The Growing Role of AI in Credit Scoring

August 2023

Fintech & Payments

Cara Malone

Senior Research Analyst

The use of AI, more particularly of ML in credit scoring, is not entirely novel. However, their growing presence to allow for both more accurate models with better predictive capabilities, as well as models that can be customised by clients or for particular purposes/verticals, is emerging, especially with the increased sources of data.

As we describe in our latest credit scoring research, AI allows lenders can better identify specific variables, such as risk factors to define existing new customer segments, to which new and more tailored products can be deployed. It is also leveraged to increase financial inclusion by extending traditional credit products, such as mortgages and auto loans, as well as alternative ones (ie, micro loans) to underserved consumers.

AI’s utilisation in credit scoring creates more accurate and personal offers simply by utilising considerably more varied sets of data and characteristics. Compared to traditional credit scoring models, which use over 10 characteristics to allocate points and develop a credit score for each customer, ML-driven credit scoring models use over 200 variables and a range of data to take a more complete view into a customer’s behaviour.

Furthermore, ML models use new data sources such as Open Banking, leading to the generation of self-calibrating models which ‘are able to learn patterns and variances online as the data is streamed [that] can be used for outlier detection, [which often serves] as a strong proxy for financial crime events. As such, continuous inflow of data does not only help these models to compensate for the lack of historic data (ie, credit history), but also increases their predictive capabilities achieved through their constant refresh and refinement. The automation of selection of explanatory variables and the ability to evaluate complex interactions among those are what differentiates AI models from traditional, or static, models.

One of the biggest impacts of AI is the streamlining of underwriting processes, helping organisations save considerable human and financial costs. This is due to AI’s dynamic nature in obtaining and processing data and greater integration capabilities to existing systems and processes (ie, connection of various legacy or new systems into single data lake or pipeline).

AI models can process both routine and complex cases, and decisions on multiple cases much faster and without the need for manual intervention (ie, for document checks, filling gaps), enabling underwriters, brokers, and operations teams to respond to customers very quickly and accurately. The decisioning process includes generation of data-driven insights which would be impossible to collate and evaluate manually. These models also contribute to customer centricity as better product propositions can be built around faster and better decisioning (time-to-decision and time-to-cash).

The developing use of AI in credit scoring is intertwined with transparency and explainability. According to the OECD’s definition, explainability is ‘enabling people affected by the outcome of an AI system to understand how it was arrived at [which entails] providing easy-to-understand information to people affected by an AI system’s outcome that can enable those adversely affected to challenge the outcome, notably – to the extent practicable – the factors and logic that led to an outcome.’

In addition, explainability also needs to be actionable, which means that once an explanation is provided, individuals or businesses should be able to take an action to change their behaviour or remedy the wrong data about their behaviour that is included in scoring models. This also holds true for lenders which leverage AIbased models in risk and credit decisioning. The lack of transparency and explainability, and therefore, accountability, can lead to detrimental outcomes for consumers, and pose financial, as well as reputational risks to lenders.

Linked to transparency and explainability, the use of AI has been the subject of the much-debated bias issues, specifically when its claims to improving financial inclusion are considered. To achieve fairer outcomes at lending to pockets of subprime or thin-credit file consumers, these models are increasingly being trained to eliminate discriminatory variables from the source data. Although this does not entirely guarantee zero bias, it can pave the way for lenders to actively revise or remodel their scoring systems. 

Juniper Research anticipates AI’s share in credit scoring model development to grow exponentially, in line with the developments in generative AI. Data protection and privacy will be key considerations in moving forward, and human intervention, for instance data scientists within organisations, for ethical and responsible use of AI, will be needed for the foreseeable future.