Housing Industry Innovation: 5 Ways AI Can Help Boost Supply and Affordability

As the use of artificial intelligence (AI) grows more ubiquitous, the housing industry is looking to these new tools as an opportunity to improve housing access, reduce costs, and speed up housing production. At the same time, both states and the federal government are considering regulatory measures to safeguard against unintended consequences stemming from AI, while also fostering its potential benefits. In this explainer, we examine how AI is changing five areas related to housing: predevelopment, construction, creditworthiness, home appraisals, and property tax assessments.

To learn more about the basics of artificial intelligence, see BPC’s AI 101 explainer.

AI and Predevelopment

During the predevelopment phase of housing projects, engineers, designers, and architects can leverage generative AI, a subset of machine-learning technology. The longer a project takes, the more costs can rise. AI tools offer automation capabilities, streamlining various predevelopment tasks and potentially cutting down on both time and expenses. In addition, several AI tools focus on swiftly generating multiple design options, as well as more sustainable designs by taking into account local climate, energy usage, and available building materials. Generative AI can learn from prior projects to help project managers create realistic development timelines.

When selecting a building site for a residential property, developers must analyze whether the site can support their proposed use. Compliance with zoning and land-use regulations is critical. Attempts to rezone a site can lead to lengthy delays or even stop a project from moving forward. To avoid these roadblocks, developers can use AI tools to assess their project’s adherence to local zoning requirements and building codes. Local governments are also beginning to take advantage of AI to assist with property zoning and land-use efforts. For example, the Los Angeles City Planning Department uses AI to analyze zoning data and identify potential issues as it reviews development applications.

Combining AI with modular and prefabricated housing development processes can also benefit developers of affordable housing. In California, one AI startup partnered with an affordable housing developer to more efficiently complete several tasks, including the planning, design, unit configuration, and approval of their modular affordable housing project. Another partnership between an AI software company and modular housing manufacturer cut down the design process timeline for a mixed-income property by six months by generating thousands of design options within hours. Design and planning account for an estimated 5%-10% of total project cost for developments in the Bay Area, where this project was located.

Policymakers and researchers can also deploy AI to analyze and mitigate restrictive zoning and land-use rules, which are often significant barriers to increasing housing supply in high-demand regions. Harnessing AI for the collection and analysis of zoning and land-use requirements nationwide could lay the groundwork for meaningful policy reforms. Currently, there is no centralized, standardized, comprehensive dataset on zoning across the United States, which can make it challenging to evaluate the impact of zoning on housing supply and affordability. Although some states have attempted to build out a national dataset by contributing to the National Zoning Atlas, the development and maintenance of this resource requires significant effort. Researchers recently attempted to utilize AI to collect and digitize zoning laws. Despite the limits to what AI could accomplish, researchers saw benefits of a hybrid approach, combining human review with automated techniques.

Outlook: For machine learning to be effective at the project level during the predevelopment phase of housing development, and more broadly for zoning and land-use innovations, data inputs must be accurate, up to date, and in compliance with current regulations. AI tools cannot completely replace human involvement in zoning and land-use efforts, at least in the near term.

AI and Construction

Although estimates vary, the United States has “underbuilt” housing by millions of homes over the past two decades, contributing to today’s housing affordability crisis. AI is poised to help homebuilders build more homes and close the supply gap by significantly streamlining the construction process.

Homebuilders can leverage AI by employing drones, cameras, and mobile robots to analyze footage, monitor progress on construction sites, and confirm in real time that projects are aligned with design plans. During construction, AI can analyze historical project data on safety to preemptively flag potential hazards on work sites. This risk mitigation strategy might lead to more affordable insurance options, which currently constitute up to 10% of a project’s cost. Developers in New York City recently utilized AI during the construction of a high-rise, mixed-use development, using helmet-mounted cameras to automatically review site data and generate real-time progress reports with accompanying visuals for the project team. The AI tools helped reduce costs by improving labor and project management efficiencies, and the AI platform won awards for its ability to do so.

AI can bring greater efficiencies to the construction industry as well, with technological innovations reducing construction costs. AI’s ability to streamline processes and enhance safety and compliance can further improve the construction process and lower costs. The construction industry is currently ranked second to last in terms of industry digitization in the United States, according to McKinsey Global Institute’s digitization index. In addition, AI could help attract new talent to the field and help to address a lack of skilled construction workers, a barrier to expanding the housing supply. AI’s integration will enhance many construction-related technologies, such as drones and robotics, potentially creating job opportunities, particularly for younger workers seeking opportunities in emerging fields.

