Appraisals and ValuationsServicing

Automated valuations models: A modern assembly line

How AI and machine learning changed AVMs

In today’s housing industry, logistical efficiency and attractive price points sit at the top of priority lists for both homebuyers and sellers.

When it comes to a home’s price, all parties involved rely on an accurate home valuation. And credentialed professional valuation experts shoulder most of the burden to meet the demands for accuracy and efficiency. They are traditionally seen as the most credible and objective providers when it comes to property valuation in the home buying process. Home evaluations need to be as precise as possible and a home’s condition history needs to be up to date to generate the most accurate value.

However, in many cases, lenders can sometimes wait weeks, and even months, for valuation reports and the homebuying and selling process cannot be completed until the report is received.

Delays and interruptions in the homebuying and selling process often begin at the valuations stage, where a lack of valuation professionals and an overload of demand for property valuations have, for over a decade, led to long delays in getting a completed home valuation. Nothing is more frustrating once a homebuyer has settled on a dream home than having to wait for a valuation to be completed. Simply put, it’s a bottleneck in the homebuying production line.

So, how can the mortgage industry combat this slow down and create a better experience, free of disruptions? Often times, we can look to the successes of the past to find inspiration for solutions of the future.

The past as inspiration for the future

In the early 20th century, Henry Ford found motivation in the continuous-flow production methods of brewers and flour mills, and was determined to “build a car for the great multitude.” Kindled by methods lacking production interruptions, by December 1913, Ford had installed the world’s first moving assembly line.

This new moving-chassis technology allowed several mechanics to work on one vehicle simultaneously as it moved down the assembly line, reducing production interruptions and the amount of time a team spent assembling a single vehicle, effectively reducing the time it took to build a Model-T from around 12 hours to just over two hours. The assembly line cut time and cost, and did so while improving the quality of his product.    

These days the idea of an automated workflow may not seem revolutionary to most. But when it comes to home valuations, perfecting the modern valuation “assembly-line” could be the game-changer many of us have been looking for. Automated Valuation Models are tools that use mathematical and statistical modeling techniques applied to large sets of data to establish a point in time estimate of the price of a real piece of property. As the housing industry continues to be transformed across all sectors, AVMs are more commonly being leveraged to assist mortgage professionals in removing human bias from appraisals and providing more timely and objective valuations.

In the residential world, AVMs have been used for underwriting, loan modifications, portfolio and valuation risk monitoring, home equity lines of credit, loss mitigation, marketing and pre-qualifications – at nearly every point in the ongoing review and analysis of collateral risk. Given their lower cost and ease of fulfillment, in addition to physical appraisals, originators, servicers and investors often leverage AVMs as part of their quality control process.

More and more, companies are investing in the application and use of AVM technology to substantially cut long delays and improve accuracy of data.

These technologies are dramatically changing the way valuations are generated and delivered, providing clearer, more precise details per home, showcasing facts over opinions and ensuring a consistent review regardless of who is analyzing the property. While AVMs don’t require an expert to interact in the value decision, AVM technology is also assisting appraisers and non-appraiser evaluators in unearthing additional risks and saving valuable time for everyone involved.

Richer data and expanded coverage

In essence, AVM technologies have become the moving-chassis assembly line of the valuation process, leading to positive impacts on the overall homebuying process. The key is the combination of expanded processing power, newer and deeper sets of data and an expanded selection of statistical tools such as artificial intelligence and machine learning to empower the tools to grow more knowledgeable, ultimately providing faster and better information to users.

AVMs take advantage of hardware processing power to assist valuation professionals in their analysis, creating a result that considers and reviews hundreds of thousands, even millions, of data elements simultaneously. AVMs organize and evaluate more elements of risk and value than any appraiser or analyst could ever consider manually. And given the speed of computer processing today, this data can be used to deliver real time information to valuation professionals.

Recent improvements in the breadth and depth of data available for modeling have greatly enhanced the AVM’s value to the industry. As a modeled value, the data available for the model to consume has a direct and meaningful impact on both accuracy and coverage.

Today’s valuations are not just about the count of bedrooms and bathrooms or the square footage, but rather consider multiple environmental factors such as distance to mass transit, elevation and topography of lot, local demographics and outside influences of things like nearby rental or foreclosure properties.

