ING Automated Credit Models enhance efficiency and value for clients and employees
Credit risk decisioning demanded a lot of manual work; employees did not have sufficient time for high-risk cases
Expert-based learning enables more frequent assessment and can quickly adapt to new situations
The models were validated and approved by ING’s Model Risk Management department, IT Risk Management and Credit Risk Management
80% of reviews and 50% of loan extensions were automated, they are now piloting the acceptance of new applications
Customer journeys were improved and employee satisfaction increased
At a glance:
ING Real Estate Finance wants to remain a front-runner in the market while keeping regulators satisfied. Manual credit review processes required a lot of human capacity. Approval would take up most of the employees’ time and effort with repetitive work, hindering fast client-facing decisions. This hurdle was even greater for high-risk cases as there was not sufficient time to assess them.
Business rules or Artificial Intelligence…why not both?
ING initially asked IG&H to develop a rule-based model for credit risk review. The rule-based model eventually became too complex; the many rules made it difficult to maintain and subtle differences were not identified, ultimately making it less accurate. This led to a lot of uncertainty in the application. IG&H’s preferred approach was to capture the knowledge and experience of the experts using AI. Turning this vision into a reality required close cooperation of the risk department with the business and the data science team. An Artificial Intelligence (AI) model was developed in this way to prove its added value. Based on this success, additional models for credit extension and acceptance were created.
Expert-based learning that stays in-house
Data science projects like this can be a challenge. Both risk managers and data scientists must be on board. Therefore, a balance between the business and risk is vital to the project’s success. IG&H can connect the business, risk and the ICT part of the project, and has a lot of experience in managing stakeholders in complex organizations.
Expert-based learning (EBL) gets experts invested and engaged. This engagement is based on transparency: the model explains its outcomes and visualizes them in a dashboard for the user. By having experts in control of what the model does as well as understanding how it works, the model can easily adjust to new circumstances. The owner and sponsor of the AI model is the Credit Risk Management department itself.
AI models learn from historical data and results. However, historical data is not always available. And what happens when something changes in the outside world, such as COVID impacting the value of office space, inflation or new policies? This EBL model was trained in just a few days without the need for historical data. With expert-based learning, adjustments can be made in days or weeks compared to months. Due to the many environmental factors that impact real estate finance, the power of the model lies in its quick adaptability.
“Together with the data science team of IG&H we developed within ING a unique decision model for the real estate financing market for reviews, extensions and applications. Hereby we create added value, by using our real estate financing expertise where specific expertise is needed, like risk exceptions. By extensively reducing repetitive work, we create more efficiency and value for our colleagues.”
Hein Wegdam | General Manager ING Real Estate Finance