Plug into the world of home valuations and learn about the latest trends in AVM Software

July 18, 2002-- Our interview this week is with Darius Bozorgi, President and CEO of Veros Software, Inc.

Veros Software is a company best known for predictive technology solutions for the financial services industry and is the host of the annual Predictive Methods Conference held each June. Veros Software offers decision support applications including its VeroValue AVM used to obtain property values on an individual or portfolio basis.

Q. Your AVM product, VeroValue, is relatively new. What value are you bringing to the marketplace?

Veros is very different from other AVM vendors. We started as a technology company with a mission to apply our core competencies, such as modeling and technology implementation, to a variety of financial services applications. VeroValue was designed, from the start, based on several of our proprietary predictive technologies, drawing upon in-house expertise across the entire spectrum of analytic-based predictive methods. Our competition is just starting to claim the use of “predictive” capabilities in a variety of blended models vs. more traditional one dimensional and static methodologies. We consider this a great compliment, but do not take all the credit. Many business intelligence or enterprise analytic applications have performed exceedingly well in numerous financial services sectors. It was simply a matter of time before they made their way into the automated valuation world. The marketplace is finally recognizing that much more should be expected from automated valuation technology. VeroValue and other newer valuation models provide the benefits of advanced technology in the form of very dynamic and robust tools.

Q. How important is technology in achieving an accurate home valuation?

I always tell people that there are two sides to delivering high performance valuations: data and technology. If you ignore either one, you’re in the wrong business. That being said, technology is very important. Results can differ substantially depending on the type of methods used and the designer’s skill with that technology. It’s difficult to put exact numbers on this, but results could easily differ more than 20% to 30% by using either inappropriate or inadequate technology.

Q. Is the number of models the key to AVM accuracy?

VeroValue definitely has numerous models, but the number of models is not the whole story. Without getting into the exact science of model development, the key is to have the tools and skill sets for the design of solutions from completely different perspectives. Simply having two models or 50 models that perform essentially the same or similar task is of little value. For example, adding an additional neural net or multiple regression model when you already have one of these in place adds little value and is only a slight variation of the same basic theme. Vendors need to build these models based on different disciplines. We are running approximately ten different valuation methods on any given subject property. Many of these models are themselves based on multiple valuation methods. The key is that these methods are based on a wide variety of backend predictive analytics. Innovation is another key element. We are constantly updating and refining existing methods while researching and adding new models to our systems. As market conditions evolve and technology is advanced, you simply cannot sit back and rely on a model that was developed several years ago.

Q. So would you say the success of VeroValue is largely due to your combination of multiple technologies?

Well that’s only half of the equation. Technology is very important because it allows us to optimize our analysis of the data and do things that would have been impossible just a few years back. However, as I mentioned earlier, technology alone will not make a successful AVM in my opinion. We rely on both predictive technologies, quality data sources and excellent data handling techniques. We have a rigorous internal process that includes cleaning, standardizing, mapping and mining all data that is imported into our databases. Having control of both the data side and the technology side of the equation is imperative in this industry. Although there have been tremendous strides in the data procurement, matching and blending world during the past five years, there is still much to be desired. No one can afford to take what they get from any data provider, slap it in some database, and then cross their fingers and hope for the best.

Q. What should end users consider as important issues regarding overall AVM performance?

The answer to that question could easily take up another interview. Let me just say that if you ask ten people in the industry what is important to them regarding AVM performance, you will likely get ten different answers. Reasons for this include a lack of standardization and that AVM usage is still in its infancy. Nevertheless, I believe that coverage, hit rate, accuracy, value-added analysis, service issues, and last but certainly not least, consistency are all issues that should be considered. Testing for these factors and evaluating the corresponding results is a major and unfortunately all too often easily manipulated effort. The downside is the end-user may have little or no idea of what they are actually getting in return for their dollars. Personally, I am a big advocate for a high degree of accuracy within an acceptable hit rate range. Virtually anyone can provide a generic area average or median with extremely high hit rates. The question is should you make a decision based solely on that information? We have authored several whitepapers on objective AVM testing and performance metrics, some of which have been incorporated by leading financial institutions as part of their own internal procedures. I would invite those that are interested in this issue to contact me directly.