Outlook: Like AI’s usage in predevelopment tasks, to improve construction processes, AI must rely on accurate, up-to-date data that complies with all relevant laws and regulations. AI could generate erroneous or nonsensical results, referred to as “hallucinations,” necessitating human oversight. Not all developers, particularly small firms or those with limited resources, may be able or desire to incur the cost of AI products. Similarly, a lack of knowledge may be a barrier until AI technology is more widespread. To take advantage of any AI products they purchase, companies will either need to hire personnel with AI expertise in-house or find contractors with that knowledge.

AI and Creditworthiness

Each year, mortgage lending and the housing finance system that supports it give millions of people the ability to purchase homes. Lenders evaluate borrowers’ creditworthiness and make loan decisions based on several factors, including credit scores. Credit scores are calculated using information in a borrower’s credit history and provide a simple, standardized measure of how likely it is that a borrower will repay a loan. Those without a credit history—meaning no reported history of activities, such as credit card usage, student loans, or car payments—likely do not have a credit score and are therefore much less likely to qualify for a mortgage. Lenders see individuals meeting the mortgage eligibility criteria but holding a lower credit score as higher-risk borrowers and will likely give them a higher interest rate, increasing the cost of the borrower’s mortgage. Traditional credit assessments have been shown to provide less accurate information about Black, Hispanic or Latino, and low-income borrowers, and they are unable to provide information on individuals who do not have a traditional credit history (sometimes referred to as the “credit invisible”). In 2015, the Consumer Financial Protection Bureau (CFPB) found that 1 in 10 adults in the United States lacked a credit score.

Incorporating AI into these calculations can help increase the accuracy of credit ratings, mitigating the risk of losses for lenders while also broadening access to credit for more individuals. In fact, AI is already being used to incorporate additional information beyond traditional credit scoring components to better evaluate creditworthiness. These new types of data include on-time utility or rent payments and, in some cases, personal financial habits via bank transactions. At the end of 2022, the Federal Housing Finance Agency (FHFA) approved two new credit scoring models—FICO 10T and VantageScore 4.0, both of which integrate rent and utility payment data—for use by Fannie Mae and Freddie Mac. Both Fannie and Freddie will require lenders to furnish credit scores based on these new models by the end of 2024.

AI and Home Appraisals

A home appraisal involves an assessment to determine its fair market value, taking into account such factors as the property itself and recent sales prices of similar homes. Conducted by third-party appraisers, appraisals are most often done as a condition of receiving or refinancing a mortgage. Generally, prospective homebuyers bear the expense of the appraisal, which averages around $400. When someone applies for or refinances a mortgage, an appraisal ensures that the value of the home is adequate to serve as collateral for the loan. Differences in accuracy can exist between appraisals for a home purchase and those for a refinance; homes assessed at the time of a sale are more likely to be appraised at or above the sales value because the human appraiser is aware of the price for which the property is being sold, which can influence the valuation.

The selection of comparable properties, often called “comps,” for an appraisal represents an area vulnerable to human bias. Conducting accurate appraisals becomes more challenging when recent nearby sales are limited, as that limits the number of comparable properties available to determine the appraised value. Over the past three years, home sales have slowed significantly, exacerbating this problem. Additionally, about 97% of home appraisers are white, raising concerns regarding the potential of inherent bias in the process against nonwhite property owners. Another concern is that a white appraiser might have a limited understanding of neighborhoods made up primarily of nonwhite residents, which can lead to inaccurate appraisals. Research from Fannie Mae and Freddie Mac has found some evidence of appraiser bias against Black- and Latino-owned homes, though research from the American Enterprise Institute indicated that widespread explicit discrimination based on race in appraisals was unlikely.

Automated valuation models (AVMs), which leverage large datasets to calculate a home’s value, have brought AI into the home appraisal process. AVMs that rely on statistical analysis to calculate a home’s value have been used for decades, but recent advancements in machine learning and larger data sources have improved their accuracy and increased their usage, including through consumer-facing AVMs available on websites such as Zillow or Redfin. In addition to those available on public websites, both lenders and appraisers use AVMs: Lenders rely on AVMs to estimate loan-to-value ratios and provide mortgage quotes, while appraisers can use AVMs in conjunction with traditional appraisal methods. Historically, Fannie Mae and Freddie Mac have required in-person appraisals for mortgage loans they purchase, leading most lenders to adopt the same requirement, though that is beginning to change. AVMs are more likely than human appraisers to give accurate valuations in the case of purchase transactions, while research finds mixed results when appraisers are unaware of a home’s sale price, as in the case of refinance appraisals.

