How to Design the Perfect Pilot with TrueAccord

By on June 5th, 2018 in Industry Insights
TrueAccord Blog

We like to tell our new clients that working with TrueAccord to manage debt collection will be an entirely different experience than they’re used to. Traditional collection agencies are usually fast out of the gate, calling the highest value debtors and pushing for short-term plans or liquidation settlements. The problem with this approach, apart from the lack of learning mechanism, is it sacrifices long term gains from consistent payers for short term settlements. It may move quickly but it also fades quickly and puts a damper on long-term results.

TrueAccord takes a more steady and consistent approach, both in learning and offering payment arrangements that customers stick to. Our decision-making engine, Heartbeat, is based on machine learning and it is the driving force that helps us find connections between different types of debt along common similarities. It learns as it interacts with consumers, and ends up beating our call center-based competition. It does take some time for the machine to learn and tune itself; it starts behind at first but rapidly picks up steam as the decision engine learns more and more and we begin to make adjustments based on the inputs that are unique for each client and portfolio.

Our piloting process is built to help the machine to learn and fine tune its strategy. It is the first step in determining the right debt collection approach for a type of debt As we gain more experience and work with millions of consumers, we leverage our experience working with major banks, issuers and lenders to give our subsequent clients the best results early and consistently based on those accumulated lessons.

Here’s how we build the perfect pilot process.

  1. Gather Client Data

Our new client first gives us the base collections information and content, including the number of accounts, the number of placements and the duration of each. While some clients have giant placements of 10,000-100,000 accounts per month, others have smaller sizes of 1,000-5,000. Ideally, we look for at least three placements of 3-6 consecutive months, with at least  5,000-10,000 accounts per placement. This ensures we’re analyzing enough information to ramp up the algorithm, perfect the experimentation model and enhance performance of future debts.

TrueAccord uses a dynamic, self-service model for collections, which means we try to get the customer to pay on their own without any direct contact from an agent. Only two percent of our customers ever speak to an agent on the phone or via email, and more than 90 percent of customers with balances of over $300 choose payment plans. Since we focus on getting more customers on payment plans with lower breakage rates, the liquidation curve appears slow at first because of the time it takes to optimize customer-plan fit, but it catches up and exceeds the competition’s curve in roughly 45-60 days .

  1. Build the Ideal Initial Model

When we onboard a new client, we look for specific policies and procedures they have, disclosures, content and communications to come up with the right recipe for our pilot, including duration of payment plans and settlement authority. That’s where Heartbeat’s decision-making engine comes into play. Heartbeat is always evolving based on all the data it receives.

If the client has a product that we haven’t worked with before, we gather anecdotal information and do a manual review to see what new features we can add to Heartbeat to optimize collections. We validate the anecdotal information with data by monitoring direct consumer engagement rates. For example, if more than two percent of consumers are still reaching out to our engagement team that’s an indicator that we need to improve the flow to service them manually. We build a feedback loop into the process to continually evaluate if we have the right configuration settings and determine how to modify our strategy.

  1. Experiment, Learn and Optimize

Once the pilot is up and running we continue to execute on a suite of performance enhancing experiments. Our goal is to gain data that comes from real-life testing of each strategy and content. Over time, Heartbeat learns automatically how each scenario will impact liquidation, giving us valuable insight on how the product and process are working for each customer, not just on hard outcomes such as liquidation or conversions.

Here’s a quick example of how we would adjust the pilot based on higher-than-normal engagement rates. On new types of debt, we might not be using email content that is easily understood by each audience. The consumer might not even know why they are in debt in the first place, or to whom they owe the debt (especially with debt buyers who collect on debt that may be old, and using a name that the customer doesn’t immediately recognize), and that’s why they reach out to the engagement team. When we detect a spike in contact rate and run through a manual review, we discover how to refine the messaging and target each consumer better to lower the direct engagement rate. Over time, Heartbeat learns to automatically classify each engagement so there’s no need for a person to take the time and do it.

After testing, the final step is to roll the learnings back into the machine learning models. In the end, it’s all about creating a cadence of placing accounts, monitoring performance, building consistency into the model and maximizing liquidation.

TrueAccord’s highly successful debt collection model all begins with a well-oiled piloting process designed to kick-start the machine learning engine, enhance the consumer collection experience and optimize long-term liquidation rates. To hear more, please tune into our podcast.