In what is billed as a pilot of the company’s AI technology, Cary-based SAS worked hand-in-hand with the Wake County tax administrator to determine how much every one of the county’s 400,000 properties should be valued. Its model constantly evaluates properties across the county, meaning it could have a theoretical value ready every single night.
Marcus Kinrade, Wake County’s tax administrator, said the partnership with SAS began in the summer of 2017, when the company reached out to Wake County about using its Viya analytics program.
With the appraisal cycle in Wake County shortened from eight years to four years, Kinrade said his office was slammed with work and figured that whatever help it could get would be beneficial. To his knowledge, he believes Wake County is the only jurisdiction using AI in the U.S. to determine property values.
“We were looking for a solution because they had shortened the appraisal cycle to four years, which really piled up a lot of work for us,” Kinrade said in a phone interview. “We were adding staff at two [workers] per year but we needed a lot more.”
So far the partnership has worked, and the county has extended its contract with SAS. Wake County has paid the tech firm $707,587 since 2018 to use the technology, according to figures The News & Observer obtained through a records request.
However, Kinrade argues, that contract is far cheaper than bringing on additional staff. The tax administrator’s office currently has 25 appraisers, which is about one per 16,000 properties in Wake County. Kinrade said industry standards say it should be about 12,000 parcels per appraiser — a rate that would likely require a doubling of staff.
Just one part of the toolbox
In its first revaluation since 2016, homes in Wake County increased 20% in value from four years ago, and commercial properties increased 33%, The N&O reported.The county has yet to determine what its new tax rate will be, so it is still unclear what property owners will pay. The current tax rate is 72.02 cents per $100 of assessed property value.
It’s important to note, however, that the new values were not arrived at solely through robots. By state statute, the tax office cannot rely just on the AI and must follow certain guidelines.
“SAS’s recommended values are just that — a recommended value,” he said. “They were just a tool.”
SAS pitches it as a source of comparison, rather than an elimination of the traditional assessments.
“Assessors use this tool to validate their own expertise and to identify neighborhoods and homes with assessed values outside the expected range which may require a deeper analysis,” Jennifer Robinson, a SAS industry consultant for state and local Government, said in an email.
The SAS model tracks 140 variables, which across 400,000 properties equates to about six gigabytes of data. Assessors have to look at everything from square footage and location to the number of bathrooms in a house to determine a property’s value. With each new input, SAS’s AI learns how any one of those variables contributes to its value.
“A typical modeling approach would require substantive knowledge of how these inputs interact, for example, how square footage rates vary by neighborhood, or how the value of an additional bathroom is greater for houses with few bathrooms than for larger houses with many bathrooms,” Robinson said in an email. “However, using SAS machine learning software, such knowledge and assumptions are not required — the computer learns as it processes each home sale.”
Million-dollar homes and gentrifying neighborhoods
Kinrade said SAS’s model was accurate for most properties in the county, though there were two exceptions where the tax administrator’s office felt it was off. Because of a lack of data — or in some cases too much — the AI was either too aggressive or too conservative with homes that were valued at more than $1 million, and in neighborhoods that were seeing values increase rapidly.While there are more homes than ever in the region selling for over a million dollars, there still isn’t enough data to feed the AI about that price range.
“In Wake County there is a dearth of data on sales at very high price points, compared to lower priced sales,” Robinson said in an email. “The SAS solution is dependent on sale prices and makes no assumptions about home prices in Wake County and works best at price points that are well represented.”
However, as the number of homes in Wake County selling at those price points grows, the AI should become more accurate.
The AI also lacked context when it came to neighborhoods that appear to be gentrifying rapidly, suggesting values in those neighborhoods that were higher than the county felt comfortable with.
“Maybe not a lot higher but quite a bit — 8% to 10% higher in developing areas,” Kinrade said. “We were conservative there.”
While property values in quickly-changing neighborhoods are indeed rising quickly, Kinrade said, there is an understanding that a lot of people in those neighborhoods might experience sticker shock when their tax bills come due.
“Some of the biggest increases were in some of the areas where home values were quite low and are rapidly increasing,” he said of the recent re-evaluation. “Individuals are expressing concern that they won’t be able to pay the taxes. We are trying to get out to the areas that are being developed rapidly and let them know about certain tax relief available.”
Despite that, the county hasn’t received as many informal appeals — or people challenging their property values — as it expected. So far it’s received around 6,000 informal appeals, though Kinrade expected something closer to 28,000.
Future of appraisals?
Kinrade said he thinks Viya would be a useful tool for other large municipalities that have a lot of data for the model to crunch. For it to work in small communities, SAS noted, it would likely require collaboration with other communities to share data.SAS said AI has become a big topic in the appraisal industry, and already it is in conversation with many other counties and cities across the country.
“Given the types and amounts of data involved, AI technologies will become more and more prevalent in property assessing,” Robinson said.
©2020 The News & Observer (Raleigh, N.C.). Distributed by Tribune Content Agency, LLC.
