Modernizing debt collection with automation and compassion

Finding a way to unify data efficiently for actionable debt resolution insights

Debt collection agencies have traditionally employed outdated methods — such as conducting home visits, sending intimidating letters or using a rigid, one-size-fits-all communication strategy — to recover outstanding debts. Much of the focus, until now, has been on recovering money rather than helping customers to reach the best solution. These strategies are typically expensive, rarely successful in recouping payments and can hugely damage customer relations. What’s more, creditors are often left in the dark when it comes to data and insights, leading to poor decision-making and increased compliance risk.

Ophelos is bringing the industry into the 21st century with a tech-first model. Customers are first encouraged onto the debt-resolution journey with an ML-driven communication strategy, which is tailored to each individual. They can then choose between a range of flexible options — such as paying now, paying later or paying over time — to resolve their debt, and ask for help in just a few taps. By focusing on delivering a great customer experience, Ophelos helps to rehabilitate individuals back into the customer journey and improve customer loyalty. All throughout the process, Ophelos loops data and insights back to businesses.

Being a young company, Ophelos had to establish their data infrastructure to support their business needs. This originally consisted of different solutions for storing data and orchestrating data pipelines, and for model exploration and model deployment, but the solution quickly became unfeasible as the requirements grew. “Our issues were mainly around how slow it was to develop and deploy to production. It was taking way too long, and orchestrating different data sources was completely unmanageable. A lot of the tools we were using weren’t user friendly, didn’t allow for collaboration, and we couldn’t test ML models while they were being built,” explains Jacob Goss, Data Lead at Ophelos.

Without a centralized data platform, model exploration was done locally in Jupyter notebooks. This led to increased time to deployment and stifled Ophelos’ ability to scale data exploration and model training. The result was time-consuming DevOps work and limited compute resources needed to explore and train large data sets. Despite their clear vision for data-driven debt collection, Ophelos didn’t have the tools necessary to develop the complex ML models they required with efficiency. Deployment took weeks and innovation had to take a back seat to infrastructure management.

Lakehouse architecture enables speed and productivity with ease

Ophelos made the decision to migrate to the Databricks Data Intelligence Platform so they could fulfill their ML use cases with ease. Ophelos stores all of their customer and third-party data in a unified lakehouse, which allows them to accommodate GDPR regulations while supporting the organization at large with immediate analysis and quick insights for smart decision-making. Since moving from Apache Parquet to Databricks Data Intelligence Platform, Jacob says, “we have schema reinforcement and we can merge into the tables so that we no longer have duplicate customer records — which can be pretty devastating to downstream processes. No issues with rollbacks or versioning. Delta Lake is so simple and easy to use — it has been a massive feature for us. It’s essentially the backbone of our workload.”

With Databricks Data Intelligence Platform, the data team at Ophelos can use integrated capabilities like MLflow to easily retrain, store and manage hundreds of models every day. They also migrated all of their feature engineering to Feature Store, which has a significant impact on productivity. Jacob explains, “We’re no longer recreating features for a specific model since we now maintain an up-to-date set of features on all our customers. Now, all we have to do is pick a target variable and we can train a new model straightaway.” The uptick in speed, productivity and capabilities has enabled Ophelos to develop the ML solutions necessary to impact debt collection.

Elevating engagement through data while lowering costs

Ophelos has developed multiple data products to support their customers and clients using Databricks Data Intelligence Platform. At the core of Ophelos’ product is the Ophelos Decision Engine — a fully automated ML-powered outbound communication strategy. The model leverages reinforcement learning to send the best communication to each customer at the right time to help them resolve their outstanding balance. Using this model, Ophelos achieved a 25% increase in customers reaching a solution, with a reduction in the number of outbound communications sent. They found that 88.3% of customers were self-served, without having to speak to anyone unless they wanted to, and 89% of customers had rated their experience as 4 or 5 stars. By allowing customers to self-serve at a time that suits them, Ophelos drives customer engagement while also freeing up time for customer support agents to help those who need it most.

Ophelos also created OLIVE (Ophelos Linguistic Identification of Vulnerability) to better support their customers in vulnerable circumstances. OLIVE is a natural language processing (NLP) model built and served using MLflow. It scans inbound customer communications to identify people in vulnerable situations and predicts the likely causes. Jacob says, “OLIVE helps our customer support agents to prioritize those in the greatest need and alerts them that they may be dealing with a vulnerable customer. We are also able to analyze our customer base to decipher macro trends in the vulnerabilities faced by individuals. Which helps to ensure we can continue to provide the best support.”

To support their clients, the team has developed Ophelos Analytics. This allows businesses to examine real-time data, make quick business decisions based on those insights and ultimately prevent more customers from falling into debt. Within the solution, Ophelos’ clients can create and download their own dashboards and reporting tables with the live data running from Databricks Photon. This is a huge improvement over typical debt collection reporting, which only occurs monthly or quarterly. Now, businesses get a fully transparent and customizable view into performance, processes and customers in real time.

Now that Ophelos can deploy models in a matter of minutes rather than weeks, they plan to continue improving engagement and support for their customers while optimizing reporting and dashboarding for their clients. Jacob says, “Databricks has given us the platform to build complex data products fast enough to keep up with the demands of an early-stage tech company. It feels so natural now that I can’t imagine going back to developing any other way.” Looking ahead, Databricks will continue to serve as the foundation and catalyst for Ophelos’ mission to transform debt collection and empower individuals to better manage their finances.

Learn more about Ophelos here.