Customer is looking to employ more advanced testing mechanisms than our current use of just A/B splits (at least in some situations that need it). That includes some multi-arm bandit experimentation and more dynamic A/B testing.

I want to use multi-armed bandit to dynamically allocate traffic for my experiments. For example, I have a number of different promotions I can show, and want to know what allocation optimizes my click-through rate.

Customer today uses Optimizely for their feature flagging and web experimentation. They are looking to move away from Optimizely for web experimentation and when asked what would be one feature that they would really miss out from Optimizely, they said Contextual Multi-armed bandits. They are quite mature in their experimentation program and is generating revenue for customer in terms of driving AOV etc. They don't use visual editors as much.