Man and Machine

This article was first published in the Appraisal Buzz Magazine.

No doubt appraisers have been hearing the warning for the past couple of decades that Automated Valuation Models (AVMs) would someday replace them. You’ve heard all the recent buzzwords- big data, machine learning, regression and the platform revolution. Fear not! There is one very critical factor that the very smart quants have simply overlooked.

Appraisals and collateral risk analyses are part art and part science. As an appraisal professional, you can actually perform some tasks better than machines such as using your human senses and judgment. But when it comes to large dataset analysis, machines will win every time. If appraisers are going to remain viable, change will indeed happen, and in a very big way.

I think you are going to be witness to a giant leap in both the appraisal process and in the technology. Efforts to date have been around creating IVMs (Interactive Valuation Models) whereby appraisers are provided with regression modeling and comp scoring. While the motives have been in the right place, regression models simply don’t work very well especially with “big data.”

The next big thing is Artificial Intelligence or AI. The future has arrived. Machines really can think like humans. Intellectually it has hard to grasp at first, but the concepts should seem more familiar. The first concept is Classification. As an appraiser, you do that all the time. Next, comes Categorization. Think of how your “appraiser brain” works in filtering data from MLS, inspections, and public records, and how you file that in your brain. Machine Learning is where things get interesting. The knowledge you have gained from years of experience can all now be learned by a machine. It’s both scary and exciting at the same time. Collaborative Filtering is where science meets art in action. Human preferences for features and attributes of a home and market area can be measured quite easily.

Hopefully, this is all beginning to make sense. The exciting part is where man and machine intersect and complement each other. Below is a comparison of functions which separate man from machine:

Appraiser                                                                    Machine

Verify transactions with parties                                 Analyze big data

Report seller concessions                                           Cheaper

Observe and report age /condition                            Faster

Reconcile conflicts in data

Identify functional obsolescence

Measure (for now)

Identify non-arms- length transactions

Weapons of Math Destruction

There is a lot of chatter that machines are superior because they aren’t subject to bias. That is simply not true. AVMs, as we know them today, are mislabeled. They do not estimate market value. They are sales price models.

Before the mortgage crisis, AVMs were being actively used occasionally on originations but mostly on home equities, second liens, defaults and especially Mortgage Backed Securities (MBS) pools. AVM Cascades were developed in the 90s which offered multiple AVMs by different providers on a single platform using a waterfall algorithm to determine the “best” results. Lenders seeking not only a faster and cheaper alternative could also hit the number they “needed” every time. The AVM providers would be paid only on the “winning” AVM. AVMs were “juiced” like a New York Yankee at Spring Training. Those machines were subject to the worst of human elements, corruption. Algorithms are developed by humans and are subject to bias.

AVMs can scale up at a rapid rate potentially earning them the moniker of “weapons of math destruction” if they are deployed on a large-scale basis. You should be concerned if there is no independent third party monitoring the use of these models. Also, keep in mind that appraisers are licensed and subject to oversight. We are entering new ground here. Can machines be held accountable if their outputs are unreliable or at worst have been corrupted? Let’s hope that regulators employ experts in this burgeoning field rather than focus on relieving lenders of their regulatory burden. Their focus should be the “safety and soundness” of housing finance.

Intersection of Man and Machine

This is where it starts to get exciting. When you “mash up” the best of what appraisers do with the supercomputing power of a machine, everybody wins. The future of valuation appears to hinge on the humanization of “big data.” The hierarchy begins with data but ends with knowledge. In between is information. That only happens when there is human interaction with the local market expert, the appraiser.

In today’s world, we do not enjoy the benefits of a national real property standard. Each MLS system is unique in data formatting. There are over 3000 counties in the US each with their own methods and standards for collecting data for assessment purposes. For now, at least, appraisers are that filter that can verify, clean and scrub data and transform data into meaningful, actionable information.

Why AI Now?

