Advertising · Large-scale randomized field experiment and heterogeneous-treatment analysis
Personalizing Ad Load to Optimize Subscription and Ad Revenues: Product Strategies Constructed from Experiments on Pandora
Marketing Science · 2025
Summary
Consumption responses took more than a year to stabilize, while subscription responses stabilized in under six months.
For your marketing
Possible founder application
Shows why ad load can have delayed effects and why short tests may miss them. Treat the numbers as a measurement-design example, not a target for SaaS paywalls or sponsorship inventory.
Reported result
What the study found
Consumption responses took more than a year to stabilize, while subscription responses stabilized in under six months. A personalized allocation was estimated to raise subscription profits by 7% without reducing ad profits; the modeled strategy lowered average consumer welfare by 2%.
Research question
What the authors investigated
Can ad load be personalized to improve subscription and advertising revenue, and how quickly do users respond?
Design
Field experiment varying the number of ads per hour, followed by modeling of heterogeneous impacts and alternative allocation policies.
Sample and setting
More than seven million Pandora users in an 18-month experiment.
Scope and transfer
Limitations to keep in view
A single audio-streaming platform and its advertising/subscription economics. The personalized policy results are model-based extrapolations from the experiment, and the reported welfare effect is a platform-specific estimate.
Research sources
INFORMS publisher page and abstract retrieved with Firecrawl; no full text inspected for this profile.