Upper Confidence Bound

An algorithm for the explore-exploit dilemma that balances choosing the best-known option (exploit) with trying uncertain options that might be better (explore). UCB selects the option with the highest upper confidence bound—a combination of estimated value plus an uncertainty bonus. Options that haven't been tried much get a large uncertainty bonus, encouraging exploration. Well-tested options compete on estimated value. Over time, the algorithm naturally shifts from exploration to exploitation as uncertainty decreases. It formalizes the intuition that you should try unknown things proportionally to how uncertain you are about them.

When to use it

When deciding how much to invest in exploration vs. exploitation; when resource allocation needs to balance proven approaches with promising unknowns; when designing experimentation programs that determine how many resources to allocate to new ideas; when the explore-exploit tradeoff is explicit in strategy.

How it can help

Apply UCB thinking to resource allocation decisions. New, untested approaches deserve investment proportional to your uncertainty about them—not because you think they'll work but because you don't know yet. The practical version: allocate a 'exploration budget' (time, money, attention) to untested approaches proportional to how little you know about them, while concentrating the majority of resources on proven approaches. As you gather data, the exploration budget naturally shifts toward exploitation of winners. Google's '20% time' and Amazon's experimental product launches are organizational UCB.