- 23 Dec
Using AI In Invoice Finance
I had the opportunity to discuss Artificial Intelligence (AI), and its application to the invoice finance sector with James Beck, the CTO of Optimum Finance.
Interview With James Beck CTO Of Optimum Finance
Glenn: James, what's your role at Optimum Finance?
James: I am the Chief Technology Officer (CTO). My background is in IT development and technology, and I am working to integrate cutting edge technologies to enhance the existing processes at Optimum Finance.
Glenn: I have read a lot about how Optimum are applying AI and machine learning to the invoice finance sector - how is that progressing?
James: AI, or artificial intelligence is not really the right term to use here, it would be better called machine learning (ML) - a process by which huge data sets can be analysed by a program that develops a model based on what it learns from that data e.g. by identifying patterns within the data. The ability to record large amounts of data regarding aspects of invoice financing e.g. the occurrence of financial fraud, allows us to leverage that data when trying to spot similar patterns in the future.
Often the difficulty is accessing a big enough data set to enable the machines to spot the patterns, and develop a predictive model. This requires collaboration but has been aided by large data sets being made available by parties such as Companies House, and other financial institutions.
Glenn: Is machine learning set to replace human efforts?
James: In a word "no". It's more a case of putting powerful tools at the disposal of our staff to assist them in analysing large volumes of data, and identifying the patterns.
For example, considering the underwriting process for new invoice finance prospects. Machine learning can't reach the accuracy of a human underwriter, but it can play an important role in the risk assessment process, which is concerned with the early identification of potential fraud. Clients can be assessed on an ongoing basis to identify the early signs of potential problems, so that we can head those problems off.
A good example of how we can apply machine learning is the scanning of published articles on the internet. A machine can be continuously scanning newly published news articles that could relate to the customers that we are helping. This can be an early warning system if we identify a new story about a client.
It's time consuming and difficult for a human to manually analyse all the data so this is were a machine can help the human effort.
Glenn: How are you developing this type of capability?
James: At Optimum Finance we are working with PHD students that are on the cutting edge of this type of technology. They are helping us to build the models to analyse these large volumes of data. The data can comes from a mixture of public sources, supplemented with further data sets that we purchase.
One application is that we can then use these models to identify enterprises with a propensity to use our services. In these cases we can approach them, and offer the financial support that they need. In time it may even be that our systems visit potential customer's websites and analyse keywords in order to identify the suitability of their business for financing.
Glenn: Are there limitations when deploying machine learning?
James: Yes, for instance machine learning can be prone to bias. When a particular credit card was launched, ML was used to set credit limits for customers. An anomaly was observed whereby the ML granted male applicants credit limits multiple times larger than female applicants. It emerged that there were biases in the decision making of human underwriters, and as their decisions were used to educate the ML, the same bias was introduced. We have to be vigilant for that type of problem.
We have already launched an API (Application Program Interface) that determines the suitability of potential customers within a few minutes. The API can be integrated into the platforms of our introducers, and several major introducers have already adopted our technology. The good news is that every deal that our API accepted has been able to be honoured by our human underwriters, so our technology is proving reliable.