Will we ever see complete collections automation?
Automation can roughly be divided into operations automation – handling communications, payments, complaint registration and other day-to-day tasks automatically – and decision-making automation – making choices about which customers to call, what payment plans to offer, etc.
Operations automation is easy to automate, with the notable exception of situations where a human-in-the-loop is a key feature, whether due to customer service concerns (e.g. the perceived quality of service of talking to an actual human) or due to the unpredictability of exceptional circumstances (e.g. breadth and unpredictable nature of complaints). It is hard to see how these obstacles can be overcome by currently foreseeable technology, but these processes are likely to be disrupted in the near future by real-time assistive technologies – e.g. voice monitoring to provide suggestions on tone and approach, intelligent anticipatory data retrieval to more effectively leverage knowledge bases, etc.
Decision-making automation has traditionally been the hardest to automate, but is also the one where modern analytics (including machine learning) have had the most impact. The collections industry is rather conservative in adopting such technologies but I do expect empirical treatment workflow design to be a thing of the past by the middle of the next decade – whether through a combination of process-based and predictive modelling, or through a full-on data-driven approach like reinforcement learning.
How has QUALCO evolved its communications strategies in recent times analytics-wise?
We are striving to optimise our communications strategies using an evidence-based approach, through a series of A/B tests in which the “B” or “challenger” strategy differs in a minimal, well-defined way. This is a never-ending process faced with significant challenges: a portfolio is not a laboratory, customer circumstances and macroeconomic conditions are not under our control, collecting the necessary data may require system changes that lag far behind the pace of strategy evolution, and all of this needs to be done within ethical and regulatory constraints – but the confidence obtained from measurable results is hard to ignore, even in situations where the gains are marginal.
Is it inevitable that we will see ever more robo-advisory services and chatbots in financial services?
It is inevitable that we will see humans being replaced, for tasks that machines can do just as well, or just as cost-effectively – and this is a good thing, as long as we do not sacrifice our quality-of-service standards for the sake of cost-effectiveness. There is a danger of entering a “race to the bottom”, where perceived short-term competitive gains override longer-term objectives and values, but our regulators have shown their willingness to help us stay honest, and the public is also now aware enough to push back.
To what extent have increased regulatory demands decreased the scope for innovation in the collections industry, if at all?
Not at all. On the contrary, they have offered opportunities for innovation in areas that are not necessarily in the mainstream of current analytics practice.
A few examples:
A common theme in recent regulations is the demand for “understandability”: having an obligation to explain why a decision was taken. This is a weak point of currently fashionable approaches to predictive analytics (e.g. deep learning), and has forced us to innovate in order to catch up to the automation and predictive power expected of modern analytics, while maintaining or even improving on the understandability of “traditional” modelling.
Good regulations tend to focus on outcomes, and keep their process constraints broad and common-sense-based. Innovation is needed in the way process-based and data-driven analytics are combined to ensure that outcomes are optimised, the optimisation leads to a service level above what the regulator prescribes, and the process leading to it stays within the regulator’s process constraints.
What regulatory issues are at the top of the agenda: vulnerable customers, regulation, accurate billing, data, and help schemes?
Fairness – not just towards vulnerable customers, but towards any customer category the regulator wishes to protect – is an upcoming concern. This is particularly critical as automated, data-driven modelling and decision making expands in scope, because all data-driven approaches will, by default, perpetuate all our past biases. Again, innovation is necessary to ensure that our data-driven predictive analytics only automate what is best about us, leaving the foibles of human behaviour behind.
Is the current regulatory framework fit for purpose for the digital world?
The great thing about regulations and standards is that there are so many to choose from! There is a definite distinction between modern, trendy regulatory and advisory frameworks, like GDPR, SLP or (zooming out to less customer-centric frameworks) Basel III, and more traditional frameworks. Modern regulations tend to focus on outcomes and on common-sense provisions: even if these are sometimes open to interpretation (ultimately by a court of law), let’s be honest – we all know what they are aiming for.
Frameworks of an older or more bureaucratic vintage focus on micromanagement of the process and largely miss the point. The fact that they are not digitally aware is a side effect of this fact: if they had focused on customer rights, obligations and expectations instead, they would have aged much better.
Having said that, modern disruptive technologies do pose issues of their own, and will pose more as the technology evolves. European regulators have shown a knack for anticipating these – again, by focusing on outcomes rather than technologies or specific processes – but it is inevitable that regulations will need to track technological evolution, and equally inevitable that there will sometimes be a lag between tech adoption and regulation.
How is self-serve impacting upon the work you do?
Self-serve makes it easier to capture good-quality data, by taking human operators out of the loop. As such, it can only be a good thing for analytics.
Can analytics technology be used to help vulnerable customers and customers with mental health linked to debt?
Technology can certainly be used to enforce regulations concerning the treatment of vulnerable customers, to develop financial solutions better suited to their needs, and to provide automated monitoring of our compliance with these regulations, but care needs to be taken to ensure that the right data are being captured so that our view of the customer captures their circumstances without violating privacy constraints. However, in some areas – especially where mental health is concerned – we need to admit that we are not the experts, that other people know best, and that it is better to follow appropriate guidelines rather than try to innovate from a technical viewpoint, especially given that experimentation is a key part of innovation.
What is the best way to measure performance and incentivize collections staff? Can analytics technology help with this?
Technology can have a significant impact on the accuracy, fairness and relevance of performance measurement: ensuring that the right data are captured, that the collection staff’s performance is fairly and accurately reflected in performance metrics, and that the metrics are linked to the desired outcomes. This is not a trivial problem: there may be technical obstacles to bringing all performance-related data together from multiple systems (portfolio management, dialler, system of record, and even originator systems), and the link between metrics and outcomes is not always obvious or linear. It is important to have a mechanism in place that evaluates such needs on a par with other operational and analytics requirements, and prioritises their implementation accordingly.
Technology can also help by improving the immediacy of feedback on performance – essentially helping collections staff to do their job better, rather than just providing incentives. There is a balance to be struck between providing immediate feedback and getting in the way, but there exists a range of technologies – ranging from simple real-time performance reporting to more exotic approaches like interactive, predictive communication script guidance and real-time voice analytics – that can be adapted to fit an organisation’s priorities and culture.
What can collections learn from other customer focused industries?
From a southern European perspective, I would say: respect for the customer. However, if we regard this as a solved problem, I think that other industries have as much to learn from collections as we do from them: after all, we have a unique problem in that our customers typically don’t want to be our customers. If I had to point out one area where other industries are doing better, that would be “after-sales support”: it’s all very well to determine an optimal payment plan for a customer and set it in motion, but I feel we could be doing more in terms of monitoring that plan, providing sufficient flexibility in adapting to changes in the customer’s circumstances, meeting new needs as they come up (whether those include dealing with new debt, or providing appropriate additional credit), and helping with a customer’s longer-term rehabilitation.
One area not typically thought of as a “customer-focused industry” is healthcare. In collections analytics for data-driven decision making, we tend to worry about the ethics of essentially experimenting on our customers, and we tend to rely on regulatory guidance and personal experience when dealing with such issues, but one can’t help feeling that our approach is still relatively heavy-handed and lacking in nuance. Medicine has a long history of grappling with similar issues at much higher stakes: I feel we have a lot to learn from the field of medical ethics, especially as it applies to e.g. trials of new drugs and medical protocols.
