Can AI determine what is good advice and what a good submission is?
It’s only natural for people to assume that Artificial Intelligence (AI) can and will make informed and qualified decisions, just like a trained and qualified human would. But while 2023 showcased AI’s growing capabilities, 2024 raised more fundamental questions about its responsible use, and 2025 will assert pragmatism into how it may be used.
The allure of AI in mortgage compliance: tasks, efficiency, and limits
AI’s ability to streamline workflows is undeniable. It shines in automating repetitive, detail-oriented tasks like verifying payslips, cross-checking employment histories, or ensuring documentation is filled out correctly. Traditionally, these tasks are time-intensive, involving meticulous data comparison and entry checks. By handling these straightforward jobs, AI frees up businesses to focus on higher-value work and shortens processing times, a key advantage in a fast-paced environment.
Yet, AI’s role here is not a full-fledged replacement but rather an assistant for smaller, routine tasks. When it comes to more complex compliance evaluations—asking if a particular case constitutes good advice or assessing a nuanced financial history—AI falls short. At best, it can help verify that the data going to the lender is accurate. But deeper, more nuanced compliance checks demand human insight, judgement, and an understanding of context that current AI simply doesn’t possess.
AI can indicate where there may be an issue, flagging certain patterns or discrepancies, but the ultimate decision-making remains firmly in human hands. The “pyramid” analogy is useful here: AI can handle basic, foundational tasks at the bottom of the compliance pyramid, yet the peak—determining the quality of advice or compliance decisions—requires skilled professionals. Attempting to start from the top down would be both impractical and risky; AI needs to be built into compliance gradually, layer by layer, with a strong reliance on human oversight at every level.
Human oversight remains essential
While AI is highly effective in certain areas, it’s important to remember that it operates solely on the data it has been trained on. Its decision-making is data-dependent, and the potential for bias exists because the data it uses reflects patterns from real-world information, which is often imperfect. For example, bias in criminal profiling has shown a tendency to disproportionately affect minorities—a clear cautionary tale for relying on AI uncritically. AI algorithms, especially those that involve machine learning, have demonstrated biases—whether unintentional or inherited from historical data. This makes it dangerous to treat AI’s suggestions as final decisions without human verification.
The proper approach to AI’s role in compliance is to treat it as a tool that helps test hypotheses rather than a decision-maker. For instance, AI could flag unusual spending patterns or alert compliance teams if the data being submitted to lenders lacks certain elements. But relying on AI to provide an authoritative ‘good advice’ judgement, or to complete nuanced compliance checks independently, is not feasible. As it stands, AI should be given smaller, clear-cut tasks to manage while compliance professionals continue to oversee the broader context and interpretation.
The analogy of driverless cars is relevant: while they handle much of the driving, there’s still a steering wheel for humans to take control if needed. In the same way, compliance teams within mortgage intermediary businesses will need to “steer” AI in compliance, ensuring its findings are verified and that it hasn’t missed subtleties only a person would notice.
AI’s role in building a layered compliance system
In the future, AI could contribute to a more layered compliance approach, where it performs discrete tasks that serve as the foundation of a larger, human-led compliance structure. Tasks like ensuring data accuracy, performing document checks, and generating compliance reports are useful entry points for AI in mortgage intermediary businesses.
Over time, these small, well-defined tasks will form the base of an AI-driven compliance system, freeing compliance teams to focus on the interpretative, high-level tasks at the top of the pyramid.
This layered approach allows mortgage intermediary businesses to implement AI gradually, testing and refining its capabilities at each level. Rather than expecting AI to immediately tackle complex advisory roles, businesses can use it to support and enhance their current processes. This layered model also keeps compliance teams in control, with each new AI responsibility adding another layer of efficiency to existing human oversight without compromising the trust clients place in the business’s judgement.
Challenges in trusting AI with compliance decisions
As AI’s role expands, an essential question remains: are we ready to trust it completely? While automation can handle straightforward tasks, trust is built over time, with repeated demonstrations of accuracy and reliability. Mortgage intermediary businesses cannot start from the ‘top down’, expecting AI to replace the human touch in high-stakes compliance tasks.
The human element in compliance is not just a matter of interpretation; it’s a core part of maintaining trust and accountability in the mortgage industry. Compliance isn’t just about adhering to rules; it’s about assessing the nuances and potential risks associated with each unique client profile.
