USDA’s Bot Program Saves More Than 150,000 Hours Per Year

The 2021 fire season in the U.S. has been one of the worst on record. California alone had more than 8,000 wildfires over almost 2.5 million acres. Thousands of seasonal firefighters across the country worked very long hours in extremely hazardous work conditions, using specialized equipment and communications, aircraft loaded with water or fire-extinguishing chemicals, heavy machinery to cut firebreaks, and more. Many of these efforts were planned, directed and coordinated by the U.S. Forest Service.

And all that effort also generated thousands upon thousands of invoices and paychecks, many of them handwritten, submitted to the Agriculture Department—home of the Forest Service—for prompt processing. In earlier years, all those invoices had to be processed manually but now Agriculture’s Robotic Process Automation, or RPA, program has created a bot to handle the vast majority of them.

Lattrice Goldsby, Agriculture’s RPA program and branch manager, said it would take anywhere from 5,000 to 10,000 hours to process all the data before the introduction of the bot. 

“We’re estimating saving $173,000 [through] cost avoidance, and we’re also adding AI on that in the first quarter to process handwritten invoices,” she said. “The goal is to increase being able to invoice and pay to get the supplies they need quickly.”

Goldsby said the RPA program so far “has 66 bots in production, with $5.3 million in cost avoidance and 155,989 hours saved per year.” They are spread throughout the department, used in functions as varied as contracting, procurement, IT and human resources.

On top of that, “currently we have 55 bots in various stages of development,” she said, with a dozen scheduled for deployment by the end of this quarter and another five in the queue for release in the next quarter.

Earlier this year, the department hosted its first “bot-a-thon,” Goldsby said. “It was a three-month competition with participants across the department [producing] prototype automations. We did weekly educational sessions, and got 11 prototypes, with one now in production. We’re going to do it twice a year.”

With all those hours of labor freed up from manual processing, Agriculture took the next step: creating a reskilling program for employees to direct their energies toward more productive work. Agriculture CIO Gary Washington said the program has graduated its 27th cohort, including 19 during the pandemic.

“USDA partnered with [the Office of Management and Budget] to create the cohorts,” Washington said. “First in New Orleans, where we attempted to get IT people [to enroll], then in D.C. We went out and advertised the opportunity.”

He said the intent of the reskilling program is to give current Agriculture employees opportunities to stay with the department. “Currently, I have four people on my team from the cohorts,” he said, including the RPA program manager for implementation and an RPA subject matter expert for infrastructure. “He built the whole RPA infrastructure.”

Goldsby said employees in the reskilling program learn such things as business analysis, assessing business processes for automation, return on investment automation, and specific knowledge about topics such as data types and controls.

Agriculture is a federated department, comprising a number of standalone agencies. The RPA program, however, is mostly centralized, offering its automation services to all of the agencies. An agency can develop its own bot(s), but they have to fit within the RPA program’s governance.

“We have strict governance,” Goldsby said. “It’s a nine-step process … We use it to document whatever is being automated, whether the data is structured or unstructured, any [standard operating procedures,] whether it’s attended or unattended, and how many applications it will be touching.”

The rule of thumb for developing a bot is that it shouldn’t take any more than 12 weeks to complete, she said. “Anything that [takes longer] may not be suitable.” So far, the RPA program has been able to keep up with its requests. The agencies who ask for a bot to be created pay for the work, so the program is able to staff up to meet the need.

Goldsby said deciding whether a bot should be created takes qualitative benefits into account, as well as cost and labor savings.

“Velocity—will it increase the speed? Is this one of those processes that we need to get done sooner? Also, are we improving compliance? Will it increase data quality?” she said. “We also look at the estimated amount of code reuse – when we have a [process] owner come to us, say in procurement, we’ll [ask] is this process similar” to another one already automated? And the program will give the customer an estimate of how much code will be reused, since that benefits both turnaround time and the cost to the customer.