While there is always the potential for bias when using AI, it can be easier to detect and correct than unconscious bias exhibited by individual appraisers. The Urban Institute found no consistent trend of models undervaluing or overvaluing homes in majority-Black neighborhoods when compared with majority-white neighborhoods. However, it did find that the models’ appraisals of homes in majority-Black neighborhoods were less precise than appraisals of homes in majority-white neighborhoods. Ultimately, AI models’ objectivity relies on the quality and representativeness of the data they utilize.

Most regulation of the appraisal process happens at the state level, but with the increasing prevalence of AVMs, there is mounting interest in promulgating more standardized federal regulations. An Interagency Task Force on Property Appraisal and Valuation Equity (PAVE), launched in 2021 by the Biden administration and consisting of representatives from 13 federal agencies, developed recommendations to reduce racial bias in home appraisals. Its recommendations included establishing a nondiscrimination quality control standard through AVM rulemaking and creating a working group focused on researching possible bias in AVMs.

Outlook: A combined approach of human appraisers working with AI tools will likely yield the most accurate and efficient results as opposed to relying solely on either method. AMVs can use larger amounts of data than would typically be feasible, resulting in more accurate estimations of a property’s market value. Incorporating AVMs can also help address the problem of a declining number of appraisers and allow for a quicker appraisal process. However, such appraisals might reduce transparency into how an appraiser arrives at a valuation.

AI and Property Tax Assessments

Similar to appraisals, property tax assessments estimate the value of a property, enabling the calculation of property taxes owed annually to state and local governments. Property taxes constitute the majority of most local governments’ annual revenue and fund essential public services, such as public schools, infrastructure, and emergency services.

Each jurisdiction conducts property tax assessments differently, with some areas requiring updated assessments each year and others requiring new assessments every few years. Tax assessments tend to be less accurate than appraisals because they rely on less information. Moreover, government assessors cannot physically visit every property in the city or county annually for evaluation.

Inaccurate tax assessments can directly affect both homeowners and local governments. An overvalued property can require a homeowner to pay a higher-than-anticipated tax bill, while an undervalued property can lead to lost tax revenue. According to the Brookings Institution, Black-owned homes are more likely to be over-assessed than white-owned homes, resulting in higher tax burdens. Increased accuracy in assessments could reduce this disparity.

Wake County, NC, transitioned recently from conducting assessments every eight years to a four-year cycle, doubling the workload for local assessors. To manage the increased workload, the county partnered with a local company to leverage an AI tool to more quickly validate property tax assessments and identify neighborhoods that require additional review. Appraisers rely on the AI tool to provide an objective analysis of each property’s value using machine-learning models that can quickly and accurately account for numerous variables.

Outlook: In addition to reducing the workload for local governments, employing AI to aid in property tax assessment calculations makes it possible to utilize more data and conduct more frequent property tax assessments, and it could increase their accuracy.

Looking Ahead

In October 2023, the Biden administration issued a landmark executive order on the use of artificial intelligence across federal agencies. The executive order aims to establish standards for safety and for privacy protection, encourage the use of AI to advance equity and civil rights while protecting consumers and workers, and support innovation. The order includes calls to action for more than 20 federal departments and agencies, including the Department of Housing and Urban Development, which will issue guidance in 2024 on the use of AI in housing-related actions.

Congress is also assessing the risks and benefits associated with the increased adoption of AI across industries. Both chambers have prioritized hearings on the topic, with quite a few bills introduced, though none has passed to date. In the Senate, a bipartisan group, led by Majority Leader Schumer (D-NY), hosted a series of nine AI Insight Forums at the end of 2023. These forums supplemented traditional committee hearings and gave senators an opportunity to speak with experts on the state of AI, including its potential in housing. In the House, a bipartisan Task Force on Artificial Intelligence made up of 24 members was announced recently. The task force plans to work with relevant House committees to develop a series of “guiding principles” around AI legislation, as well as concrete policy proposals. Meanwhile, the House Financial Services Committee has formed a bipartisan working group on AI to examine the technology’s impact on the financial services and housing sectors, assess the applicability of existing regulations to AI, and evaluate the need for new regulations.

AI presents a host of new opportunities for innovation across the housing industry. From reducing project timelines and costs, to improving construction-site safety and home appraisals’ accuracy, to increasing access to credit, the potential is significant. Nevertheless, policymakers must address numerous challenges to ensure the safe and effective utilization of this new technology for the benefit of all Americans.