Machine Learning in Debt Collection Our goal is
Machine Learning in Debt Collection Our goal is to create a scalable and empathetic collections experience built on technology.
What is it? Field of computer science and statistics focused on giving computers the ability to make decisions they haven’t been explicitly programmed to make.
What does that mean and why does it matter? Scenario 1: Buying an Apple ? How do you explain to a computer what an apple is? ● Size ● Color ● Shape ● Texture Machine Learning: Computer Vision (Image Recognition )
What does that mean and why does it matter? For TA Scenario 2: Communication with Debtors ? How do you explain to a computer how to collect a debt? ? ? ● Payment Offers ● Channels ● Communication Style / Tone ● Frequency ● Time Etc.
What does that look like? Digital First ● < 5% of customers will receive a phone attempt on our platform ● 80% of customers engage with our product solely on mobile platforms Coded Compliance System ● ● Hard coded backend compliance handling allows us to handle diverse state and creditor restrictions seamlessly. 003% complaint rate! (+4. 7 Google Score) Automated Decision Making ● < 10% of customers on True. Accord’s platform ever interact with an agent ● 94% of payments are made without any agent interaction ● 88% of initial commitments are made self-service via the website
How does it work? ● ● Behavioral Analytics Coded Compliance Machine Learning Experimentation Pairing Machine Learning with Data Driven Heuristics Compliance Restrictions Algorithmic Optimization A customized and effective experience. Engagement Group
Types of Machine Learning Supervised Learning f(Labeled Data) = Desired Outcome Mapping the relationship between features and a desired outcome 1. 2. Classification Regression Un-Supervised Learning f(Unlabeled Data) = Grouped Data Finds patterns in provided unlabeled data and group accordingly 1. 2. Clustering Association
How do our algorithms make these decisions? Clients Example: ● Debt Size ● Debt Type Little to no customer information is used. Customer Interaction Example: ● How are they engaging? ● How often and when? ● With what content? Reach customers where and how they want to be reached.
Customer Testimonials I absolutely love that I was contacted via email & not harassed with a million phone calls, which usually end up being at a very inconvenient time. I'm really appreciative & thankful that I was able to pay off this debt at a price which I felt was fair. Thanks! Resolved 2/17 - no agent interaction I have severe anxiety so I never answer my phone which makes it harder to pay my debts. True. Accord made it easier for me to pay my debt. Google Reviews First of all. . . I am proud to have paid off my debt with your company!!! You’ve never treated me like a monster or a bad person. . . but as someone who fell on hard times and needed the kindness, care and encouragement to get back on track!! I have other debt, and I wish I could pay off every balance through your company. THANK YOU for doing things differently. I appreciate it!! Resolved 7/18 - no agent interaction You guys are so sweet, respectful in spite of the owed debt. Patient, not-overbearing or threatening. The list goes on. I was inclined to pay this off no matter how hard it was. You guys were like a friend that I didn't want to let down lol. Resolved 4/19 - no agent interaction Resolved 7/18 - no agent interaction
Questions?
Appendix Additional Reading on ML: ● Intro to ML for Non-Technical ● Machine Learning | For the Fresh Bloods ● Supervised, Unsupervised, and Deep Learning ● Radio. Lab: Talking to Machines ● Experimentation at True. Accord ● Using NLP to Classify Inbound Emails in Debt Collection
- Slides: 11