Artificial Intelligence (AI) is making a big splash in the credit industry with lenders able to reap rewards through faster production of more powerful models and automation of manual processes. In this blog we focus on the specific benefits of using AI within the Collections sector by analysing how it can solve challenges that arise from using traditional approaches.
The need for faster production of models in Collections
Collections is a highly dynamic sector with strategies often having to keep pace with operational and regulatory changes. The models that underpin collections strategies should be developed in sync with the changing environment. Traditional modelling methods, which can require substantial manual effort to produce acceptable results, can mean that you are constantly playing catch-up. Slow model development and deployment mean that collections strategies are frequently sub-optimal as they lack the most suitable configurations. Collections strategies are typically assigned using segmentation approaches.
Segmentation organises accounts into homogeneous groups exhibiting similar payment behaviours enabling contact strategies to be assigned correctly. More collections effort is generally applied to higher risk segments which drives more positive outcomes in these segments (which is of course the point!) but this also drives a reduction in the risk ranking of the overall segmentation. Combined with the effects of model deterioration through time, this means that collections segments can quickly become misaligned - with frequent calibrations or even a redevelopment necessary to correct them.
AI introduces more powerful models, quicker
One of the ways in which AI is revolutionising modelling is through an acceleration of the model development and deployment process. Recent advances in both theory and technology mean that AI techniques such as deep neural nets can rapidly produce very powerful, robust predictive models. While concerns around the black-box nature of AI models have historically limited their adoption, new methods now support the introduction of model constraints to control how the model responds to changes in the input data. This means that intuitive results can be guaranteed, enabling model governance, if required, to be agreed upfront and maintained for all future redevelopments.
With AI approaches, the change in performance outcomes can start to inform the model straight away. The ability to deploy models more regularly enables new risk drivers to be identified and the contribution of existing variables to be reconsidered. As data builds up over time, the underlying relationships influencing the model will become stronger so step changes in operational activity can be readily delivered at the earliest opportunity. AI increases the performance of collections strategies by providing more discriminatory power and enabling them to be maintained at an ‘always optimal’ level.
How rich is your data?
Collections models tend to utilise customer characteristics and account performance information, which provides good insight into payment behaviour. However, Collections operations activity also produces a myriad of additional data such as Dialler contact attempts, call outcomes and agent notes. This data can be difficult to access and interpret, meaning that traditional modelling methods can struggle to accommodate it
AI can exploit the use of these additional data sources such as operational data and online footprints, finding powerful patterns within the data. Advances in technology and processing power mean that deep neural networks can incorporate data that may have previously been disregarded. AI techniques such as Natural Language Processing will enable firms to pick out key information in agents’ notes and customer communications, bringing this rich source of data into scope for modelling and segmentation.
Collections well placed to take advantage
AI is set to deliver significant benefits to the collections sector. It can significantly improve the pace at which models are developed, optimised and deployed. The powerful non-linear techniques employed outperform traditional linear approaches, ultimately driving better collections performance and better customer outcomes. In addition, AI provides enhanced models with more data brought into scope, such as the unstructured text in agent notes, that could not previously be incorporated easily into predictive models..
Find out more about Jaywing’s AI credit risk modelling software Archetype and AI Consultancy Services.