Clearing hurdles in the race to automate

Agencies across government are experimenting with robotic process automation, technology that uses software robots to quickly perform common, highly repetitive tasks previously assigned to employees. NASA has developed RPA tools to help with financial management, procurement and human resources. The Treasury Department's Bureau of the Fiscal Service is testing the technology to populate forms by copying and pasting information from different financial management systems. Despite its obvious benefits, agencies have not flocked to RPA.

Part of the challenge is the amount of time it takes to map out the process to be automated, develop it, scale it and maintain it. Many organizations underestimate what's needed to scale and maintain an RPA solution, according to a recent Forrester report. It requires keeping code up and running, dealing with errors and making adjustments to accommodate changes in technology or processes.

Another issue is finding staff that can implement the technology. “Having enough qualified implementers who know how to do the architecture for RPA, who know how to build the bots and know how to maintain them well … [is a] limit for growth for the category as I see it now” said Craig Le Clair, a Forrester analyst who studies the RPA market.

One way agencies can overcome these hurdles is by working with an industry partner that offers a variety of RPA software and support services.

“We do not just sell the software directly to the agencies and let them deal with the implementation,” said Paul Dillahay CEO of NCI, an IT services provider in the federal market.

Instead, he said, NCI works with agencies “from a change management perspective," helping identify and prioritize the workflows that should be automated, training the bot and then providing the ongoing operations and maintenance support that will be required as business processes or underlying systems change.

The company has developed a solution it calls Shai, for Scaling Humans with Artificial Intelligence. Shai was designed as an American-made “operational AI” solution, coded entirely on U.S. soil and, therefore, meeting federal agencies’ rigorous cybersecurity requirements, company officials said.  It has been trained to focus on highly repetitive tasks in three strategic focus areas: service desk support, back office support and data entry/data migration.

Shai can handle basic, standard and advanced tasks using both RPA and AI, according to Brad Mascho, the chief AI officer at NCI. “Basic level tasks are purely robotic process automation … you can use RPA to [create] new accounts,” he said.

“The next level is where you do machine learning and AI on past data, so something you have to train against, and you are able to make a higher level decision. As an example, think of it as coding census forms,” he said. “The most advanced work is where you’re doing constant [machine] learning based on data that’s coming in at a constant speed.”

RPA processes are easier to implement because they require much less data and training than AI solutions.

Still, it’s important for RPA solutions to start to add more AI, Le Clair said, because there are only so many processes that will benefit from the automation. Plus, as technology evolves, there will be less of a need for RPA because gaps between systems will become less of an issue, he said.

“If RPA stays in the task-harvesting mode … where you have the simple repetitive processes that are in high volume, at some point you’re going to run out of low hanging fruit,” he said. “So the digital worker, the bots, need to get more intelligent, they need to start to rip apart emails and unstructured content coming in, they need to have knowledge bases and machine learning that can make decisions.”

Work in the health care field helped prepare Shai for the authentication hurdles that come with a highly regulated industry like government, Mascho said.

“We’ve been dealing with sensitive data for seven, eight years now and have aligned the system to be able to handle that securely,” Mascho said. “We have everything from enterprise password vaults to the ability to work with password vaults that are already existing within the enterprise.”

NCI is beginning to pilot Shai with the Department of Defense and the Centers for Medicare and Medicaid Services, Dillahay said.

The DOD pilot involves a service desk application to review DD 2875 forms, which users must complete to gain access to computer networks as they move from one military base to another. Shai reviews submitted forms to ensure everything from dates to signatures are correct and kicks them back if anything is wrong, Dillahay said.

“If that form is not filled out correctly it creates a backlog," and it could take a service member weeks or months to get access to system if the process is not managed correctly, he said.