Thompson Sampling
A Bayesian multi-armed bandit algorithm that chooses arms in proportion to their probability of being optimal.
Last updated: 2026-05-04
Definition
Thompson Sampling sits at the heart of WatEase's budget rebalancer and A/B winner-promotion logic. Each arm (channel, creative, audience) has a Beta-Bernoulli posterior over its success rate. On each tick: sample one value from each arm's posterior, allocate to the arm with the highest sample. The result: high-performing arms get more traffic, but exploration continues for arms whose posteriors are still wide. Better exploration-exploitation tradeoff than ε-greedy or UCB in most marketing settings.
How it applies in India
No India-specific behavior.
Related terms
- Multi-Armed BanditA class of algorithms (Thompson, UCB, ε-greedy) that balance exploring uncertain options against exploiting known winners.
- Budget OptimizerA solver that proposes a per-channel budget allocation maximizing predicted total ROAS subject to constraints.
- Bayesian MMMMarketing Mix Modeling implemented with Bayesian inference (typically MCMC sampling), producing posterior distributions over channel contributions instead of point estimates.
Try WatEase free
Run your WhatsApp commerce on the platform built for India — Cloud API, GST invoices, UPI checkout, opt-in tracking, and a Free Forever plan.
Start Free Today