AI and Real Estate Part II: When It Comes to Employment, AI Is More Friend Than Foe

The concern about AI causing unemployment is nothing new: For hundreds of years, technological advances have provoked anxiety about mass unemployment and harm to society that turned out to be unjustified. While the Luddites, skilled textile workers in the early 19th century who protested against the introduction of power looms and knitting frames that they believed would cause unemployment and worsen working conditions, are the most famous of these critics, the history of anti-technological sentiment goes back much farther than that. Around 1589, an English clergyman named William Lee invented the stocking frame, a device that assists in knitting: He soon asked Queen Elizabeth for a patent, which she refused out of fear that his device would put weavers out of work. Even the earliest steps along the path to industrialization were opposed by people who feared its destabilizing effects on society.

Historically, technological improvements have increased both productivity (and in turn wages) and employment. The unemployment effects have been transitory, and in line with what I said above, no one has ever argued for banning a technology to remedy unemployment. If rolling back technology to cure unemployment is a bad idea, why do so many consider stopping technological improvement to prevent unemployment to be a good one? Obviously, this is a case of cognitive bias: We by our nature fear the unknown and question the utility of changing something that already works. However, there is a sense in which the questions are not all that different.  While new technologies have led to new forms of employment, it is easier to see how machines can displace workers than how they might produce new jobs: Discovering how labor can complement a new technology is, as my colleague John Schuler noted, an essentially entrepreneurial activity while seeing how the machine slots into the task a human previously performed requires much less creativity. This is also part of the reason no one proposes technological rollback as a remedy for unemployment: Once entrepreneurs have shown how labor can be reallocated, people become more aware of the jobs the new technology created than they are of the jobs it replaced. Indeed, in those sectors where AI becomes a tool that supplements labor, we can expect wages to rise—just as they do in other sectors. Who makes more, a man with a shovel or the man who operates the steam shovel? Similarly, who is more productive, a human with an AI assistant or that same human without the assistant? Technology overall increases productivity and, thereby, wages. Its effects might not always be linear; some may experience wage declines, but the wage-increasing effects overwhelm its wage decreasing ones.

The general reaction I get to this is: Yes, that is quite nice, but AI is different. Those machines are all supplements to human labor, but AI is a direct replacement—more importantly, AI replaces that which is most unique about humans, their intellectual capacity. Of course, there are many machines that directly replace human labor. However, even if we grant the premise and assume that AI will eventually produce robots that are better than humans in every way, there is still reason to believe human labor would persist: After all, there are humans that are better than other humans in every way—economically speaking: We can think of an Ivy League record-breaking athlete; such a man may be better than a man of merely normal intelligence and physical prowess: Yet, because of the theory of comparative advantage, these two could gain by trading with each other.

Consider the example of two men, one 35 and the other 75, who live next to each other on virtually identical plots: John and Dan. Each of these men weeds and mows his own lawn, but John is much more efficient at both tasks.  Yet can John and Dan gain anything from trading with each other?

So, can John and Dan make a trade that will reduce the amount of time each has to spend gardening even though John is better at both weeding and mowing? The answer is, of course, yes. While John is better at both tasks, he has a comparative advantage in weeding, he can weed two gardens in the time it takes him to mow one while Dan has a comparative advantage in mowing as he can only weed 1.2 gardens in the time it takes him to mow one. If both men want to save time, one possible exchange would involve John weeding his whole lawn and half of Dan’s while Dan mows his lawn as well as half of John’s and weeds half of his lawn. Let’s compare the total time spent between this exchange and the autonomous situation where each person mows his own lawn.

If each person weeds and mows his own lawn, John spends 120 minutes, and Dan spends 220. If they exchange, John spends 40 + 20 mins weeding and 40 minutes for a total of 100 minutes mowing while Dan spends 50 minutes weeding plus 120 + 60 minutes mowing for a total of 210 minutes. The exchange can save John 20 minutes and Dan 10 minutes. Now, why would this logic change if John owned a robot that could perform these tasks instead of performing them himself? (Or perhaps the robot is autonomous and makes this exchange himself).

A counter is that that is all well and good, but what happens when robots/AI become so inexpensive that it is cheaper to hire an AI than a person—or the cost of the AI is lower than the cost of feeding, clothing, and housing a human being? After all, hardware becomes less expensive over time. This argument, however, ignores Say’s Law. If AI is this productive, you can expect demand, as measured in terms of real goods (let’s say you track the goods a person making a true market minimum wage and working 35 hours a week can buy), to go up. Of course, the effects of greater productivity on labor demand are harder to think about than the analogy of simply being undercut by a “better and cheaper laborer.” Comparative advantage and the greater productivity this technology would create are easy things to forget about.

