Case StudyPartner Page

Read about how we used machine learning to help the Director of Billing Services at a global leader for storage and information management services. Our client wanted to improve manual payment processing and streamline collections outreach to over 100,000 customers worldwide. We applied a machine learning model to the customer database to create lists of customers for focused collection follow up. The finance team was able to save time and improve revenue with proactive customer communications.


Late payments were impacting revenue at a logistics and transportation company with over 100,000 customers and thousands of facilities worldwide. Slim margins for services rendered meant our client had a low tolerance for late payments. The director of billing believed that on-time payment required education and proactive customer communications. But the sheer volume of late-paying customers made the task of following up labor-intensive and inefficient for the business. They needed better clarity into which customers required follow-up and a better way to predict customer behavior related to on-time payments.

The director of billing had recently completed a machine learning course taught by a Synaptiq consultant. He approached us about piloting a project he believed would improve his team’s productivity. With machine learning to help delinquency tracking and behavior analytics, he hoped to focus his team on obtaining payments, reducing overall late payments, and increasing revenue for the business.


The Synaptiq team built a test model on obfuscated customer data followed by a scalable model in the production environment to learn which customers finance should target with calls and emails.  Unlike traditional service projects, we were initially asked to provide advice only — there were numerous IT and security hurdles to overcome for us to build a model for them. Our client assigned a billing automation specialist to the project as he had access to their sensitive data and interest in learning how to build a model himself using a tool called Orange.

We met with the billing automation specialist biweekly to guide his approach, provide recommendations, and answer any questions.  From the pilot we discovered about 400 delinquent accounts. Using these results, the billing automation specialist set up a pilot team in the organization that used the output of the model to target nearly 200 customers with emails and phone calls.  He then measured the outcomes of these process changes.

After several rounds of pilots, the client asked us to rebuild the Orange model using a scalable, modern machine learning stack.  We wrote Python code in a Jupyter notebook and used model data to train it for the production environment. They provided us obfuscated data to train and test the model on our own then they run the model on real data after we delivered it.

Based on the model, our client learned they could create a highly accurate method for identifying problem accounts. But the finance team still needed a way to target outreach. We created a hypothesis that those accounts at higher risk for delinquency than others could be best reached via email.

How Machine Learning Helped

By using learning model to predict which customers would be late on payments, the business could focus their efforts on outreach to critical accounts meeting specific criteria. According to our client, finance successfully:

  • Limited those accounts in collection for over 180 days;
  • Targeted calling campaign to customers with outstanding invoices above a certain threshold;
  • Improved collection times by two to four weeks;
  • Created a prediction model for customers paying over 10 days late;
  • Deployed an email and calling plan manageable for the finance team.

The model we built together showed great promise in improving revenues for the business.   Our client is rolling it out to other customer segments and has asked us to help with another area of their business in 2019.