Appraisals get transparency boost from AI, according to exec

June 10, 2024 5:15 AM

Artificial intelligence is making the appraisal space, which has been marred by instances of bias, more transparent.

According to Kenon Chen, executive vice president of strategy and growth at Clear Capital, his company is using machine learning and other AI tools to attain fairer assessments of properties.

"[We're] trying to solve a national problem, but also make it accurate at a community level," he said. "These types of tools help us build a national standard for how this should be approached and in the underwriting process, ensure that we're measuring the quality of the appraisal and the quality of the condition ratings against a repeatable standard."

Apart from implementing machine learning and computer vision to analyze and compare property values, the real estate valuation technology company is also pondering how generative AI can be integrated into the appraisal and underwriting process.

"There's a lot of great new possibilities that are being explored with generative AI and we're certainly looking at that as well," Chen said.

READ MORE

"There's still obviously very real challenges that we as an industry need to tackle, particularly in the racial housing gap," he added. "We still have a long way to go in closing that gap for especially black and brown homeowners and it's important to ensure that the techniques that we're using for underwriting for other types of decisioning like appraisal are consistent and accurate in every community."
National Mortgage News sat down with Chen to talk about how artificial intelligence is helping to create fairer standards of valuing homes and also how technology is set to change the appraisal space going forward. This interview has been edited and condensed.

How is artificial intelligence being used by Clear Capital?

Machine learning has been very prevalent within the industry for quite a while. At Clear Capital we embraced it back in 2017 as part of our automatic valuation model strategy. We found that the traditional statistical model approach to building AVMs was cumbersome. New models had to be created for every single major area of the country [and they had to be updated very frequently.] This problem really lended itself well to machine learning, which is a really effective discipline for doing accurate predictions.

We really like using machine learning to predict property values because it can be tested. We're testing against what the actual sale prices and appraised values of properties are as benchmarks and that can be measured to see how accurate we were in our confidence.

Another area that we've leaned into very early in terms of AI is the use of computer vision…We can really preserve that original foundation of the home by using computer vision to capture the home, scan it and then have consistent data points from that imagery that can be used to ensure consistent standards across the country.

Is the impact of artificial intelligence tools significant in reducing potential bias and building consumer confidence in real estate transactions?

You're trying to solve a national problem, but also make it accurate at a community level. These types of tools help us build a national standard for how this should be approached, and in the underwriting process, ensure that we're measuring the quality of the appraisal and the quality of the condition ratings against a repeatable standard.

We have an automated appraisal underwriting tool that we call AURA and that uses photo AI to read the photos in the appraisal and understand whether all the photos present. It's able to identify different rooms to know how many bedrooms and bathrooms are being represented in the photos and compare that data to what's in the appraisal and also compare to what is in the sketch or the floor plan to make sure all of that data is consistent.

We also are able to give a prediction on what is the actual condition of the property based on the model looking at the photos. That has been a process that's needed to be completely manual before with humans looking at photos and giving their best interpretation of what they're seeing in the photos. So from an efficiency standpoint, it really helps to have a machine be able to do that. But also it provides these checks and balances because condition is one of those areas that can be inconsistent community to community.

What are ways generative AI can be used in the appraisal space?

When you're in the world of generative AI and large language models, we're talking about the interpretation of language and the ability to generate content and imagery. Language is a big part of how we how we communicate what's happening on a particular property and what's happening with the market, providing effective summaries and information. If you're doing appraisal work these are all really interesting potential capabilities to be able to quickly summarize a lot of written information and language. I'm excited about the idea of understanding objective language versus subjective language because we're striving to ensure that the appraisal practice is more objective.

The CFPB has been skeptical of AVMs in the past and has proposed regulations to place safeguards around these types of systems to minimize bias/ blindspots of developers. Is this necessary? What issues can the industry run into if safeguards aren’t implemented?

The proposed rulemaking calls for accuracy and testing, ensuring there's no discrimination or disparate impact. These are all part of our testing process today. We've also seen these be adopted by a number of clients that we work with to ensure that there aren't issues with accuracy, consistency, or fairness. We welcome the idea of consistent testing standards where lenders can be confident in the results and confident in how to use the information given to them by different models.

How might human dynamics be impacted by more reliance on machine learning? Will the industry rely less on appraisers themselves?

I think we'll continue to see the progression of using technology to capture the home and moving the home from an analog format or physical format into a digital format. That trend is clear and providing a way that can be done more consistently without specialized training, I think is part of that…If we have improved data, then yes, we'll be able to improve not only models such as AVMs, but also improve the data that's used by humans to analyze and do appraisal work. There's more room for technology, but there's also room for a partnership.