Customer-friendly debt management adds value and increases loyalty
Through randomized experiments that included an extensive customer base, hallon was able to confirm what they already suspected – it pays to treat your customers well. Now the mobile operator is using machine learning models that predict customer behavior to target customers with the right action at the right time. This achieves two important goals: reducing household debt and preventing customer churn.
Solomon Seyoum is Credit and Collection Manager at the Swedish mobile operator Tre and their sister brand hallon. He explains that the group’s commitment to a more sustainable society includes efforts to help households reduce their debt. One way to achieve this is to hold off on pulling the trigger when you are considering sending a late payment case to debt collection:
“It’s important that we don’t lose sight of the people behind the numbers. Someone may be going through a tough time and have a temporary loss of income that affects their purchasing power, only to later return to a more stable life situation,” he explains.
Solomon and his colleagues have long been convinced that a ‘gentler’ approach to debt management is good business, and it pays off through increased customer satisfaction and loyalty while reducing the customer service workload.
But what exactly is the link between the number of debt collection cases and increased churn? And how can you know which measures are appropriate for which customer – and when to take them?
An experimental workshop to predict customer behavior
To get to the bottom of these questions and optimize the customization of invoice flows, hallon has enlisted the help of its digital payment partner Billogram.
David Hallvig, VP of Data Science, is heading up the project for Billogram. His PhD in machine learning has earned him the nickname ‘Doctor Data.’ Here’s how he describes the background for the experimental workshop co-run by the companies:
“Among your customers, you always have people who want to pay but don’t have the funds in their account when the invoice falls due. Others are willing and have the funds to pay, but occasionally forget. And a small minority simply won’t pay, even if they receive several reminders. The challenge is to be able to predict which customers will pay and which won’t, so that your company can take the right action.”
David recalls that they started testing various measures on a few smaller customer segments at hallon, only to successively expand the scope to include the customer base for the entire group.
“We conducted randomized experiments in which customers who were behind on their invoices were divided into different groups. In the control group, the cases were sent on to debt collection as usual. Another group received gentle communication, such as a reminder by text message, while for a third group, no action was taken at all.”
“80% would end up paying even without debt collection”
What has the experiment taught you so far?
“We can see that about 80% of those who paid after their case was sent to debt collection would have paid anyway. And if we look at churn, it was 50% lower in the group that paid without being sent to debt collection compared to those whose cases were sent. We have therefore concluded that debt collection is a blunt instrument that significantly contributes to the risk of customer churn,” explains David.
Over at Tre and hallon, Solomon Seyoum agrees that taking the right action for the right person is simply the best approach. He also emphasizes the importance of timing:
“If you take a certain action when the customer doesn’t have the funds in their account, there’s a lower chance that it will produce results. By trying a new approach and sending the invoice and various reminders on different dates, we can increase the likelihood that people will pay once they actually have the funds.”
Hallon and Billogram are continuing to use machine learning to make further adaptations and optimizations of the automated flows that are triggered by late payment. The vast amounts of data collected in the tests are being used to build machine learning models that can predict which customers will pay even without debt collection – so that these customers are automatically managed with a gentler approach.
Fewer debt collection cases = less churn
Thanks to the lessons learned from the experiment, hallon has been able to cut its debt collection cases by 80%. This is fully in line with the group’s mission to reduce household debt and promote a sustainable society. As an added bonus, the reduced number of cases eases the burden on customer service. And as Solomon points out:
“Instead of debt collection fees, our customers can spend more on our products and services, resulting in added value.”
It also means that hallon is able to offer a pleasant customer experience, without the hassle:
“The actions we take shouldn’t unnecessarily disturb our customers. We should only contact them when we really need to. In the wrong group, even a minor inconvenience can have a major effect on churn. And we now have clear data to back that up,” says Solomon.
Deeply entrenched approaches must be challenged
What advice does Solomon have for other companies that want to experiment with machine learning?
“What will yield the best results for your business depends on your industry and customers. But I would really recommend testing the waters and making adjustments until you find the optimal solution in the area you want to improve. Of course, it helps to have a partner like Billogram, where we aren't afraid to challenge each other.”
Billogram’s ‘Doctor Data’ is also looking forward to continuing the collaboration and to applying the same experimental approach at more businesses:
“Invoice flows are substantial integrations, and there are a lot of ingrained notions about payment. In a project like this, a key to success is that important stakeholders in the company must be prepared to try novel approaches. Solomon and hallon certainly are, and everyone here at Billogram is really grateful for that,” he concludes.
Would you like to know more about how your business can benefit from AI and machine learning? Don’t miss our interview with AI expert Errol Koolmeister!