Losing customers, also referred to as churning, is something that any company wants to prevent. But not by predicting churn, assuming correlation is causation, or by acting on prescribed actions. Let me show how to combine techniques from uplift modelling, causal inference and reinforcement learning, into one contextual bandit system that balances exploitation & exploration and deals with biases.
Losing customers, also referred to as churning, is something that any company wants to prevent, especially in industries with many subscribed customers, like Telco, Media, Finance and Insurance. The challenge then is to determine which, if any, intervention (also referred to as treatment or countermeasure) can retain a customer. In the Telco case this could be giving a discount or sending specific emails. In common approaches I see three issues: to predict churn, to assume correlation is causation, and to act on the prescribed actions. Please join me in constructing a more effective solution, combining techniques from uplift modelling, causal inference and reinforcement learning. Let us solve the correct business problem, deal with biased historical data, and balance exploration and exploitation. Along the way the audience will get a good grasp of possible design choices and techniques, with their implications. The end result will be a full contextual bandit solution, utilizing various techniques like inverse propensity scaling, Bayesian neural networks and Thompson sampling.