Attribution Decay and Ad Stock

Also known as: Ad Stock · Adstock Half-Lives · Carry-Over Effects · Decay Curves · Broadbent Ad Stock

Attribution decay and ad stock is the measurement framework documenting that advertising effects do not occur instantaneously at the moment of exposure but accumulate, peak, and decay across time-horizons that vary substantially by channel, by category, by creative, and by audience. Broadbent's foundational ad-stock concept formalized the carry-over structure underneath media-mix modeling and provided the half-life calibration framework that subsequent econometric attribution depends on. The framework operates as foundational measurement-architecture across contemporary marketing-effectiveness practice, with channel-level decay-curves directly determining the ROI estimates that media-mix decisions rely on. The framework matters strategically because attribution-windows mismatched against channel-decay-curves systematically misestimate channel effectiveness — short-window attribution under-credits long-decay channels (TV, OOH, podcast brand-build) while over-crediting short-decay channels (paid search, retargeting), producing allocation pressure that the underlying total-effect data does not support.

The intellectual lineage runs through marketing econometrics and cognitive-memory research. UK marketing-research executive Simon Broadbent's 1979 Journal of Advertising Research paper introduced the ad-stock concept — the carry-over weighting of past advertising into current effect-estimation. American econometric researcher Prasad Naik and colleagues' 1998 paper "Planning media schedules in the presence of dynamic advertising quality" extended the framework into wear-in / wear-out dynamics. Dominique Hanssens, Leonard Parsons, and Randall Schultz's 2003 Market Response Models synthesized the broader econometric attribution-tradition into the foundational practitioner-and-academic reference. Gerard Tellis's 1988-onward meta-analysis tradition extended the framework across category-context calibration. The intellectual antecedent runs further back to German psychologist Hermann Ebbinghaus's 1885 forgetting-curve research, which established the exponential-decay form that ad-stock subsequently inherited.

How it works

The mechanism operates through systematic differences between exposure-moment and effect-occurrence-moment that aggregate into channel-distinct decay curves. Sustained empirical research documents that brand-building channels produce longer ad-stock half-lives than activation channels, that channel-decay-curves are stable enough across categories to function as media-mix-modeling priors, and that attribution-windows must match channel-decay-curves to produce unbiased effectiveness estimates.

The framework operates through three structural features.

The first is channel-level half-life variation. Different media channels produce different ad-stock half-lives — TV typically produces 4-8 week half-lives, OOH and podcast 4-12 week half-lives, programmatic display and social-feed 1-3 day half-lives, paid search 0-2 day half-lives. Half-life variation directly determines how attribution-windows should be calibrated across channels. Operations applying uniform attribution-windows across channels with different half-lives produce systematically biased channel-effect estimates that media-mix allocation decisions cannot easily correct.

The second is wear-in and wear-out dynamics. Advertising effects do not respond linearly to frequency — initial frequency builds effects (wear-in) up to a saturation point beyond which sustained over-frequency erodes effects (wear-out). The Naik-Mantrala-Sawyer 1998 framework and subsequent dynamic-advertising-quality research formalize how creative-quality, frequency-distribution, and channel-context interact in wear-in / wear-out dynamics. Operations ignoring wear-out produce sustained over-frequency campaigns that creative-effectiveness erodes faster than incremental-reach can replace.

The third is decay-curve estimation bias in MMM. Media-mix-modeling channel-effect estimates depend directly on ad-stock parameter calibration. Misspecified decay parameters produce channel-effect bias that subsequent allocation-decisions inherit. Bayesian-prior approaches (Meta Robyn 2020-onward, Google Meridian 2024-onward) explicitly calibrate ad-stock priors against channel-benchmark data to reduce the bias risk, but the framework requires sustained methodology discipline rather than uniform-default-parameter application.

Variants

Geometric decay (constant proportional)

The simplest form — each period, ad-stock retains a fixed proportion of the prior period's stock. Easy to estimate, widely deployed in practitioner-grade MMM, and the form Broadbent's 1979 framework canonicalized.

Negative exponential decay (Broadbent canonical)

The continuous-time analogue of geometric decay, with explicit half-life parameterization. The form most commonly cited in academic econometric attribution literature and the foundation underneath most contemporary MMM tooling.

