Federal agencies buy the most recent technologies such as artificial intelligence, satellite broadband and supercomputing services. Yet too often, they do so with antiquated procurement systems in which simply finding applicable policy for a given procurement feels like nothing so much as a scavenger hunt.
“It’s disjointed; it’s hard to find; it’s not organized,” Amanda Price said, describing the process by which contracting officers and others in the buying function must discover the best way to structure a given buy. Price, the director of the virtual acquisition office at Unison, said it’s like rummaging through card files in an old-fashioned library.
Much of the information is online.
“There are SharePoint filed and portals we house things in,” Price said. “There’s Acquisition.gov which has the FAR [Federal Acquisition Regulation] and agency supplementals. But the problem it, it’s static. You have to know where to find it. I like to call it a scavenger hunt, but probably less fun.”
Plus, many agencies still rely on paper documents here and there.
“There’s even the true horror story,” Price said, “where someone sent an email in 1999 that got printed out and hung up in someone’s cubicle. It’s treated like guidance.”
A new twist adds fresh difficulty to procurement policy discovery, Price said.
“There’s a unique problem right now in federal acquisition,” she said. “Knowledge is walking out the door. Folks who have 30, 40 years of experience – that experience, that context they have, are valuable.” Anyone can recite the FAR, but nuance, experience, context and intuition of how to put the right pieces together has left.”
She added, “Federal acquisition is nothing if not complicated and nuanced. And there’s a reason why it’s complicated and nuanced.” For instance, agencies, in addition to the specific requirement for an acquisition, have small business and competition mandates to follow.
Beyond that, different regulations enhance differing situations, Price said.
“You might use FAR Part 13 for an acquisition, because speed is key and that’s worked for you in the past, Price said. “But there are times where you may say, ‘You know what? I’m going to look at Part 15 and work through that process to avoid the complication on the back end during post award.’”
Enter AI
Nuance and complexity, Price said, “is really where I think artificial intelligence steps in.” AI can capture human experience “and scale it across to multiple people, multiple agencies. That’s where the real magic happens.”
In the example of a choice between FAR Parts 13 and 15, a large language model (LLM) “might ingest agency supplementals. It might ingest the FAR. It might ingest policy, templates and guidelines,” Price said. It won’t replace, but rather amplify human expertise, and supply some of the knowledge otherwise gained through long experience, or from the scavenger hunt.
Price cautioned that the general purpose, public-facing LLMs won’t suffice when seeking insight into something as specific as federal procurement.
“You’re not going to go to a tool like ChatGPT and say, ‘Should I use part 13 versus 15? And what are the pros and cons?’” she said. You’ll get a definitive answer all right, but even if close to the mark, the answer won’t have the specificity or nuance needed in the federal domain. And it will lack auditability.
“Close enough is just not going to cut it. What you need is no hallucinations,” Price said.
Unison’s answer consists of an agentic, or autonomously-operating, AI application tied to an agency’s workflow and using highly specific data in the LLM with which the agents interact.
Integration into the workflow means users avoid elaborate prompt engineering or going on those scavenger hunts for information outside of the procurement system.
“You’re putting in what you’re buying, and the system itself is ingesting all of that context that you’re giving it,” Price said. The agent may flag items or procedures, or it may operate on its own in the backend to “to take you exactly where you need to go.”
The basic Unison procurement LLM has the FAR and other widely applicable documents, but users can tailor the model to include their agency’s supplements and specific policies.
Price noted that the administration has undertaken a comprehensive update of the FAR. It’s also centralized some acquisition categories in the General Services Administration and its contract vehicles. These moves affect how agencies – who may have lost people experienced in procurement – go about their buying.
“That pivot is not always the easiest pivot to do,” Price said. “Rather than having to spend time getting really smart on that type of acquisition, you’ve got this safety net when you’re using this agentic AI that is purpose built for federal acquisition to guide you and carry you through.”
As for the tool itself, Price added, “Here at Unison, we are moving to a place where agentic AI really has that human expertise, and prompt engineering isn’t as important.” That means faster usability for a greater number of people.
For instance, if told the agency wants to make a commercial buy, “it’s making those recommendations to you, depending on what your thresholds are, depending on your IGCE [independent government cost estimate],” Price said. “It’s flagging when you’re over those thresholds. You really don’t have to do much engineering.”
All the AI-powered flagging and recommending occurs within the regular workflow.
“Purpose-driven AI should connect into your contract writing system,” Price said. “It should connect into wherever you are in that process. It should connect into your budget office and whatever financial management system that you’re using.”
She added, “And it really gets to yes on acquisition, yes on the program, yes on the mission, but then to the downstream effect of auditability,” all accompanied by an automated logging of the steps for transparency. Moreover, the model, she said, “continues to ingest what you feed it, and it continues to get better and better, based on those nuances.”
Copyright
© 2025 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.
