AI and Machine Learning in Credit Control
Artificial Intelligence (AI) is poised to transform many digital processes. In particular, machine learning, as a subset of AI, presents a step-change in credit control processes.
Through the application of machine learning, credit controllers will be able to identify credit risks more rapidly. Moreover, this will guide credit control processes to improve cash flow and limit credit risk.
What is machine learning?
Machine learning is an evolution beyond the use of algorithms in business processes. Specifically, machine learning is where a machine refines the algorithms it uses in data processing. This process of learning is typically supervised by a human. However, it has the capacity to learn and evolve to process data with greater speed and accuracy than its human supervisor(s).
Machine learning is categorized in a number of different ways, depending on how supervised and reinforced the learning is. Supervision and reinforcement improve efficiency of the output but require considerable human input. For instance, the vast number of humans manually assisting the learning of driverless cars.
As computing power becomes more readily available, less supervised machine learning becomes more practical. Deep learning is an advanced subset of machine learning, where the machine learns from various interdependent layers of data, rather than the more simplistic models required for traditional machine learning.
How is credit scoring normally done?
Traditionally, credit scoring was done via calculations using a set of rules, or a model. Namely, the Altman Z-score and its modifications and variations. This approach uses linear discriminant analysis based on several accounting indicators [Altman et al, 2014]¹.
Credit scoring undertaken in this way typically requires a specific set of accounting entries as an input for the calculations, at the exclusion of any companies which do not provide this. The exclusion of some companies in this way restricts the breadth of the data in the analysis.
Looking at the pattern of typical data availability for credit scoring, we can see that a lot of additional data is also available. This is described as depth of data. However, this data is not normally taken into account – even if there is potential relevance for credit risk calculations. The main reason for this is fluctuation in the availability of data. With a rigid model, fluctuations in data availability result in exclusions from the analysis. To summarize, increasing the depth of data in credit scoring reduces the breadth of data.
Why is machine learning relevant to credit control?
Fundamentally, machine learning enables a far greater degree of flexibility in modelling. Introducing flexibility needs to be paired with an ongoing review of performance and iterative improvements resulting from this. Teaching a machine to modify its models based on results from a feedback cycle like this is where machine learning really comes to the fore.
As machines weigh up the relative impact of various data on credit risk, they can dynamically adjust the weighting from each area. This enables a wide range of different data sets to be used effectively for credit scoring. In conclusion, this means that greater data breadth can now be achieved when using greater data depth.
Initially, such analysis needs to be supervised and reinforced, but over time the machine learns to make its own judgements. Consequently, the complexity of such machine learning approaches can far exceed what is feasible for a human to regulate and supervise.
Invoice payments and credit scoring
At Corcentric, we are particularly excited to see this taking effect through the utilization of invoice payment timeframes and billing interactions. Late payment, delayed response and the requirement for payment reminders are all factors which can be utilized in credit scoring – providing greater depth than traditional scoring models.
As sellers shift to electronic invoicing, this data becomes easily available via the e-invoicing platforms. These can be interrogated via API or data export, by credit scoring systems, to furnish their machine learning processes with this data.
A pincer movement via e-invoicing
One the one hand, data from e-invoicing platforms can be used to better inform credit scoring tools. On the other hand, e-invoicing provides early warning signals and better controls over accounts where there is a higher credit risk.
Through a web-based e-invoicing platform, such as Corcentric, users can indicate their intent to pay – even if they are not paying immediately. Combining this with real-time visibility of whether buyers have received, read or paid invoices gives a better insight into payment status.
Higher risk accounts can be managed more tightly through e-invoicing. Early payment discounts can be used to incentivise payment. Combined with easy online payment gateways and clear communication processes, e-invoicing equips businesses with the tools to more proactively manage their credit risks.
- [Altman et al 2014] Altman, E.I., Iwanicz-Drozdowska, M., Laitinen, E.K. and Suvas, A., 2014. Distressed Firm and Bankruptcy Prediction in an International Context: A Review and Empirical Analysis of Altman’s Z-Score Model.