In addition, the coverage of data now allows for automated values in areas and on properties previously not able to be modeled. While public record and tax assessment data is more widely digitized and more easily accessed online, recordation data is still typically released weeks or months after a transaction is completed. The inclusion of transactional data from multiple listing services has allowed for values to be generated in rural areas or states that don’t disclose property transactions publicly. Moreover, MLS data is often updated daily or sooner. Using technologies like natural language processing to evaluate MLS listing information, including broker comments, AVM developers are finding new ways to capture as much information about a property that is available.

However, not all AVMs are built the same. Many lesser-equipped AVMs can only capture data weekly or monthly, while the best AVM tools are able to analyze the latest data in near real-time. Like the steel, rubber and leather of the Ford factory of the past, the breadth and depth of data available today are the ingredients that fuel that the AVM assembly.

AI and machine learning: better and more informed values

Remember the #10YearChallenge that made waves on social media earlier this year? News feeds were filled with these photo-mashups — two photos, side-by-side comparing the user’s appearance over a 10-year period.

Now apply that same comparison to AVMs. What would a 10-year challenge tell lenders about the value, or risk, of a property through the eyes of an AVM? AVMs weren’t nearly as powerful in the past as they are today. Ten years ago, the valuation process revolved more around traditional logistical statistics generating mathematical estimates. Data on properties and neighborhoods was limited, it was non-standard, difficult to aggregate and transform and lacked timeliness. Modeling relied almost exclusively on publicly recorded data about sales transactions with a three-to-six month delay. Digitized photos were almost non-existent.

While earlier conceptions of AVMs could churn data with a mathematical formula, without photos and modern AI, they couldn’t determine the interior or exterior condition of the property, or how the property has changed through construction or remodeling. Lenders would wonder: Is the home run down? Does it need repair? Was the kitchen updated? All of these details were a mystery to models and set limitations on the automated process. This major pitfall is all changing with new technology intelligence.

General AI, technologies like those used in products like Google Photos or various search engines, has already been commercialized to successfully identify differences when comparing images. But rather than comparing photos of cats and dogs, or humans, some AVMs are utilizing these types of technologies to compare past and present photos of a property or images of comparable properties to better understand the relationship of the different qualities in ultimate value estimations. These tools provide lenders the ability to forecast the property’s value, through past-present analysis or comparable property analysis, without even entering the property.

In addition to more data, two of the biggest changes in valuation technology include the ability to process larger amounts of data and the use of newer mathematical and statistical tools. A decade ago, images of kitchens or renovations, for example, were manually compared by the appraiser, leading to subjective adjustments to valuations. Today, instead of having to do their own research to find image comparisons, evaluators are using AVM tools and data, along with AI technology, to evaluate the property’s condition based on image classification and review, saving a great deal of time in the valuation process.

Accelerated improvements

Back in the 20th Century, Ford didn’t stop at the first generation of his assembly line.   Rather, he continued to improve it, accelerating the pace of production, further improving his product quality, and recording fewer and fewer mechanical delays. By 1924, just 11 years after the conveyor belts on Ford’s assembly line began to roll, the 10-millionth Model T came off the Highland Park, Ill., assembly line.

Similar to Ford’s assembly line for autos, AVM technologies will continue to find new ways to incorporate additional features that will improve their accuracy, reduce their delivery time, expand their use cases and lower the cost of residential transactions. As AVMs continue to improve upon their use of cloud-based architecture, expand their use of ML and AI and as hardware becomes more powerful, lenders can expect the pace of the valuations fulfillment to accelerate. Rather than waiting weeks for a single property valuation, AVMs are now assisting appraisers and non-appraisers in mere seconds.

In addition, like all technology, the rate of improvement in AVMs will continue to accelerate as well. And unlike non-automated valuation products, the ability of AVMs to be tested independently for accuracy makes them well suited to weather the changing landscape of home valuation in the coming years.

Whether as a tool to assist the appraiser or a stand-alone valuation product, the homebuying process should be proud of the modern AVM.

Most Popular Articles

3d rendering of a row of luxury townhouses along a street

Log In

Forgot Password?

Don't have an account? Please