These days I start every conversation with, “Yes, I have used ChatGPT,” just to get that out of the way for the sake of efficiency. The speed of discourse about Generative AI in all aspects of life from personal to business topics has been astounding. And while the use cases for generative AI as a consumer, such as internet searching, customer service and random poem writing for hours (who, me?) have been immediately obvious, specific application of this tech in the mortgage industry has been a more involved discussion.
The thing that separates Generative AI from other types of artificial intelligence is that it can create new types of patterns and data such as narratives, images and even code, based off of existing data used to train the model. The possibilities are endless; however, the risks are also plenty since models that are not properly managed and tested can produce biased or incorrect results.
The highly regulated nature of the mortgage industry, paired with the mandate of ensuring fairness in the process for borrowers, tends to give pause in terms of introducing new tech that is not always easy to explain. Like any other form of automation and modeling, effective controls that curate data input and robust regular testing of output results are essential to this tech being adopted. That being said, there are a couple areas where Generative AI could have a transformative impact, such as increasing underwriting explainability without adding inefficiency, and breaking out of our old patterns of thinking when it comes to solutioning.
The cost to originate a mortgage loan has continued to rise annually. According to the MBA, it cost an IMB $12,450 to originate a loan, on average, in the fourth quarter of 2022. So, the last thing lenders want is to add additional steps or cost to the loan process. However, recent public statements from regulators suggest that demand for visibility and transparency into underwriting decisions is increasing.
Lenders are being pulled between the demand to show their work and the very real need to embrace automated underwriting technology for efficiency, consistency and quality. When reviewing the appraisal specifically, underwriters increasingly have a wealth of data at their fingertips. From GSE collateral underwriting tools to third-party appraisal review solutions to good ol’ Google searches, there is a lot of data an underwriter is considering in their thought space to analyze whether an appraisal is sound in its quality and accuracy.
That thought space, in which we rely on the training and experience of a human to consider (or not consider) available data, is difficult to document efficiently. Understanding what comparables were reviewed but ultimately not included in an appraisal and why could give even more confidence in the soundness of the analysis.
But with generative AI, a quick summary of available data and how it may or may not compare to a subject property could be generated on the fly. In the same way that Microsoft and others are creating “copilots” to automatically create slide presentations and docs, generated content relevant to an underwriting decision is within reach.
Today we trust the credentialing and experience of the person rather than asking them to always document their thought process. But what if we had access to a summary of the subject property’s local market environment on the fly in plain language? Instead of just a black box score, we would have a narrative of the data considered to generate the score.
I’ve been thinking a lot about the creative solutioning process and how that might change with generative AI. As a musician and songwriter, it is a common thing to build new ideas off of existing patterns and previously created content. Pablo Picasso is often quoted as saying, “Good artists borrow; great artists steal.”
On an album I recently released, I used generative AI to create the album cover and some of the loops in the songs. I let the AI produce the raw material but then edited the design and sounds to further refine it and add my own style. It has spawned a whole new vein of inspiration and patterns that I would have never conceived from scratch. Instead of stealing from another artist, in a way I’m stealing from a machine. Oh, and I finished the entire album in weeks, not months (this doesn’t mean it is good, but it was super-efficient and fun).
My point is that I don’t believe AI is going to replace existing jobs on a massive scale, but with a thoughtful approach to keeping humans in the loop we could see productivity and new solutions exponentially increase. With the ability to generate code, AI could help business experts finally create their own apps the way they have always hoped for without the ideas being lost in translation through a requirements process. A seamless path from creativity to execution is now forming. This democratization of tech capabilities could really benefit smaller lenders that don’t have the massive tech organizations of the top 10 lenders.
So while the risks of implementing generative AI without proper controls are real and well documented, the potential of finding solutions that finally reduce the cost of originating a loan is also real. Like any other modeling technique, data quality and curation are absolutely essential.
As an industry we can spend a lot of time being paralyzed by fear of the black box or we can develop standards to test the output of these models and their impact in a way that allows innovation to continue while mitigating risks. Yes, there is a high likelihood that this type of transformative tech will change the way we work. But by embracing the human in the loop approach, new patterns of creativity and innovation can emerge from generated raw materials. Regardless of whether generative AI makes you terrified or excited, the future is going to be anything but boring. And no, I did not use ChatGPT to write this article…maybe the next one.
