Using data science to predict customer value

Knowing the expected lifetime spend of new customers fundamentally informs retention efforts.

Using data science to predict customer value

This article is part of Faraday's Out of the Lab series, which highlights initiatives our Data Science team undertakes and challenges they solve.

Businesses strive to both engage existing high potential customers and acquire new valuable customers who fit the right profile. Does a customer have high or low potential for purchase? Do certain customers have a high probability of cross- or up-selling? Customer lifetime value (CLV) can help a business answer these questions.

Purchase count is critical for CLV prediction

CLV is the net monetary value (read: net sum of discounted cash flows) a company earns from a customer over the customer's lifetime. Calculating how much a customer has spent to the present is trivial: sum the revenue from purchases and/or subscriptions and subtract any discounts or customer acquisition/retention costs, taking into account the time value of money. But predicting CLV out into the future is much more complex, since the lifetime of each customer is oftentimes unknown.

A critical component of a CLV calculation is the number of purchases. Predicting a customer's number of purchases out into the future significantly differs from the contractual setting to non-contractual setting.

See how credit unions can target high-value leads and build member loyalty with AI

Contractual settings

In the contractual setting — for example, a subscription business – the company knows the cadence of recurring purchases and if/when a customer churns. Thus, for each customer, predicting the total number of purchases into the future can be done by fitting a set of probabilistic models to predict the customer's likely last purchase.

Non-contractual settings

The non-contractual setting is more complex. Say we have an online retailer selling shoes: there is no guarantee of recurring revenue. Whether a customer is a repurchaser with longer inter-purchase times or single purchaser, it is very difficult to tell whether a customer has truly churned based on their last purchase. Due to these added variables, in addition to predicting the purchase after which a customer will become inactive, similar to the contractual setting, we have to fit another set of probabilistic models to predict the number of purchases while the customer is still active.

Although the topic is easy to understand from a non-technical standpoint, modeling CLV by allowing for heterogeneity in purchase frequency, dropout, and revenue is very challenging. But knowing expected lifetime spend of new customers fundamentally informs retention efforts.

Want to know more about how you can track customer journeys and lifetime value? Check out Faraday's Maximized Customer Value solution.