Before turning to the real estate sector in detail, we should deal with one possible concern. People reading this post might declare that brighter minds than mine have argued for the dystopian unemployment effects of AI. For that reason, a brief review of the literature might help clarify matters. The following literature review will demonstrate that even the AI pessimists agree that the unemployment effects tend to be largest in fields that involve non-interpersonal work performed by males with medium amounts of education; even if the pessimists are right, real estate is not likely to be seriously hit. 

Studies Favoring AI (Arguing in Favor of No Significant Unemployment)

1. Georgieff and Milanez (2021)

Analyzed jobs considered at high risk of automation in 2012 and found that while these occupations experienced weaker employment growth and sometimes modest declines, automation did not lead to net job destruction at the aggregate level.

2. Felten et al. (2019)

Proposed an “AI Occupational Impact” measure and found that AI had a small positive effect on wages but no impact on employment, indicating that AI adoption did not result in significant job losses.

3. Aghion et al. (2020)

Utilized French data and discovered that automation had a positive impact on employment at both the firm and industry level, reinforcing the idea that AI does not lead to widespread unemployment.

4. Koch et al. (2021)

Analyzed Spanish manufacturing firms and found that robot adoption led to a net increase in jobs, while firms without robot investments suffered job losses, suggesting that AI adoption promotes job growth.

5. Dauth et al. (2018)

Explored the effects of robot adoption on the German labor market and found that it changed job distribution across industries without decreasing the aggregate level of employment.

Studies Opposing AI (Arguing in Favor of an Unemployment Effect)

1. Acemoglu and Restrepo (2020)

Analyzed the US labor market and found that one additional robot per thousand workers correlated with a slight drop in the employment-to-population ratio and wages, particularly affecting lower-skilled workers.

2. Chiacchio et al. (2018)

Employed a similar methodology across six European Union countries and estimated a negative impact on the employment-to-population ratio, with a particularly strong displacement effect on young males with medium levels of education.

3. Autor and Salomons (2018)

Investigated the effects of robotization on the labor share and concluded that automation contributed to its decline since the 1980s, potentially suggesting a connection between automation and certain labor market changes.

There is a tendency even within the pessimistic literature to see the effects as non-catastrophic and limited to sectors that do not entail much interpersonal interaction, sectors that are nothing like real estate. 

When ATMs were introduced in the late 1960’s, there was fear that they would put human tellers out of work. However, the opposite was true: In a 2015 article, James Besson of Boston University School of Law explained that ATMs actually made it feasible to open more branches by reducing the costs associated with operating a branch. These additional branches lead to more tellers being hired in total—even if there were fewer tellers at each branch.

Having dealt with the question of AI and employment generally, let’s turn our attention to the real estate sector. Real estate is an expensive commodity—and the more expensive a commodity is the more reluctant people will be to automate the process of selling it. Humans are still deeply involved in the process of selling cars: Until we see car salesmen replaced by AI, realtors and appraisers can rest assured that there will be demand for their services.

Real estate workers are often required to switch from task to task in unexpected ways. For example, a realtor must decide how the house should be staged, what pictures should be posted on the MLS, how long to keep the property on the market, and must handle a variety of customer interactions. Loan officers similarly must manage a wide array of relationships. Juggling this many tasks and providing people with satisfying personal interactions is still well beyond the capabilities of AI.

That said, I do see AI altering the real estate employment landscape in a number of ways. Let’s consider the major categories of real estate worker each in turn.

Realtors

I don’t see online platforms pushing realtors out of the business. AI might make platforms like Redfin and Zillow a little more user-friendly, but it is unlikely enhanced chatbots will ever make someone comfortable with listing his home on one of those platforms when he wasn’t before. Rather, I see AI as being a benefit to realtors who can adopt it while hurting the income of those who aren’t able to utilize the new technology. Those realtors who embrace AI will benefit from the following:

Enhanced Customer Insights

AI can analyze vast amounts of data to provide real estate agents with detailed customer insights, such as preferences, behavior patterns, and affordability, enabling agents to offer personalized and targeted services to clients.

Automated Lead Generation

AI-powered tools can help agents identify and qualify potential leads more efficiently, saving time and resources in prospecting efforts.

Predictive Analytics

AI can assist agents in predicting property value trends, demand patterns, and market fluctuations, aiding them in making informed recommendations to clients. They can also help them make informed decisions about how long to leave properties on the market, etc. For example, one could use an AVM that includes days-on-market to discover how long the property should be listed to maximize the sale price.

Virtual Tours

AI-driven virtual tour technologies can enable agents to showcase properties remotely, offering clients an immersive experience without the need for physical visits.

