The problem with a pure machine learning approach to prioritising which customers need attention in post sales is that it only works where there is a concrete outcome that is recorded in the data. For example, a churned accounts or accounts that start using additional features. As this is what the model is trained against.

However, the reality isn't that clean cut. You can have high revenue customers that will never churn but the cost of solving escalations can be a massive time sink and reputational hit. It's not clear what you train the model for then?

This is where the incredible intuition of the front line comes in to building out the agent. It doesn't take a ML model to understand that a customer is struggling when there is a massive spike in support tickets and revenue isn't as high as expected from the customer at this time of year. The frontline know the tell tale they just need the tools from Delegate (Formerly 42ai) to build that out.Now, this is where things get interesting.

Once we have a foundation of signals build by the Frontline it improves the performance of the ML models. Still lots to learn and build and this is why we favour a hybrid approach.