Two forces are driving the AI revolution. One is computation power. The explosion in machine power has made feasible all kinds of breakthroughs in applications of artificial intelligence that were mere glimmers in the eyes of researchers for decades. AI technologies in use today were invented decades back, but we lacked the computation power to bring them to realization. Self-driving cars contain between 5-10 Tera Flops of power. ASCI White, the top supercomputer in June 2000, consumed 6 MW of power and weighed 106 tons and had less computing power than that contained in a self-driving car. The unique ability of AI algorithms to keep improving with increasing computing power distinguishes them from older statistical models. We believe this will keep yielding accuracy gains for a long time to come.

The second thing driving the AI revolution is the explosion of data brought about by the internet and the widespread availability of computing technologies. Today the equivalent of Netflix’s streaming library is being added every week to the world’s data store. This data provides valuable datasets for training AI algorithms. This data is referred to as the rocket fuel of the AI explosion.

How Does AI Impact the Valuation Process?

AI models with use of large data sets have exceeded human-level performance in several fields, e.g., Chess and other games, speech translation, image classification, facial recognition and so on. The next generation of Interactive Valuation Models (IVMs), with enough data sets and computing power, can produce estimates that exceed any AVM in reliability. The incorporation of daily satellite imagery, street-level imagery and of course local, state and national trends will improve accuracy to levels previously unheard of in the valuation space.

This, however, is not a question of human vs. machine. But rather human with machine. The nature of technological progress over the past 200 years has been for humans to take on more sophisticated tasks as the more mundane tasks get automated.

An IVM can approach human-level accuracy in its valuation for most standard properties, but given the nature of real estate where every property is unique, we will still need an expert human appraiser to verify the machine outputs and provide adjustments as needed. By definition, the nature of real estate is to be idiosyncratic, i.e., unique, and this makes valuation especially tricky.

The combination of man and machine holds the potential to streamline the process, speed up appraisals and provide appraisers a valuable unbiased ally in the current landscape. Think of it this way. Warren Buffett, Ray Dalio, and other world-class investors are using AI to manage billions of dollars. They do not rely simply on the models they build. They have teams of analysts who filter the outputs and apply human judgment. They rely on the experts.

What is the Solution?

Appraisers……you are; if you embrace new technologies.

While it may sound incongruous to talk about a return to fundamentals in a discussion of emerging technologies, that is precisely where the opportunities are. The three approaches to value have long ago been abandoned which has fundamentally destroyed the appraisal process. The appraisal of today is bloated, and the appraisers’ tasks focused on many of the wrong things. The current forms are antiquated and don’t ask the right questions.

Sales prices sometimes reflect irrational behavior. If we don’t inject rationality into the appraisal process, we will surely have yet another crisis ahead. The demand for faster and cheaper previously could only be satisfied by taking shortcuts. With access to “big data” and deep machine learning capabilities, leveraged by local market experts, exciting things can happen.

Interested in submitting an article? E-mail us at


About Ritesh Bansal, Joan Trice

Ritesh Bansal is the co-found and CEO of Appraisal. Appraisal harnesses the power of Artificial Intelligence (AI) and Big Data to produce intelligent valuation models which achieve human level accuracy. Ritesh studied Mathematics and Computer Science at Carnegie Mellon University and Economics at CUNY Graduate Center in NYC. Prior to Appraisal, he worked for 15 years in finical institutions developing quantitative valuation models for trading and risk management. Joan N. Trice is the founder and CEO of Clearbox, LLC, publisher of Appraisal Buzz, and host of the annual Valuation Expo, the largest conference for the valuation community. Joan also hosts the Collateral Risk Network, a members-only group of more than 500 dedicated chief appraisers, collateral risk managers, regulators, and valuation experts who are focused on resolving the many challenges facing our profession.

Check Also

In the Trenches, Crazy Appraiser Stories

This article was first published in the Appraisal Buzz Magazine. Are you receiving your print …