In-house tools sped up tax refunds in this county

Home to one of the country’s hottest housing markets, Travis County, Texas—particularly the city of Austin—has seen the volume of property tax refunds increase by 25% annually since 2018. To keep up and meet requirements to audit the refunds for accuracy throughout the year, the Risk Evaluation and Consulting Division of the county’s Auditor’s Office  relies on automation and analytics tools built in-house to perform continuous auditing. REC has reduced the time it takes to process audits of property tax refunds by 91%.

It used to take weeks to analyze the large volumes of property tax refunds, but the model can do it in less than five minutes, said John Montalbo, data scientist for the county. “It can detect anomalies, double check for accuracy and write findings to audit standards with incredible efficiency,” he added. 

“We’ve gone from 1,000-plus auditor hours per year to [being] at a pace right now for under 40, and we continue to trim that down,” REC Manager David Jungerman said. “We’ve made a lot of progress [in] being able to dedicate folks to more interesting, less mundane work.” 

Last month, the National Association of Counties, or NACo, recognized REC’s work with an Achievement Award for Financial Management.

Even as Travis County’s operating environment and services grew increasingly sophisticated, additional funding for audit compliance was unavailable, according to NACo. Developing innovative, automated auditing techniques allowed auditors to improve their effectiveness and increase their coverage.

The move from a time-consuming, paper-based process has been several years in the making. In 2018, REC began using a dashboard for remote auditing, but the COVID-19 pandemic really showed the office what was possible.

“It pushed forward how much more data is being collected during that whole refund process,” said John Gomez, senior data scientist at the county. “It allowed us to use data to verify when the check was scanned into the system or when the refund application was received and scanned in.”

It also enabled auditors to see the metadata so they could determine  who looked at and verified an application. “There’s a timestamp that gets tied to it … recorded and stored,” he said. 

Since then, the data science team has integrated algorithms into the review process to automate it. Now, human auditors are needed only to review audits that the system calls out as anomalous.

Before the algorithm could be deployed, the data scientists built an extract, transform and load process to collect and organize the data needed for all property tax refunds. Then the county’s senior auditor walked them through all the steps she takes and what she looks for in processing the refunds.

“We have our algorithms sitting on a virtual machine that will run itself,” Montalbo said. “Every time that it needs to run, it goes and it gets all the information, does all the tests with which it needs to do, notes exceptions when it finds them, and then starts compiling work documents.”

Those documents are put into an email that goes to auditors who spot-check what failed.

“It’s basically a multi-tab Excel spreadsheet that they get,” Jungerman said. “We keep one senior [analyst] dedicated to the audit and rotate staff, and basically, they just work the tabs of the spreadsheet if there’s any exceptions on there.”

Currently, REC is working with the data scientists to automate system-generated receipt testing to streamline audits. “We’re in the process with 12 county offices right now—and portions of a 13th—of looking at all of the system-generated receipts and tracking them to the elected official’s bank account and then tracing them to the posting in the enterprise accounting software,” Jungerman said. The automation would mean “being able to turn around findings to offices within a couple of weeks.”

It would also mean processing tens of thousands of receipts every week across all county offices. Currently, receipt testing typically samples only about 80 out of 20,000 receipts, he added. 

Automation could be applied to any type of audit, Montalbo said, although the exact mechanisms won’t translate seamlessly every time.

“We have 50-plus departments [in the county government] and most departments use a different application for their day-to-day business activities, which means different data is being stored for each transaction that is being receipted,” Gomez said. “So, we have to mine the data for each department to extract the information we need to verify each receipt is recorded correctly and deposited in a timely matter.”  

Despite the efficiency of automation, Jungerman said that he doesn’t foresee any processes running without some form of human interaction. “The vision is to automate all of our processes that we can and free standard auditors to just look at exceptions and to look at a whole lot of other areas,” he said, adding that “you need a human being to verify the potential findings.”