Erlang-distributed decay (delayed-peak)

Multi-stage decay forms (Erlang-2, Erlang-3) capturing the delayed-peak dynamic — effects build for several periods before peaking and decaying. Brand-building campaigns frequently exhibit Erlang-distributed decay rather than monotonic decay, and operations applying geometric forms to brand-building campaigns produce systematic peak-effect underestimation.

Saturation-with-decay (Hill function)

Combines saturation-curve and decay-curve into a single response function — captures both wear-in saturation and post-peak decay. The form Meta Robyn and Google Meridian have canonicalized in open-source MMM tooling since 2020-onward.

Channel-specific decay-with-shape-parameter

Bayesian MMM frameworks parameterize decay with channel-specific shape parameters calibrated against benchmark data, allowing decay-curves to vary by channel, category, and campaign-type within a unified-modeling framework. The methodology represents the current frontier of MMM tooling.

When it breaks

The primary failure is uniform attribution-windows across channels with different half-lives. Operations applying single attribution-windows (typically 7-day or 30-day) across channels with different ad-stock half-lives produce systematically biased channel-effect estimates. Short-window attribution under-credits long-decay channels (TV, OOH, podcast brand-build) while over-crediting short-decay channels (paid search, retargeting), producing allocation pressure that the underlying total-effect data does not support.

The second failure is zero-decay assumption for digital channels. Operations treating digital channels as zero-decay-attribution (last-click, single-touch, instant-conversion) produce digital ROI estimates that ignore the carry-over effects digital channels produce. The failure mode systematically over-credits the final-touch channel while under-crediting upstream-influence channels, producing allocation pressure toward bottom-funnel digital regardless of actual full-funnel ROI.

The third is wear-out neglect. Operations sustaining frequency past wear-out thresholds without creative-rotation or audience-rotation produce campaigns where creative-effectiveness erodes faster than incremental-reach can replace. The failure mode produces sustained over-frequency outcomes that effectiveness-benchmark data does not support.

The most expensive failure is decay-parameter misspecification in MMM. Operations running MMM with default-parameter ad-stock specifications (typically uniform-decay across channels, geometric-form, no shape-parameter calibration) produce channel-effect estimates with systematic bias inherited by subsequent allocation decisions. The failure mode is methodologically subtle and rarely surfaces in conventional MMM reporting, making it one of the most-frequently-encountered measurement-error patterns across contemporary effectiveness practice.

In the wild

Played straight. A brand integrates ad-stock-calibrated MMM with channel-benchmark priors, runs attribution-windows matched to channel-decay-curves, monitors for wear-out across sustained campaigns, and revises decay-parameter estimates against post-campaign measurement data. Most successful effectiveness-disciplined operations sit here.

Inverted. A brand explicitly rejects ad-stock methodology and runs attribution through last-click / single-touch architectures alone. The DTC era post-2018 produced sustained inversion across multiple high-profile brands.

Subverted. A brand engages ad-stock methodology meta-textually with finance organizations and trade-press — typically through MMM-publication or analyst-day disclosure of allocation-correction following ad-stock-calibrated effectiveness research.

Averted. A brand declines to engage decay-curve calibration at all, allowing attribution-windows to default to platform-reported defaults regardless of channel-decay-curve mismatch.

Canonical examples

Broadbent 1979 ad-stock concept

UK marketing-research executive Simon Broadbent's 1979 Journal of Advertising Research paper introduced the ad-stock concept and provided the foundational carry-over weighting framework underneath subsequent econometric attribution-tradition. The work has remained the canonical reference for ad-stock methodology across global marketing-effectiveness research.

Naik, Mantrala & Sawyer 1998 dynamic-advertising-quality

American econometric researchers Prasad Naik, Murali Mantrala, and Alan Sawyer's 1998 paper "Planning media schedules in the presence of dynamic advertising quality" extended the framework into wear-in / wear-out dynamics and creative-quality interaction. The work has remained foundational reference for frequency-effect curve calibration.

Hanssens, Parsons & Schultz 2003 Market Response Models

Dominique Hanssens, Leonard Parsons, and Randall Schultz's 2003 Market Response Models synthesized the broader econometric attribution-tradition into the foundational practitioner-and-academic reference. The work has remained primary reference for cross-channel attribution research-tradition across global advertising-effectiveness practice.