Appraisers

Appraisers face a similar situation but with more of a downside: Those appraisers who can embrace new technology will be able to massively increase their productivity as the process becomes bifurcated—with lower-cost workers performing the inspections and taking photographs while appraisers produce an evaluation by looking over those photos from an office where appraisal software aids their evaluation process; they will be paid less per valuation but they will be able to perform many, many more. Those appraisers who cannot embrace new technology, however, will find less and less work over time. With confidence scoring and assignment complexity scores determining which appraisals can be handled by an AVM and which require a human touch, appraisers will only be needed for about half of valuations. That said, with the average age of an appraiser running between 49 and 55 depending on the source, the older ones may respond to the pressure by retiring earlier than planned—as they are the ones who are least likely to adapt. This means younger appraisers might actually see their standard of living improve. Their overall earnings will increase and the work itself will become easier as AI aids them in the following ways.

Comparable Property Search

AI can help appraisers by sorting properties based upon their suitability as comps—easing the search for properties that have particular characteristics. Loan officers often ask for additional comps to “bracket the subject” or that have features in common with the subject that the top comps don’t have; an example of bracketing the subject would be adding a comp that has a larger lot than the subject if the three best comps all happen to have smaller lots.

Guidance Around Adjustments and Other Aspects of the Appraisal 

AI-based AVMs should allow appraisers to provide better, more market-driven adjustments than they have in the past. Indeed, many appraisers would learn an adjustment factor when they were appraising, let’s say for gross living area (i.e. square footage that is “above grade” meaning not a basement; there are other complexities here, but this gets the idea across), of 75 dollars a square foot and apply that factor throughout their career despite massive home price appreciation.

Image Analysis 

Image analysis can help appraisers identify issues that they, or the inspectors, might miss. They can also help appraisal review by flagging appraisals that have unaddressed issues like mold, differed maintenance, and other issues.

Enhanced QC Efforts 

AI will be able to create increasingly efficient quality control processes: As AVMs become more accurate and fraud detection algorithms are refined, more and more defective appraisals will be detected without requiring costly human reviews. That said, some number of random reviews will still be necessary to ensure that people have not discovered ways to outsmart the AI.

Fewer Low Complexity Assignments

AVMs should be able to handle an increasingly large percentage of low-complexity assignments where the homes in question have many strong comparables and no unusual features. Appraisers will focus on more complicated assignments. This should result in lower turn times for the industry.

Loan Officers

Home loans are, by their nature, expenses that require just as much due diligence as the home purchase itself: Both on the part of the borrower and the lender. As such, we can expect loan officers, as well as underwriters, to remain a key part of the process.

Personalized Loan Products

AI can analyze customer profiles and preferences to recommend personalized loan products that best match borrowers’ needs and financial situations.

Efficient Lead Generation

AI-powered algorithms can identify potential leads and prospects based on specific criteria, enabling loan officers to target their marketing efforts more effectively.

Streamlined Applications

While much of this will rely on data standardization and more traditional computer science techniques, we can expect loan officers to focus more on human interactions and less on paperwork.

Underwriters

Automated Data Analysis

AI algorithms can quickly analyze vast amounts of data, including financial records, credit histories, and other relevant information, to assess risk and determine the insurability of applicants. This automation can significantly speed up the underwriting process by flagging matters of concern without the underwriter having to spend as much time on “fact finding.”

Natural Language Processing (NLP) 

NLP allows AI to analyze unstructured data, such as text in medical records or customer applications, enabling underwriters to extract essential information quickly and accurately. This should increase their efficiency. That said, pressure to close loans quicky is increasing, esp. given the uncertainty of interest rates, so it is likely that lenders will prefer faster and deeper risk analysis to simply employing fewer underwriters.

Fraud Detection

AI can assist underwriters in detecting patterns of fraud or suspicious activity, reducing the risk of issuing policies to fraudulent applicants. This logic can alert underwriters to fraud before the loan closes: Saving lenders money in the long run. That said, AI will “cry wolf” often enough that it won’t replace the underwriters outright.  

Enhanced Due Diligence

AI can perform in-depth due diligence on borrowers and properties, cross-referencing data with external sources to ensure accuracy and compliance with regulatory requirements. This means less manual checking for the underwriter.

Artificial intelligence promises to enhance productivity in the real estate sector and the economy more generally. It will have some mild and short lived disemployment effects, but the real estate industry will be among the least seriously affected because of the high costs and complexities associated with real estate transactions. High volume, low-value tasks are much more likely to be replaced. 

Franklin Carroll

Franklin Carroll is Kukun's VP of Modelling and Analytics; After finishing his studies at the University of Chicago, he worked at Fannie Mae where he helped build Collateral Underwriter, roll out the Value Verify initiative, and specialized in collateral related and credit related policy.

Published August 13, 2023