Q. What problems do you see with AVM performance to date?

I think the problems are twofold. The first is a performance issue related to analytics. Up to now there has been a lack of true valued-added analytic information provided to the end user. One complaint we have heard is that, in the past, a vast majority of the information received by customers was a regurgitation of information they already had in their possession or had access to via alternative sources. Other than the estimate of value, which may or may not have been accurate, virtually no other information was being provided. For example, probably the single largest issue that was raised while we were developing our models was the lack of any correlation, if any, between the industry’s confidence score and the accuracy of the predicted market value. Second, to date, little attention has been paid to the service side of the equation. I am referring to both development and customer service issues unrelated to the core valuation engine, such as means of access and delivery, interfacing, reporting, account management, ease of use and customization. These are aspects of software development that no vendor can afford to ignore, especially in large scale implementations.

Q. How has Veros addressed the performance issues you mentioned?

In VeroValue, we provide much more than just “data” and an estimate of value. Of course customers want accurate valuation estimates but they also want value-added features and real market analysis. Today, VeroValue provides highly accurate valuations, that’s number one. But we go further, providing reliable confidence scoring that is highly correlated to the accuracy of our value estimate as opposed to some measure of data quality or quantity. We also currently provide property fitness information as a function of price and market trend information for the subject’s neighborhood and zip code. All of this is provided within our standard report and can be used independently or collectively to define levels of service. In the very near future, VeroValue reports will include numeric fraud scoring, and property fitness scoring which will address conformity issues. We are even working on the development of our forward-looking models that will predict value fluctuations in various future windows of time. All of these and other value-added analytics will be combined into an overall collateral score, which like a credit score, will provide the end-user with a complete picture of the issues surrounding collateral. This is the future of AVM evolution that we are planning for now.

Q. What can AVM vendors do to address the service issues?

I believe it is no longer satisfactory to simply provide an estimate of value and nothing else. As AVM developers, it is our job to help customers understand how AVMs are designed to work and how they can be used most efficiently. Regarding the latter, its crucial to make our products flexible and easy to use. Towards that goal, VeroValue provides very flexible means of access and delivery from web-based access with any standard browser, traditional batch processing, on-demand live batch processing over the web, any number of standard XML listeners for system to system communication and product delivery, customized or dedicated backend interfaces such as XML to PDF delivery mechanisms, to truly platform independent wireless access for any internet-ready wireless PDA or mobile phone. All of these are accessible through one user account and can be seamlessly integrated into virtually any system. Further, we want to place full control of how our AVM is used in the hands of the customer. Content delivery, account setup or modification, new user setup, security preferences, product access, reporting, and under what circumstances valuations are delivered are just a few of the aspects of use that we will place in the customer’s hands with our new account management system due to be released very soon.

Q. How are AVMs currently being used and are there changes coming in the future?

AVMs are used throughout the mortgage industry for home equity loans, second trust deeds, and, more recently for new purchase money loans. We see a wide variety of uses from QC and audit reviews, collateral valuations in the equity markets, in combination with alternative collateral assessments, to AVM only programs in the first trust deed space. Further, AVMs are now being incorporated in automated underwriting systems as a first step in alerting lenders to valuation issues and defining levels of service that they will need to provide. As a simple example, a low LTV combined with high credit score may easily qualify using just an AVM alone, whereas a high LTV or low credit score may require additional collateral assessment alternatives such as a drive-by or full appraisal. In the future, you can expect that most, if not all, properties will have an AVM valuation first and then, if required, a more traditional level of service.

Q. If AVMs are capable of such performance, are traditional appraisers obsolete?

Absolutely not. Considering my audience let me repeat that, traditional appraisers will not be obsolete. I offer and am in favor of quick, objective, consistent and accurate collateral assessments regardless of the source. Appraisers are still very necessary to provide the full range of services that the mortgage lending industry wants and needs. In fact, we get approached quite often by appraisers and appraisal management companies for a variety of cross-selling and partnership opportunities. AVMs will provide mortgage lenders with the analytic tools to make decisions on how best to serve their customers. Sometimes that decision will be to use the AVM’s value while other times full appraisals or other alternatives will be required.

Darius Bozorgi can be reached for questions or comments at dbozorgi@veros.com.