I’ve written before about how banks can streamline their collections processes and increase yield through advanced analytics and automated customer decisioning. First, smarter segmentation and predictive analytics provide deeper insight into the individual circumstances of each customer. This helps determine the appropriate contact strategy and treatments to maximise the likelihood of promises and payments.

Next, customer journey orchestration delivers personalised and seamless customer experiences and captures engagement and behavioural data to create a continuous feedback loop. By automating and adapting your collections strategies through orchestration, you can dynamically ensure a customer-centric approach while delivering increased revenues with lower operating costs. Today, we’ll look at how you can take your collections operations to the next level by applying more sophisticated mathematical techniques: constraint-based optimisation and reinforcement learning.

Collections as an optimisation problem

Collections efficiency is what mathematicians might call an “optimisation problem.” Essentially, it involves trying to maximise profitability and customer satisfaction while reducing losses within a given set of operational constraints.

With an increasing variety of communication channels available, collections operations need to work out the right channel, timing and intensity of actions to take, given the number and type of collections cases in play and the resources and budgets they have available.

Constraint-based optimisation is a sophisticated technique that allows an organisation to quickly quantify the business impact of decisions and actions on collections outcomes, such as revenue, roll rates, losses and customer retention.

Collections operations can run scenarios to model a variety of constraints and outcomes, which enables managers to make informed decisions on key questions such as:

• What is the impact on revenues if collections cases increase, given our current resources and strategy?
• What is the impact on roll rates of increasing or reducing the number of collections agents?
• How will increasing the velocity of contact attempts affect customer attrition?

Organizations can deploy optimisation across the entire collections life cycle to find the right combination of channel, timing and intensity of action for each individual customer. This approach has been proven to significantly improve performance whilst balancing operational constraints such as costs and resource.

Next-best action a good start

Today, most collections optimisation strategies focus on selecting the next-best action for any given customer at any given stage of their collections journey. These next-best actions are often chosen based on a set of fixed business rules, which apply an escalating scale of treatments based on a simple combination of factors, such as the number of days delinquent, the customer’s previous payment history and the total amount to repay.

This is a good start. But it’s not going to fully optimise your returns. Imagine the collections process as a game of chess between you and your delinquent accounts. If you only consider the next-best action for each case, you’re like a chess player who only looks one move ahead. Your next action might be a pretty good move in its immediate context. But your real objective isn’t just to make a series of individually good moves – it’s to win the entire game as quickly and efficiently as possible.

Moreover, while optimisation provides a very effective way to transform collections effectiveness with data-driven decisions, it is important to note that there is no self-learning with this type of approach.

Reinforcement learning

Reinforcement learning, on the other hand, is a hot topic in machine learning that mimics how humans learn, using trial and error to determine which actions produce good outcomes. It is sometimes considered the holy grail of artificial intelligence because it can learn autonomously without human intervention. It’s the same technique used by the AlphaZero programme developed by Google DeepMind in 2017, which was able to beat every existing AI at games like chess and Go.

Instead of being taught the rules of the games, a reinforcement learning-based AI learns how to win by analysing the state of the game as a whole. Then it develops long-term strategies based not on individual moves, but optimal sequences of moves that will deliver the fastest possible victory.

Reinforcement learning holds great promise for optimising the timing, sequence and intensity of customer communication within the collections process to rapidly transform results.

How SAS can help

At SAS, our collections optimization solution leverages sophisticated techniques, such as constraint-based optimisation and reinforcement learning algorithms to analyse existing successful and unsuccessful collections data, and iteratively develop better collections strategies. It can therefore identify not only the next-best action for a given account at any given time but the right series of actions to get the best result from that account.

Moreover, our solution doesn’t just help you optimise at the individual account level, but across your entire portfolio. For example, if you have 10,000 delinquent accounts and only 10 collections agents, you probably don’t have sufficient resources to contact every single customer. Within these constraints, SAS can generate the best set of steps to take to maximise total payments. This includes decisions about which customers to focus on, how frequently to contact them, what treatments to use, and what settlement terms you should offer and accept.

A solution for any situation

While your collections team might love the idea of augmenting its existing systems with the new capabilities of SAS for collections optimization, it might also have some concerns about introducing advanced analytics. Most collections operations teams don’t have in-house data science expertise. So the prospect of using AI for decision support may seem daunting.

That’s why we’re offering a wide range of deployment options, tailored to suit any organisation’s needs. Our Collections-as-a-Service model provides a fast-start capability with a preconfigured application. We also offer Analytics-as-a-Service options, which augment your existing capabilities with custom-built advanced machine learning models. These models help you achieve maximum business value in a short time frame.

If you’d like to learn more about how to get started, check out our e-book The AI Enabled Collections Model.