Insights Thought Leadership

Credit Underwriting: Improving Transparency and Avoiding Bias Through Automation

credit underwriting automation


Credit Underwriting

At recent financial industry conferences, the term Artificial Intelligence (AI for short) has begun to give way to a new phrase- XAI, for Explainable AI. This rebranding is in response to misgivings that algorithms could generate results that humans accept at face value, without interrogating the underlying logic.

In reality, any sound automated system- including ones that stop short of AI- should generate a clear documentation trail, providing comfort to interested parties as well as enabling the fine tuning of models to meet FI-specific needs.  

A recent ARGO white paper describes how credit underwriting automation, when properly implemented, can improve process transparency while also removing inconsistency and unintentional bias from existing manual routines. We covered the customer experience benefits of automation in an earlier blog post; let’s now explore the internal gains.

Focus on Growing the Customer Relationship

Most if not all aspects of the loan underwriting and approval process meet the foundational criteria for successful automation. They consist of repeatable- though often complex- tasks based on a wealth of data available in real time. Given the intense regulatory oversight central to the retail lending industry, institutions must be prepared to demonstrate consistency in their decision making, while also delivering with speed to meet customer expectations.

It’s not only appropriate but also desirable that financial institutions have varying risk appetites and lending policies, and employ unique risk models- it’s what keeps the market dynamic and fosters continuous improvement. Nonetheless this comes with a reasonable expectation that management can explain the decisions made and demonstrate that established rules were consistently applied.

A benefit to having credit policy automated within the lending system is that all aspects of the risk model (debt policy, pricing model, fee schedules, etc.) are documented and developed in one place rather than maintained by different departments within the institution.  This does not mean that human expertise is removed from the process- far from it. By automating the rudimentary components, trained underwriters and analysts can devote more time and energy to optimizing models, while customer-facing staff can channel resources away from document gathering and toward consultative services to best position clients- and by extension, the bank- for long-term success.

Regulatory Transparency and Eliminating Bias

Recent headlines have publicized the danger of unintentionally hard-coding bias into credit models, based on the perspectives of those entrusted with their programming. Regulations exist to police egregious cases, but subtle and often unwitting instances of bias- through inconsistent application of lending policies and risk models- are also a concern, and can also cause financial and reputational issues as well as ones of fairness.

Concerns have been voiced that if bias is baked into models, it will be even more difficult to identify and eradicate. It’s a valid concern, but the logic is flawed on two fronts. First, it relies on a subtle assumption that today’s purely human application of policies is inherently less biased than an automated process. Hopefully we can all agree that humans, even when well-intentioned, are not perfect. Second, if the transparency described above is in place- and auditors will almost certainly insist that it is- any inappropriate modeling factors can be detected and corrected more readily in an automated setting than a manual one.

There may still be valid reasons to consider information beyond data set forth to be evaluated in the established algorithm- it would be naïve to think this doesn’t occur today. However, an automated solution can enforce a “hard stop” if an underwriter attempts to exceed the latitude defined for their role, requiring higher level approval. Such manual interventions should be automatically tracked within the system, creating an audit trail that’s likely more robust than what exists today.  

ARGO’s recent Automated Credit Underwriting Analysis and Processing White Paper covers these topics in greater detail. Credit underwriting automation is far from an overnight exercise- it requires careful planning and in most cases benefits from a thorough examination of existing policies.  Institutions that embrace these ideas find that the go-forward benefits are well worth the effort.

Download White Paper Automated Credit Underwriting  Analysis and Processing