How LMI incubated its own generative AI tool before going to market

Trying out one’s own technology creation before showing it to current or prospective customers has, for the most part, become a modus operandi for companies in the government market.

In the case of LMI, that meant the company used its LIGER generative artificial intelligence solution internally for 14 months before taking it to the market.

As two LMI executives described in an interview, LIGER originated out of the company's Forge technology studio that works to help employees explore and test new ideas.

Josh Wilson, LMI’s president of markets, growth and technology, said Forge “enables us to be that pace of need provider” in a nod to how quickly tech like generative AI advances.

“What makes that different? Don’t be obsessed with where those ideas come from, be obsessed with the best ideas being put in front of your customer as quickly as possible,” Wilson said in explaining Forge’s mission. “That’s what we’re trying to build out in our culture.”

As for LIGER itself: the tool is designed to apply machine learning and natural language processing techniques for use cases like acquisition tasks, such as market research or contract writing, and comparisons of policies across organizations.

“It was important for us to put (LIGER) in the hands of our folks first and test it with our folks first,” said Alex Adamczyk, vice president for analytics and AI at LMI. “Before we ever went to any customer to say ‘we think this is useful,’ we used it ourselves.”

For one internal use case, LMI put LIGER to work in responding to requests for information out of agencies and finding details such as the relevance of past performance. The company also started using LIGER to help employees with customer-facing work early on, Adamczyk said.

Now imagine being a federal agency with 45,000 pages of policy, which alone illustrates the data challenge across government and could also show the opportunity that technology tools have to make sense of it.

As Adamczyk pointed out: no one has ever read all of that content and no one will ever read all of that content.

Neither Adamczyk nor Wilson would name that agency, but this client put LIGER to work and started to find things it had not before.

“By the end of the hour, they had found inconsistencies in pricing about how those policies interact with each other,” Adamczyk said.

Data scientists built LIGER and several other tools, but LMI sought feedback from its policy and acquisition folks to make sure that more people could use LIGER than just data scientists.

That is one facet of the government’s never-ending challenge of making sure the technologies agencies adopt can scale across the organization.

“If you don't have a robust enterprise, data management, data engineering function and a robust security function, it goes nowhere,” Wilson said. “Everybody loves their shiny object, but they all know that at some point, a CISO (chief information security officer) is going to tell me ‘no.’

“You have to go into those things thinking about the conversation with the CIO (chief information officer) and CISO, otherwise you will never scale, you’ll stay in proof-of-concept prototype purgatory.”