Shaping The Appraisal Process With AI

In reviewing over 300,000 assessed in-person appraisals, the Mortgage Outlet found 39% had data inaccuracies. At the same time, a more pernicious issue can occur when appraisers use their considerable power in the mortgage origination process to intentionally inflate valuations, which is a form of mortgage fraud.

As an example, eAppraiseIT conducted more than 260,000 appraisals for a large U.S. bank. Ultimately, the partnership led to litigation. It was stated that the bank’s loan officers pressured the company to select certain appraisers, leading to inflated property valuations and more mortgage business for the bank that subsequently failed. The settlement included $4 million in civil penalties and $3.8 million in costs from litigation. While the monetary impact in this case was relatively small, the incident bruised the reputation of the bank and eroded confidence in its practices.

These types of inaccuracies can lead to losses for mortgage banks during periods of economic contraction and increased defaults. The risk is even greater in situations where there are intentional data discrepancies. Over-inflating home appraisals is a form of mortgage fraud that could lead to serious legal implications and significant financial consequences. Guidelines vary by state, however, overinflated and inaccurate appraisals can expose lenders to default risks, regulatory penalties, and financial risks.

After the financial crisis, provisions that require appraisal independence changed many aspects of the appraisal process. These regulations were implemented to prevent lenders and other parties from influencing the appraiser’s valuation and to ensure the appraisals are objective and accurate. However, the underlying process still requires local presence, making it difficult for lenders to monitor and oversee the appraiser’s work.

Bias In The Traditional Appraisal Process

Ensuring that the appraisals are accurate and bias-free is essential to current and prospective homeowners of all races and income levels. However, given the subjectivity of the traditional appraisal process, racial or ethnic bias may occur, whether intentionally or not. Lenders have suffered reputational and financial penalties over the years due to discriminatory practices. According to a 2021 Freddie Mac study, homes in Black and Latino neighborhoods are significantly more likely to appraise lower than the contract sales price than those in White areas. Studies like the one Freddie Mac conducted have shed light to this bias that exists in the appraisal process.

Identifying and exposing bias in the appraisal process is critical to obtaining a fair valuation for minority homeowners. Although it’s required that the appraiser document the factors within the appraisal report used to value the property, there are still ways for appraisers to omit or misinterpret information intentionally and unintentionally. Despite the appraiser’s responsibility to ensure an accurate valuation, the lender is ultimately accountable for the lending decision. It is still exposed to reputational, financial, and regulatory risk if the value is deemed inaccurate.

Technology Inefficiencies

To address the shortcomings of traditional appraisal methods, many lenders use Automated Valuation Models (AVMs) in their appraisal process, a practice that has only become more common over the past few years. AVMs use advanced analytics to evaluate the available data on a property and predict its value.

While AVMs can provide a quick estimation for prequalification or appraisal comparison –sometimes even used as the only appraisal in certain scenarios - there are drawbacks to this technology. AVMs have a high maintenance burden, as the models are highly complicated and require updates to keep up with evolving market conditions. Since these models rely on complex calculation methods with numerous variables, lenders often look for third parties with expertise in building and maintaining these models.

While third-party providers offer a convenient solution, lenders who rely on these vendors often cannot fully understand or explain how valuation decisions were made to consumers. This opacity may mask issues like inaccuracies in inputs, or ineffective calculation methods that may not be readily apparent, posing potential regulatory issues if the lender cannot explain their lending decision. Last, some data has shown AVM tools can actually perpetuate and exacerbate bias issues. According to a study done by Urban Institute in 2020, AVMs in majority-Black neighborhoods produce larger errors, as a percentage of the underlying sales price, than AVMs in majority White neighborhoods. This is a weakness users of AVMs should be clear-eyed about. While machines should be able to process valuation data faster and more accurately than humans, these technologies are still far from perfect. Without proper governance, controls and data reviews, AVMs may put banks at risk of creating systematically over- or under-valued estimates.

Opportunities

With several systemic discrepancies in the appraisal process, there is a great opportunity for mortgage companies to improve their processes.

AI-Enhanced Risk & Control Framework

Regulatory pressures are unlikely to lessen, even in a tightening mortgage environment. With human error and the complexity that may be present in valuation models, lenders should seek additional risk and control measures that improve oversight and transparency into the valuation process. Artificial Intelligence (AI) can be used to analyze large amounts of data to identify potential bias, under- or over-valuation, as well as using “Learning and Adaption” to identify outliers and emerging risks. This allows lenders to focus their resources where they are needed most and improve the overall accuracy of valuations.

Leveraging AI to automate routine tasks can free reviewers to focus on more high-value and complex tasks. Using the AI model, lenders can evaluate the appraised value and provide alerts on areas of concern within the appraisal. This allows the underwriter or appraisal review team to increase their focus on potential issues that need to be addressed.

However, AI effectiveness relies largely on the quality of the data and the algorithms used and if not assessed thoroughly, it has the potential to mimic and even perpetuate human and societal bias. Therefore, lenders must establish robust data governance and validation processes to ensure that the data used for the analysis is accurate and reliable.

Appraiser Monitoring

Creative technology investments can give lenders the ability to have meaningful oversight in the mortgage origination process by creating frameworks to monitor appraisers’ behavior. Creating analytical models – potentially leveraging AI tools -- to monitor and compare appraisers can help identify overvaluation patterns and/or bias that may expose the lender to financial and reputational risk.

Lenders should consider creating robust dashboards to evaluate and compare appraisers on overall performance including quality, service and cost. Aggregating these trends and metrics is a powerful tool for lenders to assess the integrity and performance of the appraisers/appraisal management companies (AMCs). The ability to monitor appraiser activity may help banks identify appraisers who have shown instances of bias or overvaluations in their appraisal reports.

Lenders must consider an end-to-end process covering everything from monitoring, review, and enforcement and the specific data they have available to establish baselines and compare appraisal performance. Policies must clearly define the lender’s processes and procedures to add an appraiser/AMC to their panel, reconsider a value, or remove an appraiser/AMC.

Lenders must treat appraisers/AMCs consistently based on defined standards with supporting documentation to support corrective actions. With AI’s ability to evolve and improve over time, lenders have a powerful tool to continuously improve appraisal accuracy and service levels while reducing valuation risks.

AVM Selection/Data Selection

Banks should consider re-evaluating their appraisal automation tools on a periodic basis to maximize the efficacy of estimates and minimize risk. Thorough model and vendor reviews can evaluate the data elements available vs. those being used and determine how banks can improve the accuracy of model outputs, as well as identify systemic bias. Thoughtful reviews should include model competition to determine which models and methodologies perform best under a variety of conditions and geographies.

Conclusion

The mortgage origination industry is confronting some concerning headwinds in 2023. Addressing the shortcomings of the appraisal process is a challenging but critical step to helping protect lenders not just survive but thrive. Fortunately, the available AI tools may be part of the solution. They can increase productivity and speed and extract insights from data that are not otherwise humanly possible at scale. By leveraging these tools, establishing strong policies, and implementing robust, multi-dimensional scorecards to monitor appraisers/AMCs, banks can maintain confidence in valuations used in mortgage underwriting.

Additionally, mortgage bankers can establish periodic reviews of AVM models and the risk and control frameworks around them. These measures enable banks to reap the benefits of a more profitable book of mortgage business. At the same time, they can ensure that they are doing the right thing by their homeowners by providing accuracy and avoiding the unconscious and conscious bias that can emerge when individuals make decisions in a vacuum.

Vicki Bott and David Stadler are both engagement directors with Reference Point, a technology consulting firm.