Answers are grounded in this paper and cite their sources.
Last-touch attribution remains the dominant decision signal in digital advertising, despite extensive evidence that attributed outcomes can overstate causal impact by 2–10×. A key friction is the limited public body of detailed, production-scale incrementality studies that connect experimental design to operational decisions.We report geo-level blackout experiments at Dropbox across search engine marketing (SEM) and mobile app advertising and show how causal estimates are converted into an incrementality-adjusted (IA) profitability framework (IA-ROAS and IA LTV:CAC). In April 2025, Dropbox executed month-long U.S. blackouts with matched international controls and analyzed outcomes using a triangulation of Difference-in-Differences (DiD), Bayesian structural time series (BSTS/CausalImpact), and augmented synthetic control (GeoLift). We focus on incremental gross new annual recurring revenue (GNARR) and related funnel proxies. We find that mobile advertising exhibited substantially lower incrementality than implied by click-based attribution, while SEM displayed more durable causal performance with large substitution into SEO. Applying these results, Dropbox reallocated approximately $25M of FY25 performance marketing spend away from low-incrementality segments while limiting modeled full-year net new annual recurring revenue (NNARR) impact to roughly -$8M and improved portfolio IA LTV:CAC by +53% versus the pre-blackout baseline. These findings support a broader shift from attributed metrics to causal measurement as a prerequisite for reliable ML/AI-driven optimization in digital advertising.
No takes yet. Share an insight, caveat, or question.
Chivukula et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: