The application of AI as a lever of competitive advantage is preoccupying executives across industries and sectors globally, and lending is no exception. In this article we focus on potential productivity gains in the area of mortgage origination.
AI has the potential to enhance many banking operations – such as fraud detection, customer service, and process automation – by improving efficiencies and reducing costs. It has been estimated that banks could achieve cost savings of between $200 to $340 billion from generative AI alone.1 There are also associated risks, however – including the potential for bias or discrimination, lack of transparency, and the propagation of misinformation – that must be considered and if necessary, addressed when implementing AI solutions.
In a 2023 Fannie Mae survey of senior mortgage executives, lenders cited improving operational efficiency as the primary motivation for AI adoption (73%).2 This includes multiple processes throughout the mortgage originations lifecycle such as Sales and Prospecting, Loan Setup, Processing, or Closing/Funding.
Here, AI’s ability to gather and summarize documents, perform calculations, make comparisons, and answer questions has the potential to optimize employee output by shifting from knowledge work to wisdom work. For example, AI could function as a pseudo personal assistant, performing useful tasks and generating outputs for human review, verification, and integration within established workflows to increase productivity.
Note that we exclude underwriting from the list of processes where AI should be implemented. While automated underwriting systems (AUS) such as Desktop Underwriting (DU) and Loan Prospector (LP) may generate automated approvals to determine a borrower’s eligibility, AI has limited opportunities as these systems typically determine eligibility based on strict guidelines.
Automation, like DU and LP, carries out predetermined tasks using defined guidelines. Unlike automation, the true value-add of AI is not performing repetitive tasks. Instead, AI learns by observing patterns and past outcomes, enabling it to make decisions without explicit programming. At the same time, mortgage originators must be wary of AI’s potential to exhibit bias in its decision-making and to create hallucinations that clash with numerous federal and state fair-lending laws and regulations.3
As AI matures the probability of these compliance and reputational risks will decrease, thus opening the door for its use in underwriting.
To demonstrate AI’s potential, where better to start than the ‘quarterback’ of mortgage fulfilment – the Mortgage Processor. Regardless of how your organization assigns this role, whether as a traditional all-encompassing processor or dividing tasks by including a non-customer-facing junior processor, these time-consuming, labor-intensive processes are ripe for intelligent disruption.
AI has the potential to enhance human productivity by performing useful tasks and presenting its results for human review before being added to the processing workflow. In this example, the human ‘in the loop’ would perform validation to control for any quality and regulatory or compliance risks.
To illustrate potential gains in productivity, let’s see how AI tools could increase productivity in typical processing responsibilities, using appraisal handling as an example:
Step 1 – Reprioritize the fulfilment pipeline, in aggregate and by individual Processor, based on time-related triggers or actionable events with the use of Intelligent Pipeline Management (for example, sales submission to processing).
Step 2 – Read AUS conditions to capture borrower conditions and determine necessary third-party services such as title, appraisal, flood, and MI (mortgage insurance).
Step 3 – Place order for the appropriate appraisal:
Step 4 – Generate personalized status email to borrower including a list of outstanding conditions for human review prior to sending.
Step 5 – Find an answer to a guideline question from available documentation as the human processor’s Copilot.
Step 6 – Determine need for status follow-up based on various factors (appraisal type, time elapsed, etc.) and create systemic notification to AMC vendor for status update.
Step 7 – Ingest appraisal documents via DocAI:
Step 8 – Prepare loan file for final human review prior to sending loan files for Underwriter review upon satisfactory receipt of designated documents. Operational strategy may vary in order to balance key updates with underwriter touches.
Once embedded into an organization’s workflow, institutions must monitor AI continuously to ensure long-term productivity gains. Automated systems powered by modern foundation AI models are non-deterministic and produce varying outputs each time they are invoked. Firms will achieve greater accuracy and more predictable responses with appropriate controls within the application. However, we can never truly eliminate the risk. Furthermore, modern AI systems often rely on managed models provided by third-party firms or cloud service providers, which can add further risk if operational changes are mismanaged or unmonitored.
We recommend the following monitoring strategies for lenders looking to deploy AI solutions:
Monitor efficiency metrics: Productivity should increase over time as AI solutions are further refined. On the other hand, initial gains may diminish as obstacles are encountered further down the line. Lenders should identify and track key metrics to ensure baseline improvements are preserved and put in place organizational feedback loops to refine the performance of AI systems.
Detect model drift: When vendors update technology and refine training data, this can cause AI models to produce slightly different outputs. This may reduce accuracy, degrade model performance, and lead to unexpected outcomes. To mitigate this risk, we recommend setting up automated smoke testing, monitoring and alert mechanisms which are triggered when data outputs deviate significantly from the norm.
Track AI data access: Retrieval Augmented Generation (RAG) enables AI to access real-time data, improve contextualization, and provide up-to-date responses. Similar to tracking employee access to information, lenders must track the data AI accesses through RAG techniques and adhere to identity-based access management policies.
AI is a rapidly evolving field offering numerous opportunities for competitive advantage. To avoid the pitfalls outlined in this article and maximize their investment, banks should consider partnering with a firm experienced in both AI and the financial sector. When evaluating an AI partner for mortgage processing and wider banking activities, you should consider the following:
Capco is equipped to partner with business leaders as they implement AI. Alongside industry-leading AI expertise, we’ve developed relationships with other AI leaders to provide our clients with the guidance necessary to maximize AI’s potential competitive advantage. Engage with us to learn more about our capabilities and expertise.
References
1 https://www.business2community.com/statistics-pages/ai-in-banking#:~:text=AI%20could%20deliver%20up%20to%20%241%20trillion%20of,32%25%20use%20it%20to%20at%20least%20some%20extent
2 https://www.fanniemae.com/research-and-insights/perspectives/lenders-motivation-ai-adoption
3 https://www.ibm.com/topics/ai-hallucinations
4 https://www.forbes.com/advisor/business/ai-statistics/
