Propensity Modeling to Minimize Collections Churn: Insights into What’s Driving Customer Invoice Payment

Key Takeaways

• The Propensity to Pay model enhances the efficiency and effectiveness of the collections processes and contributes to broader business strategies by providing actionable insights into customer behavior.
• Key features, such as automated past-due notifications, payment methods and previous payment success rates, have been identified as critical drivers of customer payment behavior.

Opportunity

The collections process is a crucial aspect of any business in which customer churn poses a significant risk. It serves as the final opportunity for companies to salvage their relationship with customers who are at risk of churning. This creates an opportunity to provide insight into factors driving customer payment behavior, predict payments to further strategize collections processes and improve customer retention. By identifying customers earlier who are likely to have payment issues, collections or other upstream teams can take intervention measures to prevent payment defaults, including sending reminders, proactively offering payment plans, etc.

Solution

The Propensity to Pay model predicts the likelihood of a customer paying an invoice based on historical payment pattern data. This model could be leveraged to predict customer payment at various intervals of an invoice life cycle, such as whether the customer will pay on time, before 30 days or 60 days from the due date (we settled on predicting likelihood of payment 60 days from due date).

We spent significant time curating a data set from multiple sources with wide range of information, including:

• Invoice-level details covering historical payment behavior and collections performance capturing metrics such as average days for a customer to pay, how many times they ended up at risk of payment and how responsive they were to automated reminders.
• Product usage for the service period – the idea is “you use the product → which means you like it → indicating customer value → driving on-time customer payment.”
• Existing and historical subscription information such as length of subscription, products attached to the subscription, discounts offered, payment and commitment frequency.
• Interactions with customer success managers during the invoice period, indicating engagement.
• Standard customer attributes such as monthly recurring revenue (MRR), employee size, geographical location and age of the customer.

We leveraged the above feature set, using three years of customer invoice data (3.5 million invoices for training data) from January 2020 through December 2022, and leveraged h2o to train and finalize our ensemble model, which exhibited a 10%-12% false positive rate (fpr). Based on the information provided, here are the top features that the model deemed important from a prediction standpoint.

Top Features

• Past-Due Automated Email Notifications: There were multiple email reminders that go out to a customer who are at risk of payment; few reminders stood out as important.
• Discounts Bucket :  Discount money applied on the invoice.
• Monthly Payment Term : Yes/no flag for monthly paying customers.
• Last Closed Invoice Payment Status :  On-time/late flag indicating whether the customer’s last closed invoice was paid on time.
• Payment Method:  Customers’ preferred payment method for their invoices (invoice versus credit card).
• Net Terms :  Number of days a customer has to make their payment after receiving their invoice (depends on their negotiation terms).

Key Findings

Note: These findings are based on thorough analysis of top features along with their business interpretation.

1. Key Customer Touch Points Past Due Date Tend to Drive Payment. We observed two out of five touch points being effective. One of them included a form of visual notification on the portal that only shows up after the due date that might remind active/high-usage customers who may have overlooked their due date to make a payment. The other notification focused on strong termination language that encourages satisfied customers to pay their past-due invoice to avoid churn.
2. Shift in Payment Pattern. We observed that customers were hanging onto their cash for as long as possible before making a payment. There weren’t penalties for late payment, so it is possible that in this current macroeconomic environment, they were a bit more cautious when it came to spending.
3. Previous Invoice Payment Success Rate .  A customer’s ability to pay their current invoice on time is largely dependent on the outcome of whether their immediate previous invoice was paid on time. Overall, it was observed that 95% of customer invoices were paid on time whenever their previous invoice was also paid on time.
4. Discounts Offered to Customers Do Not Mean They Will Pay Their Invoices on Time.  Across each of our subsidiaries, it was observed that customers who ended up receiving no discount had a higher percentage of on-time invoice payment compared with customers who received a discount. If the customer has funds and is satisfied with services, they will continue to pay on time.
5. Credit Card versus Invoice. As expected, credit card customers did better compared with invoice customers. Customers on invoice payment terms have to take action on their payments, whereas card customers are set up to auto-pay; therefore, this additional manual touchpoint for invoiced customers can lead to delays.

Model Applications

Here are some ways to use the results of the model for making day-to-day operational decisions.

1. Targeted Follow-ups.  Update our collections team assignments to assign manual resources to customers with the highest likelihood of payment. This would increase the yield of dollars saved per collector. Research found that the collection outcomes for customers with lower product engagement were consistently low regardless of manual outreach by a collector.
2. Improve Customer Retention. Use the model results to engage earlier with customers who are likely to have payment issues and take intervention measures to prevent payment defaults. This could include sending reminders and proactively offering payment plans.
3. Use the results to predict fluctuations and forecast nonpayment cancellations churn and write-off.
4. Customer Profiling. Analyze customer base and deals to understand which attributes have adverse impacts on customer payment behavior. For example, if customers with high discounts and extended payment terms in LATAM are significantly less likely to pay, we can share that data with the sales teams and use it to inform go-to-market (GTM) motions. 

In summary, the Propensity to Pay model not only enhances the efficiency and effectiveness of the collections processes, but also contributes to broader business strategies by providing actionable insights into customer behavior, ultimately driving improved financial outcomes and customer satisfaction.

By leveraging historical payment patterns and a rich data set spanning three years, this model predicts the likelihood of invoice payment within 60 days from the due date. Key features, such as automated past-due notifications, payment methods and previous payment success rates, have been identified as critical drivers of customer payment behavior.