Policy & Practice | August 2019

technology speaks By Kirke Everson

Five Keys to Intelligently Deploy AI and Automation

J ust a few years ago, only a handful of government organizations were truly using artificial intelligence (AI) and automation. Since then, many applications have emerged with dem- onstrated results that display AI and automation in all of its forms, including robotic process automation (RPA), machine learning, natural language processing (NLP), and other cognitive tools. Examples throughout govern- ment include automating repetitive back-office tasks in human resources, finance, and procurement; harnessing AI to gain deeper insights into program data and to identify or predict anoma- lies; and deploying chatbots for better customer service. Within health and human services, recent applications span the front-, middle-, and back-offices. n California’s state health insurance marketplace, Covered California, deployed a virtual assistant during open enrollment to better serve customers, decrease call volume, and reduce seasonal call center workload needs. Utilizing a cloud- based machine learning platform and NLP framework, the virtual assistant named CiCi responded to more than 111,000 user questions within the first four months and con- tinues to evolve its knowledge and responses through training. Through interfacing directly with customers, CiCi collects valuable and actionable data such as customer sentiment and satisfaction, usage trends, as well as frequently asked and unanswered questions. n The Center for Program Integrity at the Centers for Medicare and

large application releases in eligi- bility and child welfare information systems. RPA bots build a large inventory of pre-configured cases allowing human testers to execute cases more efficiently by reducing the average time per case from 21 minutes to 2 minutes. Agency and program leaders are to be congratulated for this progress although there is still much work to be done. While agencies are clearly embracing AI tools, most are only applying them to individual use cases without a strategy to scale them across the enterprise where they would have a greater impact on productivity, cost reduction, and improved workforce morale. Given the difficulty of moving proofs of concept into production at

Medicaid Services launched a tool combining RPA, NLP, machine learning, optical character recogni- tion, and microservices to expedite a time-consuming medical record intake and review process that evaluates payment accuracy. The tool retrieves records as they are submitted by opted-in insurers and providers, then digitizes, extracts, and validates data to identify excep- tions based on pre-defined business rules and coding guidance. Records that cannot be processed with a high degree of confidence as well as prob- lematic records are routed to staff for manual review. Working around the clock, the tool is 95 percent accurate and twelve times faster than the process was before applying intelli- gent automation. n Multiple states are using RPA to support user acceptance testing for

See Automation on page 35

Illustration by Chris Campboell

Policy&Practice August 2019 24