Automatic traffic shifting to winners

In an ordinary A/B test, traffic is split evenly between the versions until the test ends. With bandit allocation, the tool watches the results as they come in and gradually sends more visitors to the version that is doing better. For example, if a new headline is clearly ahead after a week, it receives most of the traffic while the weaker one still gets a small share. Convert Experiences, Optimizely and VWO are examples of tools listed with this feature.

Every visitor shown a weaker version is a lost chance to convert. Shifting traffic reduces that cost while the test runs. This suits short campaigns, promotions and offers where there is no time to wait for a final answer. It is less suited to cases where you need a precise, trusted measure of the difference, because uneven traffic makes it harder to learn exactly why a version won.

• Can you choose between even splitting and automatic shifting for each test?
• Which method is used, and how quickly does it move traffic?
• Is there a minimum share that every version keeps?
• Can you still read clear results, with a measure of confidence, at the end?