Artificial intelligence offers automation and data analysis benefits in mortgage lending, but its implementation demands careful evaluation and subject-matter expertise. Rigorous testing, including bias checks and scenario analysis, is crucial to ensure accuracy and manage risks for borrowers, lenders, and investors.
The accuracy and reliability of AI in mortgage lending are critical for managing risks associated with borrowers, lenders, and investors. Inaccurate AI outputs could lead to mispriced loans, understated risk in securitizations, and potential financial losses for all parties involved in the mortgage lifecycle.
Artificial intelligence presents significant opportunities for the mortgage industry by automating labor-intensive processes, enhancing data analysis, and reducing manual work across the loan lifecycle. However, the implementation of AI in this sector, where inaccuracies can have severe consequences for borrowers, lenders, and secondary-market investors, requires a sophisticated approach beyond simply deploying a model.
Julia Curran, senior managing director of residential AI products at SitusAMC, emphasized that not all AI technologies are created equal, particularly in the niche and complex mortgage market. She advised mortgage leaders to involve subject-matter experts when selecting AI models or vendors, noting that many vendors struggle to perform the specific tasks required for secondary market reviews. Curran highlighted that model selection should be tailored to the use case, considering individual model strengths, weaknesses, and costs to optimize both cost and accuracy.
Rigorous testing is paramount, extending beyond basic accuracy to encompass a wide array of scenarios. This includes testing AI's ability to handle diverse borrower profiles and income documentation consistently, mirroring the outcomes an experienced human would achieve. Continuous regression testing is necessary to ensure AI performance remains stable after model updates, and bias testing is crucial to identify and mitigate any unintended biases introduced by the AI.
While AI can effectively reduce manual work in areas like data input for loan origination systems (LOS), point-of-sale (POS) systems, or servicing platforms, Curran believes human judgment should remain central to final credit decisions. These decisions often involve complex factors beyond numerical data, requiring human oversight.
AI also holds the potential to mitigate risks associated with traditional loan sampling in due diligence. By enabling the review of all loans within a securitization pool, AI can identify potential issues in non-reviewed loans, allowing for human examination of flagged cases. Curran stated that firms are ready to move AI from testing to live workflows when testing, controls, and checks confirm results meet stringent criteria, such as a minimum 95% accuracy level, which can take months to achieve.
Ultimately, Curran warned that a lack of industry-wide standards for AI model and data set quality poses a significant risk. Firms that prioritize accuracy and rigorous testing will differentiate themselves from those that rush to market, ensuring the integrity of loans sold to investors and maintaining trust across the entire mortgage cycle.
Pick the topics you care about. Get only what matters, on your cadence.