„Innovations optimize your receivables management. Subscribe to the atriga newsletter and be the first to know!“
atriga News
In conversation with Prof. Dr Florian Artinger:
How transparent AI improves processes in receivables management (2/2)
One of Simply Rational’s main areas of activity in the field of receivables management is the question of why people do not pay their outstanding invoices and how to improve the interaction with them through scientific methods and machine learning. atriga comes into play as a welcomed cooperation and sparring partner. As a pioneer in digital and customer-friendly receivables management, the company has already developed great expertise over many years and now wants to take a further step into the future of debtor communication with Simply Rational.
Read the second part of the interview with Prof. Dr Florian Artinger here.
Is it about reducing the complexity of algorithms?
With suitable means, it is possible to break down non-transparent machine learning algorithms and make them accessible to humans. Humans and machines can then make much better decisions together than humans or machines alone. This means that the judges now have another basis for decision-making with the algorithm’s suggestion. They recognise why the algorithm made its decision the way it did and can then refine the results itself. In this context, man and machine cooperate ideally with each other.
AI is supposed to support humans, not replace them?
Right. The buzzword here is transparent AI or augmented intelligence. Ultimately, it’s about understanding machine learning as a tool and combining it with what humans are good at. How does the influence of machine learning and algorithms affect the likelihood that debtors will pay their outstanding bills or not? In answering this question, we are helped by the fact that debt management has a relatively large data pool at its disposal. Among other things, they show that the reasons for debt are relatively stable and recurring. From this data, one can make good predictions about under what circumstances and when a person will pay.
In addition, interaction with the people concerned is particularly important, i.e. responding to their individual circumstances. We make the information that the algorithm has obtained from the data records available to the employees in telephone debt collection transparently and quickly. Then the employee has a good basis of information to intuitively respond to the person being called and to pick them up emotionally. This creates a strong symbiosis of data and the expertise of the contact centre employee.
“atriga offers an exciting white label solution for the commercial dunning processes in companies and groups. These are simply picked up at the already existing interface to their SAP or other legacy systems and can thus optimally and future-oriented design the company’s own processes directly after the occurrence of a payment disruption.”
Prof. Dr Florian Artinger is a researcher at the Max Planck Institute for Human Development and Associate Researcher at Warwick Business School. Artinger works and researches at the interface between economics, management and psychology. For his research, he uses Big Data from internet markets, experimental studies and computer simulations, among other things. www.simplyrational.de
To what extent does the current Corona situation change debtor behaviour?
Just look at how Corona dramatically misled Machine Learning: In February and March of this year, the prices of flights were thrown around wildly because the database was no longer correct. So something happened with Corona that has never happened before. It’s similar to a certain extent in the debt collection sector: there were big drops in consumer behaviour at the beginning of this year. At the same time, the reasons why people make debts or do not pay outstanding debts have remained very similar. However, the proportion of reasons for short-term, and in some cases long-term, payment bottlenecks has increased significantly. On the other hand, the statement that people “simply overlooked the bill” has decreased significantly. We have systems that specifically address these individual reasons and can anticipate them. We can work with them to pick up the debtor and create a win-win situation.
What is the cooperation with atriga like?
Especially atriga has been successful in this field for many years and has done exciting work. For example, using digital communication to determine where the debtor stands and where the journey can still go. At the same time, there are additional, exciting starting points. Above all, this includes making the interaction with the debtor purely automated. However, it is sometimes no longer possible to understand why which debtor was approached in which way and what really works. Our approach is therefore to achieve transparency in these processes and then to address and automate certain aspects – for example in a series of reminders. But always in such a way that you can understand why the system behaves this way or that way. And when and how I should take control in order to be able to make even better decisions for measures.
That sounds like a great benefit for both sides?
Yes, that is true. Both sides can benefit greatly from combining atriga’s expertise with our machine-based approaches. That’s exactly the point we’re talking about with atriga right now. In addition, atriga offers an exciting white label solution for the commercial dunning processes in companies and groups. These are simply picked up at the already existing interface to their SAP or other legacy systems and can thus optimally and future-oriented design the company’s own processes directly after the occurrence of a payment disruption.
Similar to pre-court debt collection, Simply Rational tailors the content to the individual customer and their outstanding receivables using methods from psychology and machine learning. We have the knowledge from psychology and machine learning to automate these processes systematically and step by step. And to design them so individually that decisive advantages arise in the dunning process compared to the usual standard process of many companies.
How much IT knowledge is necessary for this?
We model cognitive processes of people. In the meantime, we are working a lot with machine learning to understand these aspects and to be able to build similar models. That’s why we also pursue a very technical approach and have knowledge not only from classical psychology, but also from statistics and machine learning.
How does this work in detail?
We process anonymised data sets that have been prepared in accordance with data protection regulations, with which we train the algorithms. Our clients can implement this on-site in their system in so-called Docker Boxes. The process is as follows: The client accesses the input of their existing system and then uses our algorithms to process the data and turn it into useful output, which can then be processed by the classic system. Alternatively, we use a cloud service together with our client and work there for them.
Finally, perhaps a little outlook: How will the whole topic of AI develop further?
In the 1950s, the view was that the machine was not the end product, but that it was always about the interaction between man and machine. In other words, the machine is understood as a tool that humans use. Today, the focus is increasingly on designing algorithms in such a way that humans can work with them efficiently. This is certainly one of the major trends that we are currently seeing on the horizon. AI as a tool with which humans can make good decisions and which supports them in relevant aspects.
