Faster AI Adoption Starts With Smarter Federal Acquisition

Federal health agencies are finding new ways to use AI in the acquisition process, speeding up procurement while keeping human officials in the loop and responsible for critical decisions.

At the National Institutes of Health, AI helped transform a proposal evaluation process that could have taken two years into one that was completed in about four months, Ricky Clark, NITAAC director at NIH, said at the Health IT Summit Wednesday.

Clark said NIH used AI tools for offer validation, algorithm validation and document consistency. The technology identified gaps and deficiencies and provided that information to contracting officials, who remained responsible for evaluating the results and making decisions.  

“We were able to drill down from the solicitation level to the regulatory level and then have the CEO provide feedback,” Clark said. “I felt we created one of the most efficient proposal evaluations I’ve ever seen. We had an ad hoc report for everything. We reviewed each and every solicitation to ensure that everyone was totally compliant.” 

At the Department of Health and Human Services, officials are similarly experimenting with new acquisition approaches to bring AI capabilities into the department more quickly. 

Quintin McCoy, executive director in HHS’ Office of Acquisition Solutions, said the department is using Small Business Innovation Research programs and Federal Acquisition Regulation-based contracts with phased approaches. HHS is also seeking Commercial Solutions Opening authority, which McCoy said could give the department additional flexibility to test emerging technologies and work with nontraditional vendors. 

“CSO authority allows us to discover. It allows us to try out and pilot and still do it legally, but it’s not with the FAR restrictions on it,” he said. “It also allows us to move faster to bring on technology [with a] phased approach. … The key there is just we want to get things when it’s new and try it and test it with a little bit more flexibility.”   

But speeding up procurement is only part of the solution; faster acquisition must be paired with a clear path from experimentation to deployment. Ben Cushing, chief architect at Red Hat, said any pilot being developed should be developed on a production system. 

“You don’t want to build a pilot in production per se, but you want to build it on a production system that mirrors what it will look like in the real world,” Cushing said. 

Cushing also encouraged agencies to require vendors to disclose dependencies on outside AI models and services up front.

“The easiest way to garner that knowledge is to require it upfront and just make it super clear that you have to disclose that,” he said. “Otherwise, it’s really nearly impossible to figure that out after the fact until you’re well into a system design document or task design document. You don’t find that out until later, and a lot of times the bid is already done, and so you’ve awarded, but now you find out there’s a whole dependency chain.”