Becoming the Multi-armed Bandit
summary
Argues the multi-armed bandit as a practical engine for product decisions, showing concrete tactics: start with random sampling, shift to 80% exploitation once a leader emerges, and allocate top 90% traffic to the current winner. Describes building in-house infrastructure (dynamic imports, shallow abstractions) used at Dev for emails feeds and CTAs, measuring P95 performance and conversion lift, and reports that the system reliably promotes better variants without long blind experiments.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 2:43Early software development
- 5:04Multi-arm bandit algorithms
- 9:21Data-driven vs data-informed
- 11:33Applying bandit algorithms
- 13:48UX optimization with bandits
- 16:35Practical uses in development
- 17:46Explore vs exploit
- 18:56Avoid overusing bandits
- 19:40Future of feedback-driven dev
related talks
38mPybandit: A Website Optimization Framework for E commerce SMBs
14mBuilding a Data-Driven Culture: An Anthology of Strategies and Insights
27mCustomer Driven Development : What, Why & How?
41mA/B Testing with Style
24mEffective Product Experimentation: Using Data-Driven Techniques for Iteration
24mGetting Started with Multi-variant Testing
42mMasterclass: The right path to ML
35mHow to develop Growth