SUMMARY
Companies routinely experience difficulties in collecting debt incurred by their customers. As an alternative to the reactive techniques typically employed to increase a company’s ability to collect debt, preventative techniques could be employed to predict the payment behaviour of regular customers. Such a preventative technique, in the form of a machine-learning model embedded within a decision-support system, is proposed in this paper with a view to assisting companies in prioritising debt collection resources to address invoices that are most likely to be paid late. The system is capable of predicting payment behaviour outcomes linked to invoices as anticipating payment receipts during one of three intervals: 1?30 days late, 31?60 days late, or at least 61 days late. The underlying model of the decision support system is identified by selecting a suitable algorithm from among a pool of candidate machine-learning algorithms. This selection process requires the adoption of a sound methodological approach. A machine-learning development roadmap is proposed for this purpose, and applied in a practical, illustrative case study involving real industry invoice data.