AI use case: Predicting customer churn

Losing an existing customer is bad for business. Your stream of income gets interrupted and you have to devote resources into finding new ones. So preventing customer churn is good. But how can you do that? Well, there are sometimes a sequence of tiny things that indicate what is about to happen: less sales, smaller sales amounts, less reactions to marketing, less service usage, fewer services used, changed usage patterns, etc. Couple this with all available static data (name, address, industry, age, …) on the customer and you can start answering questions like:

Will this particular customer churn next month? How long before this particular customer will churn? What is the total lost sales value due to customer churn next month? How many customers will churn next month?

Why do this

How does your company handle customer churn after the fact? The customer has already changed his mind, taken action to leave and moved on. Getting this customer back is really hard. Now, if a customer is about to leave but has not taken any actions yet, ie he is still a customer, that is a much more preferable situation. The customer is not lost and so much easier to “win” back! Would you not be interested in knowing this about your customers before they leave?

Doing this would make you money because an existing customer will continue to buy from you. It will also save you money since you will not lose that existing income, and you do not have to spend money on finding replacement business.

How do I get started

The CRM (customer relationship management) system in your company normally holds data about your customer, both static data and data about all interactions. In smaller companies maybe the CRM is an Excel or some other rudimentary customer record. In this case there is probably only the static data available here - which is not enough. You also need interaction data, which you can get from POS (point of sales), website logs, phone records, sales database, etc.

The important thing about the data is that you need examples of both types of customers: existing active customers, and churned inactive customers. This way when you train an AI model it can learn what separates churned customers from the still active ones, it can follow the history of customer interactions and find the early warning signs of churning customers.

Read about more AI use cases here: https://praktisk.ai/posts/getting-started-with-ai/