We recently spoke with Restb.ai Chief Product Officer Nathan Brannen on the subject matter of artificial intelligence (AI) identifying major risks in real estate appraisals, while he also shared thoughts on the types of appraisal errors that can result in repurchase requests.
“Repurchase requests occur when an appraisal is deemed to be inaccurate, misleading or non-compliant with USPAP,” Brannen told Valuation Review. “The most frequent errors are caused by inconsistent condition and quality adjustments, which can lead to over/undervaluation. In many cases, this is due to improper comparables being selected when other more recent sales would have been more appropriate. “
We then asked the Restb.ai executive how the significant changes regarding new appraisal forms affect the overall appraisal process.
Brannen responded by saying the transition to the new UAD 3.6 appraisal standard is the most significant change to the appraisal industry since the introduction of the 1004 appraisal more than 20 years ago.
“It is designed to be a dynamic report with many conditional questions that are only required based on a particular property’s relevant needs,” he added. “Instead of having numerous report types and forms, there is a unified format, the URAR, for all appraisal data to be submitted.
“The GSEs have also taken this opportunity to create a more structured format requiring more information as well as the opportunity to add comments in-line with the relevant sections of the report, rather than having everything added as addendums,” Brannen went on to say.
Furthermore, the chief product officer indicated that while more structured and detailed information is a key benefit, there will be an adaptation period as appraisers get used to the new requirements and new software designed for the new format.
“It will initially take appraisers longer to complete appraisals in the new format and there will be a spike in quality issues that will need to be caught throughout the entire appraisal lifecycle,” Brannen told us.
The Condition/Quality report and the concerns it addresses for appraisers is another key area to understand when it comes to AI in the appraisal transaction.
The recent Restb.ai condition/quality white paper focused on the reliability and consistency of value adjustments made for condition and quality.
“The GSEs have identified condition and quality discrepancies as the most frequent issues in appraisals, and in several recent high-profile bias lawsuits, the contested valuations were due to the inclusion of inferior comparables,” Brannen said. “This study analyzed thousands of appraisals with AI-generated condition and quality scores based on images of the subject and comparable properties. Notably, more than 33 percent of appraisals were identified as having a high risk of a missing or unwarranted condition or quality adjustment, which could result in a repurchase request, which could cost lenders as much as $32,228 per loan.
“With the new appraisal standard requiring a more granular analysis (i.e. breaking out interior and exterior condition and quality), providing a more robust and efficient way to catch and prevent these valuation errors is essential,” he added.
The challenges in consistent condition and quality analysis as it relates to properties being in varying states were also outlined for us.
Unlike living area, lot size or room counts, Brannen pointed out, condition and quality are inherently subjective assessments.
“A property may have a newly renovated kitchen but outdated bathrooms, yet appraisers are required to look at properties holistically and provide an overall score. With upwards of 80 percent of properties rated a C3 or C4 and upwards of 90 percent a Q3 or Q4, the lines between categories are blurred and expecting thousands of appraisers to all agree is wishful thinking,” he said. “It gets even trickier for AMCs and lenders reviewing appraisals. The magnitude of adjustments for condition or quality vary immensely and can be difficult to validate. While an appraisal includes a comprehensive view of the subject property, it only includes a single exterior photo for each comparable. To truly confirm an adjustment makes sense, a reviewer would need to use their MLS or a portal to do a deeper comparison of the entire property.”
As a result, risky appraisals slip through the cracks because it is simply too labor intensive and costly to verify condition and quality differences, Brannen concluded.
But why will AI provide a solution with regard to automating this review process?
The answer lies in AI enabling objective, consistent and granular analysis of each appraisal.
“Using computer vision, imagery of the subject and comparable properties can be analyzed instantly to highlight areas where adjustments may be unwarranted or missing,” Brannen told us. “Additionally, AI can provide decimal-level scores (i.e. a C3.6 or a C4.4) and even break out the scores by area (e.g. exterior vs. interior or kitchen vs. bathrooms). It eliminates the ‘stare and compare’ of every property and allows the reviewer to focus on the most important aspects of each report.”
Finally, the company executive shared how this technology aligns with the appraisal modernization goal of a more transparent, efficient and consistent process and how it can be enhanced by AI. Examples include instantly prepopulating appraisal details based on insights captured in photos, optimizing comparable selection by ranking recent sales by condition and quality and automating quality checks by ensuring photos support property details and comparable adjustments.
“Fannie Mae has already noted that appraisers who embrace this technology will have a competitive advantage, so beyond being compatible with appraisal modernization, AI is an essential building block for achieving it,” Brannen said.