91% of professionals say their firm still falls short on AI, survey finds

"People are starting to work out that these technologies cost lots of money, and they don't necessarily see the value from them yet," Roth told ZDNET. "And so the conversation has turned into the classic change management one, which is, 'OK, we've got all this tech and all these tools, but how are we really changing our processes and the way we operate to be more effective?'"

The survey found that professionals have clear expectations for AI tools: 96% require safeguarding confidential data, 94% want outputs grounded in authoritative content, and 90% need explainable and defensible reasoning. However, two in five professionals (41%) who use AI at work said they lack access to high-quality tools. Even where an AI strategy exists, just over a third (35%) of professionals in firms with a named AI strategy say the approach is not visible in their day-to-day work.

Roth attributed this disconnect to "tool blast," where organizations push a broad selection of AI services to staff without a clear business outcome. "I've heard people say, 'I've been given all these things. But what am I meant to be doing?'" she said. "Too many firms aren't clear on what tools people should use. I think it's then very hard to see the uplift, other than you'll see your software costs go up significantly."

To bridge the gap, Roth advised businesses to focus on two areas: well-grounded explorations and solid production use cases. On explorations, she described Thomson Reuters' open-minded approach to generative AI. "If you had a cool tool and [were] in marketing, sales, or coding teams, and you wanted to try it, we would let you try it," she said. The firm encouraged employees to experiment with tools for about six weeks, test results, roll out successful ones across other teams, and kill those that did not deliver.

"We've said to people, 'Here are the tools. Reimagine what you can do,'" Roth said. "Now, obviously, some have done better than others, and some have needed more nudging and help than others. But we have encouraged people to think about what these tools can do in today's world. And that approach seems to have worked well for us."

On production use cases, Roth noted that firms pulling ahead in generative and agentic AI are those that convert explorations into production-level services. "I think the successful firms are now starting to say, 'OK, we have chosen, and we are going to use this tool, and therefore we're all going to change our business process to operate in this new way,'" she said. At Thomson Reuters, specific use cases center on five areas: engineering, customer support and success, marketing, and editorial. Roth cautioned that many firms remain in the "playground" phase, while those doing well are highly specific about their use cases.

The broader AI landscape remains fragmented, as Boomi CEO Steve Lucas recently told ZDNET, describing the state of play as "hyper-fragmented and hyper-fractured." Professionals now contend with terms like frontier models, private models, agentic frameworks, and agentic loops that did not exist a few years ago.