critique of attribution modeling’s
> inability to distinguish causation from selection effects and pre-existing
> purchase intent, these claims risk overstating what such systems are
> scientifically capable of determining.
>
> The danger is not merely imprecision. It is the creation of a highly
> persuasive but potentially misleading illusion of measurement validity,
> particularly when supported by sophisticated dashboards, large-scale
> behavioral telemetry, and algorithmic optimization systems.
>
> These limitations have been extensively discussed for more than a decade
> by researchers and practitioners, including within the attribution
> literature itself.
>
> Selected references include:
>
>
>    - Dalessandro, Brian et al. “Causally Motivated Attribution for Online
>    Advertising.” ADKDD 2012.
>    https://www0.cs.ucl.ac.uk/staff/w.zhang/rtb-papers/causual-conv-att.pdf
>    - Hercher, James. “Beware The Platform Incrementality Trap.”
>    AdExchanger, 2025.
>    https://www.adexchanger.com/commerce/beware-the-platform-incrementality-trap/
>
> One particularly important consequence is structural lower-funnel bias.
>
> Attribution systems systematically favor media channels positioned closest
> to observable conversion activity, especially:
>
>    - paid search,
>    - retargeting,
>    - retail media,
>    - click-oriented social advertising,
>    - and platforms with persistent identity and commerce visibility.
>
> These systems are exceptionally effective at harvesting existing demand
> signals. Modern advertising platforms increasingly use sophisticated AI and
> algorithmic optimization systems to identify users exhibiting behavioral
> indicators associated with imminent purchase intent. Because attribution
> models observe users already in motion toward purchase, such channels tend
> to receive disproportionate measurement credit regardless of their true
> incremental contribution.
>
> Conversely, media that often create demand rather than merely intercept
> it, including television, audio, out-of-home, premium video, sponsorships,
> and broader brand advertising, are structurally undercredited because their
> effects are frequently delayed, probabilistic, indirect, or not easily tied
> to clickstream behavior.
>
> This bias does not merely distort reporting dashboards. It distorts budget
> allocation, optimization systems, and ultimately competition within the
> media marketplace.
>
> The consequence is not merely technical mismeasurement. It is a systematic
> reallocation of advertising investment toward the largest digital
> platforms, particularly those with privileged access to high-intent
> behavioral data, persistent identity systems, retail signals, and
> large-scale AI optimization infrastructure. Standardizing
> attribution-centric measurement frameworks risks further entrenching these
> advantages at the expense of smaller media companies and independent
> publishers whose contributions are less easily captured through clickstream
> attribution. This concern is especially significant at a moment when many
> media organizations, particularly news publishers, face severe economic
> pressure and existential business challenges.
>
> Modern advertising platforms increasingly optimize delivery algorithms
> toward attributed conversions. This creates a self-reinforcing feedback
> loop in which platforms learn to identify and target users already most
> likely to convert, thereby improving attributed performance metrics without
> necessarily improving incremental business outcomes..
>
> The resulting feedback loop can create the illusion of advertising
> effectiveness while increasingly concentrating spend into channels
> optimized primarily for demand interception rather than incremental demand
> creation.. In this environment, advertisers themselves may become among the
> largest losers, as substantial portions of marketing investment risk being
> redirected toward measurement-favored but weakly incremental activity under
> the implicit endorsement of major industry standards bodies and technical
> organizations.
>
> For this reason, I believe there is substantial risk in industry bodies
> and standards organizations implicitly legitimizing attribution frameworks
> as authoritative measures of advertising effectiveness without clearer
> acknowledgment of their methodological limitations.
>
> The concern is not that pathway telemetry lacks all utility. Descriptive
> reporting of advertising exposures and conversion sequences can be
> operationally useful for workflow management, reconciliation, and campaign
> diagnostics. However, descriptive telemetry should not be conflated with
> causal evidence of advertising effectiveness.
>
> I would strongly encourage the Working Group to distinguish more clearly
> between:
>
>    1. privacy-preserving event reporting infrastructure, and
>    2. scientifically valid methods for measuring advertising
>    effectiveness.
>
> The former is an important engineering and standards objective. The latter
> remains fundamentally unresolved by attribution modeling itself.
>
> The advertising industry — and especially influential standards and trade
> organizations — should instead be advancing the science of advertising
> measurement toward more scientifically grounded incrementality
> methodologies, particularly randomized controlled experimentation and
> large-scale geo-based experiments capable of estimating causal lift through
> explicit counterfactual comparison.
>
> For reference, I have published an overview of one such framework here:
> https://github.com/rickcentralcontrolcom/geo-rct-methodology/
>
> I would also encourage the specification to explicitly clarify that
> Attribution Level 1 should not be interpreted as measuring causal
> advertising effectiveness or incrementality absent separate experimental
> methodology.
>
> Absent such distinctions, there is a meaningful risk that the
> specification will further institutionalize a measurement approach widely
> understood to be methodologically inadequate for causal inference, with
> downstream consequences for advertiser decision quality, media competition,
> and the allocation of advertising investment across the digital ecosystem..
>
> Rick Bruner
> CEO, Central Control
> Founder & Moderator, Research Wonks
> U.S. Vice Chair, I-COM
>