Tellis 1988-onward meta-analysis tradition

Gerard Tellis's 1988-onward meta-analysis tradition synthesized advertising elasticity research across multiple categories and produced channel-level generalizations underneath subsequent benchmark-tradition. The work has remained foundational reference for cross-category effectiveness research.

Meta Robyn open-source MMM (2020-onward)

The Meta Robyn open-source MMM framework (launched November 2020, lead developer Gufeng Zhou) provided the first widely-adopted open-source MMM tooling with explicit ad-stock parameterization and Bayesian-prior calibration. The framework has been load-bearing for advertiser-side MMM democratization across mid-market and enterprise operations.

Google Meridian (2024-onward)

The Google Meridian open-source MMM framework (launched February 2024) extended Bayesian-prior MMM tooling with explicit channel-specific ad-stock priors and shape-parameter calibration. The framework represents the current frontier of practitioner-grade MMM tooling and has been widely adopted by advertiser-side measurement operations across 2024-onward.

Ebbinghaus 1885 forgetting-curve antecedent

German psychologist Hermann Ebbinghaus's 1885 Über das Gedächtnis established the exponential-decay form that ad-stock subsequently inherited. The work has remained foundational reference for memory-decay research underneath broader marketing-effectiveness research-tradition.

Cadbury "Gorilla" 2007 multi-year decay tail

The 2007 Cadbury "Gorilla" campaign (Fallon London) produced a multi-year ad-stock decay tail that subsequent IPA Effectiveness Awards analysis documented in detail. The case has remained canonical illustration of long-decay brand-building effects that short-window attribution-architectures cannot capture.

Last-click attribution cautionary pattern (sustained)

Multiple advertiser operations running attribution through last-click / single-touch architectures alone have produced systematically biased channel-effect estimates that subsequent ad-stock-calibrated MMM has corrected. The pattern has remained cautionary reference across post-2017 advertiser-coalition practitioner-trade work.

---

Attribution decay and ad stock is the foundational measurement-architecture documenting that advertising effects accumulate and decay across time-horizons that vary substantially by channel, category, creative, and audience. The brands that understand the framework calibrate attribution-windows to channel-decay-curves, integrate ad-stock methodology into MMM with channel-specific priors, monitor for wear-out across sustained campaigns, and revise decay-parameter estimates against post-campaign measurement data. The brands that don't understand the framework apply uniform attribution-windows across channels with different half-lives, treat digital as zero-decay-attribution, neglect wear-out dynamics, or run MMM with default-parameter ad-stock specifications producing systematic channel-effect bias. The carry-over structure underneath advertising effects is also the most-frequently-bypassed methodology consideration across contemporary attribution practice.

---

Related insights

Attribution decay and ad stock is the foundational measurement-architecture adjacent to Marketing Mix Modeling Foundations (entry 214), which depends directly on ad-stock parameterization for channel-effect estimation. Media Effectiveness Benchmarks (entry 220) provides the cross-channel benchmark data underneath ad-stock prior calibration, while Multi-Touch Attribution (entry 216) provides the alternative attribution methodology whose limitations ad-stock-calibrated MMM corrects. Incrementality Testing (entry 215) and Brand Lift Measurement (entry 217) provide the causal-inference and brand-effect measurement frameworks complementary to ad-stock-calibrated MMM. The Long and the Short of It (entry 219) and Share of Voice vs Share of Market (entry 218) provide the brand-investment-allocation discipline that channel-decay-curve calibration ultimately supports. Mere Exposure Effect (entry 97) and Spacing Effect (entry 111) provide the cognitive-psychology foundation underneath wear-in dynamics, while Mental Availability (entry 145) and Distinctive Brand Assets (entry 144) provide the brand-equity foundation that ad-stock-calibrated investment ultimately compounds. Marketing Funnel Criticism (forthcoming entry 222) connects through criticism of linear-funnel models that ad-stock-calibrated attribution complicates. The broader pattern is that attribution-windows mismatched against channel-decay-curves systematically misestimate channel effectiveness, with sustained advertiser-side ad-stock methodology discipline operating as primary corrective against attribution-window distortion across contemporary brand operations.