As we enter the age of artificial intelligence (AI) we are increasingly able to create more reliable algorithms to understand, learn and adopt some human reasoning – all leading to great advances on economy and social life in general.
AI is growing fast and gaining relevance in business transactions because of huge data availability. Its ability to combine reasoning models with computer power is resulting in faster processor speeds to deal with all kinds of structured and unstructured information, lower hardware costs and better access to cloud services to enhance computing power. Since data is the new oil of enterprises, data-driven decisions now make the difference between being ahead or falling behind in the market.
One of the most well-known subsets of AI is machine learning (ML), a methodology that enables you to feed an algorithm with a large amount of data that lets the computer analyse and deliver data-driven recommendations and decisions as outputs. If any corrections are made in the model with limited or no human intervention, the algorithm can process and store that information to improve its future computational decision making and deliver more accurate results.
With learning models, you can analyse large amounts of data and be able to detect new dependencies, interactions and patterns leading to meaningful insights. In this way you can identify lucrative opportunities and avoid unknown risks.
ML is no longer a futuristic technology but is already being used in our daily lives. Typical examples include: the healthcare industry for identifying images to diagnose disease, insurance companies to price and market insurance policies, banks for fraud detection and stock market predictions, manufacturing companies for predictive maintenance monitoring, online retailers like Amazon to send consumers personalised offers and perform cross-channel marketing strategies, webshops to offer services with dynamic pricing and customer service departments to empower virtual assistants with chatbots and smart speakers.
ML in Credit Management
Along with the increase of business data and transactional speed, ML is becoming key for consolidating best practices in credit management to deliver just in time intelligent and accurate recommendations for credit decisions and to predict defaults.
Integrating decision making with ML in risk management allows you to explore unknown correlations and patterns to get to know your customers better and develop future strategies to increase profitability without increasing exposure.
Enterprises can now design their own intelligent SAP Credit Management system and drive order-to-cash processes such as a self-driving vehicle on the road. This is made possible by combining the advanced features of SOA People Credit Management Suite with SAP Leonardo services Machine Learning and Big Data.
The resulting smart tool increases credit manager’s analytic and decision-making capabilities, without replacing them at all. Credit managers remain essential to ensuring the right context is taken in the decision-making process.
SAP Leonardo Technology
SAP already offers a common platform where all data available in SAP and other non-corporate platforms such as social media, IoT networks and external data sources, can be centrally stored and transformed into a common structure for further processing, while fulfilling data governance and managing data ownerships.
SAP Cloud Platform works as a service model (PaaS) making it possible to implement different ML models for data available in your SAP HANA database with advanced service applications, powered by the SAP Leonardo portfolio. Service applications of SAP Leonardo such as ML and Big Data can be integrated with any corporate applications such as the SOA People software components of the Credit Management Suite. It allows you to integrate historical credit records with new data records available in SAP Cloud Platform to gain real-time insights and take accurate decisions with ML features.
Benefits of ML in Credit Management
Scorecards and decision workflows
ML can help businesses to improve credit decisions by specifically rating companies with insufficient credit records, such as start-ups and SMEs with no obligation to disclosure financial statements.
ML uses new models and run analyses using soft data from multiple sources captured by Big Data to detect businesses that have been safe for credit despite a lack of financial statements. Moreover, ML can check the reliability and accuracy of soft data sources by scoring every decision made and learning by each success and failure in order to extend the range of potential customers. Scoring models can self-adjust continuously without manual input based on micro and macro-economic variables, as well as deliver accurate predictions such as business default, delinquencies or payment delays.
Integrating decision making with ML enriches and optimises decision workflows when customers’ credit needs exceed available credit limits and tolerance thresholds, and when manual decisions are needed. ML models can learn from each decision made manually by the approver, as well as from the subsequent success or failure, and learn to automatically make own decisions or widen decision proposals for the approver in the applications below.
It is not unimaginable that credit limits will be approved by finance departments or ML systems even without having an specific sales order, since learning models based on Big Data will have the ability to predict customer’s demands long before a sales order takes place.
Today it is already possible to combine the power of ML with account receivables data to reduce the number of manual transactions. This is achieved by automating payment reconciliation with learning algorithms as well as providing predictions on all possible credit and account receivables scenarios, including expected invoice payment dates and customers’ default.
Enterprises can rely on ML with classification algorithms when dealing with a large amount of disputed receivables and deductions every year, in order to automatically determine the validity of deductions and to focus on resolving invalid deductions to increase productivity.
Studies suggest that more than half of all collection correspondence is directed at customers who would have paid even without a reminder or dunning letter. ML and decision analytics can help collection analysts to identify customers that do not require any correspondence since they will be paying, in order to prioritise and focus instead on chasing risky accounts. ML is able to predict payment dates for every invoice.
Collections can improve success ratios with ML by deciding the best approach to deal with delinquent customers based on the context of each designatory, for example a start-up or a large enterprise, the channel for communication, email or letter, and the type of language required.
Businesses and society in general need to still overcome various challenges arising from the use of this emerging technology to maximise its full potential without damaging business relations and social life.
ML use is still low widespread in credit management departments
Enterprises willing to use ML techniques need to keep data centralised in one system or platform, like the SAP Cloud Platform, and have access to technological expertise to apply these techniques. However, many companies are still running under separate information systems and data silos and most credit teams are not trained to apply ML techniques.
Learning models need to guarantee interpretable results in order to be useful
Enterprises using ML models should be able to clearly understand how the system makes decisions and be able to prove their reasoning behind these decisions at all times. But ML still has the reputation of being a black box. Depending on the decision model, ML can generate decisions that are hard to understand, making it difficult to provide auditors and supervisors with an explanation of a credit decision.
Data privacy and security
ML relies on huge volumes of data to learn how to make smart decisions. It must be compatible with privacy and data protections law such as General Data Protection (GDPR) and not compromise user’s personal data security and confidentiality. Issues such as anti-discrimination and cross-border regulation laws need to be proved.
And, last but not least, businesses need to address ethical questions about how ML should transform business relationships in areas such as taking credit decisions and managing receivables, in order to stay socially responsible and prevent unfairness and discrimination.