03 · Decision Systems

Bandit & A/B Testing Lab

Two variants, unknown true conversion rates. Watch Thompson Sampling learn which one wins — and pay less exploration cost doing it than a fixed 50/50 A/B split ever could.

Cumulative regret
round 0
Variant A (true rate 10%)
0 pulls
Variant B (true rate 16%)
0 pulls
0.0
Cumulative regret
vs. fixed A/B (same round)
Why this beats a fixed A/B split

A fixed 50/50 test keeps sending half your traffic to the losing variant for the entire test duration — that's the cost of certainty. Thompson Sampling maintains a Beta posterior over each arm's true conversion rate and samples from it to decide allocation, so as evidence accumulates that B is better, traffic shifts toward B automatically — cutting cumulative regret without ever hard-committing early. Toggle between the two strategies above and watch the regret curves diverge in real time.