Google Meridian | Adstock and Hill

Why It Matters

Understanding Adstock and Hill functions lets marketers allocate spend efficiently, avoiding wasted budget on saturated channels and capturing delayed media impacts for better ROI.

Key Takeaways

• Adstock models lagged media effects over a defined window.
• Alpha parameter controls retention rate in geometric decay curves.
• Binomial decay fits prolonged effects within the max lag period.
• Hill function captures diminishing returns with half‑saturation point.
• Meridian automates parameter estimation, applying transformations internally for model.

Summary

The video introduces Meridian’s two core transformations—Adstock and Hill—designed to reflect real‑world marketing dynamics. Adstock captures lagged effects by aggregating media exposure over a user‑defined maximum lag (L), applying weighted decay curves to model how influence fades over time. The Hill function models saturation, showing that incremental spend yields diminishing returns and is governed by a half‑saturation point (ec) and a slope parameter. Key insights include the choice between geometric and binomial decay curves: geometric decay, driven by the alpha retention rate, suits fast‑acting media, while binomial decay stretches to accommodate effects persisting later in the window. Meridian learns ec from data, fixing the slope at one to maintain a concave curve that guarantees a global optimum for budget allocation. The order of operations—Adstock followed by Hill—assumes saturation derives from accumulated media history, though this can be reversed. Illustrative examples underscore the concepts: a headphone ad may prompt a purchase weeks later, demonstrating lag, and doubling spend rarely doubles sales, highlighting saturation. The half‑saturation point pinpoints where a channel loses efficiency, and the model’s built‑in estimation removes the need for manual calculations, letting users simply input raw execution data and decay preferences. For marketers, these transformations enable more accurate attribution, preventing over‑investment in diminishing‑return channels and improving budget optimization. By automating decay and saturation modeling, Meridian reduces noise, streamlines analysis, and supports data‑driven decision‑making.

Original Description

Marketing impact is rarely instantaneous or linear. In this lightboard session, we explore how Meridian translates real-world consumer behavior into mathematical transformations. Learn how the Adstock function models the delayed, "carryover" impact of media over time (lagged effects) and how the Hill function captures the reality of diminishing returns (saturation).

Learn more: https://developers.google.com/meridian/docs/advanced-modeling/media-saturation-lagging

#GoogleMeridian #MarketingMixModeling #DataScience #MarketingAnalytics #MachineLearning