THE NEW ERA OF COLLECTIONS ANALYTICS: Technology, Compliance and Best Practices

Panayis Fourniotis Pavlatos, VP of Analytics at QUALCO, sheds light on the importance of explainability, repeatability and traceability for maintaining business governance and regulatory compliance and highlights how platforms like D3E can help organisations become truly data-driven.

THERE IS MUCH HYPE AROUND MACHINE LEARNING AND ANALYTICS IN CREDIT PORTFOLIO MANAGEMENT AND DEBT COLLECTIONS. HOW DO YOU SEE THE MARKET MOVING?

The credit and debt market are now aware and ready to leverage the benefits of ML, AI, and Big Data and use them as components of an end-to-end analytics process. This process consists of best practices that can be adapted to the analytics maturity level of any given business. In its simplest form, credit and collections analytics can provide clarity and repeatability in reporting and outcome evaluation, while in more advanced business cases, it can support sophisticated, ML-based customer treatment optimisation initiatives.

In my experience, the trick to becoming an advanced data-driven organisation is to strive to productise analytics and the data pipeline that supports it, by standardising processes and applying common best practices across the board. This will eliminate most of the manual effort and allow the business – and the analysts – to concentrate on what will actually make a difference.

WHICH ARE THE MOST SIGNIFICANT ADVANCES IN DEBT COLLECTIONS BROUGHT BY DIGITAL INNOVATIONS AND ADVANCED ANALYTICS? IN WHAT AREAS DO YOU USE ALGORITHMS?

Advanced analytics, including machine learning-based treatment optimisation, has the potential to customise treatment to individual customer needs and characteristics while at the same time streamlining the treatment processes from the standpoint of the business stakeholders. This can introduce cost savings and performance improvements, while still treating customers more fairly – meaning that everybody wins. And thanks to the generality of the analytics process, such improvements can be realised across the board over the complete collections lifecycle, from identification of the early warning signs of credit risk to managing and restructuring debt.

An established end-to-end analytics process will capture and maintain all the data coming out of operational decision points, allowing us to apply data-driven solutions to:

1. Fundamental decisions such as customer treatment intensity and communication channel selection
2. Complex decisions, such as allocation to specific Debt Collection Agencies or escalation to legal action
3. Strategic decisions, such as debt restructuring solution selection.

HOW CAN QUALCO HELP CREDIT AND COLLECTIONS ORGANISATIONS BECOME MORE DATA-DRIVEN?

Data is undoubtedly a valuable source of business growth; becoming data-driven will soon become the only way to operate a business. However, becoming a truly data-driven organisation is not easy because it requires a combination of cultural and technological shifts. We provide the right technology that helps clients establish a robust, analytics-enabled collections model with QUALCO Data-Driven Decisions Engine (D3E). D3E is an integrated platform that automates and streamlines every stage of the analytics workflow, from data ingestion to predictive modelling, decision-making, and action optimisation. D3E can ingest data, rebuild models and reproduce analyses as often as necessary, with minimal human effort. Unlike other software products, D3E empowers business users and analysts to design, develop and deploy strategic and operational decision-making initiatives within a single platform, ensuring end-to-end clarity.

CREDIT AND COLLECTIONS FIRMS MUST NAVIGATE A DIVERSE REGULATORY LANDSCAPE. WHICH ARE THE REGULATIONS THAT AFFECT THE COLLECTIONS ANALYTICS SPACE?

Regulations govern what we are allowed to do, and what we can do: our constraints and opportunities. In terms of constraints, GDPR is still the most important regulation governing the AI space, not because of its limitations on the use of AI but because of its constraints on the input of the end-to-end analytics pipeline – the customer data that we work on. The soon-to-be-introduced AI Act provides more structure and substance to constraints on the actual use of AI on that data and, while not without its issues, is a significant step forward in the rationalisation of AI use, since:

1. It clarifies unacceptable and high-risk uses of AI, striking reasonable compromises where necessary and introducing a risk-based framework to ensure good governance and prevent algorithmic overreach.
2. It includes provisions to minimise its impact on operations, taking care to reuse existing supervisory mechanisms and community self-governance where it makes sense to do so.
3. It reaffirms the primacy of human supervision, requiring AI users to be aware of, responsible for and capable of overriding AI-driven decision-making in high-risk situations.

For the credit and collections sector, the impact of the AI Act is mostly in the area of significant decisions such as granting credit, and its provisions are neither surprising nor unreasonable. We do expect, however, that a service industry will grow around AI Act compliance just as it did around GDPR compliance, with analytics and decision-making products or services needing to be brought into line and growing features to ensure compliance going forward.

To meet the requirements of the AI Act, our specialist solution, D3E, enables businesses to ensure trustworthiness, transparency, fairness, safety, and accountability when using and delivering AI solutions.

HOW DO YOU SEE DEBT COLLECTIONS FIVE TO TEN YEARS FROM NOW?

Regarding AI/ML, it will be no surprise to see fairness and responsible behaviour becoming more of a concern. It’s been coming for a long time, and the EU AI Act finally formalises it in actionable and enforceable ways. This brings significant change to the types of AI that will take priority.

Explainability, repeatability and traceability will be key both for business governance and regulatory compliance. Much of this involves having an automated, traceable end-to-end process, both in terms of data governance and in terms of predictive algorithms. Platforms like D3E can ensure data governance processes are up to scratch, but the currently fashionable “deep learning” branch of AI, with its black-box approach to predictive modelling, will need to rise to some significant challenges if it is to address explainability and interpretability concerns. I expect to see much more development of add-on explainability mechanisms for deep learning, as well as a renewed emphasis on more traditional, but more explainable algorithms.

Abstract from Panayis-Fourniotis-Pavlatios interview on DB Magazine