Multi-Armed Bandit Demo — Marketing Optimization Simulator

Control Panel turn the dials

Algorithm

Thompson Sampling Epsilon-Greedy Static A/B Test

Epsilon (explore rate) — 0.10

Leads per day — 50

Conversion delay — 0 days

Starting budget — $50,000

Speed — 5 days/sec

Reveal true rates

Conversions Budget Split Bankroll

Selected bandit vs. a static A/B test on the same lead stream. The gap is the bandit's edge.

Share of daily leads sent to each campaign. Watch it converge on the winner.

Money in the bank: bandit vs. static A/B, both starting from the same budget. The bandit compounds profit; the even split drains toward zero.

Conv. Regret Profit Regret Budget

Cumulative Regret

Leads lost by not always picking the best-converting campaign. Lower is better — a smart bandit keeps it nearly flat.

Profit Regret

Dollars left on the table by not always picking the most profitable campaign — the cost of chasing conversions instead of profit.

Bandit Bankroll

$50,000

Starting budget plus profit so far. A/B test: $50,000. The bandit grows the bankroll; the even split slowly bleeds out.

Net profit +$0 ROI —

Day 0 Leads served 0 Bandit conv. 0 A/B conv. 0