Companies and AI: 91% Believe Value Has Not Been Achieved

A large majority of professionals say they do not see the promised value of AI in their companies. With quality demands, market fragmentation, and rising bills, the shift towards production use cases has become the central issue, according to Kirsty Roth.

At Thomson Reuters, 87% of staff use AI daily

Thomson Reuters reports that 87% of its employees actively use AI tools in their daily work. The targeted use cases focus on five areas: engineering, customer support and success, marketing, publishing and content operations, as well as core technology operations. In support and sales, the group leverages its internal platform Open Arena and the Claude model. Approved AI services provide answers in seconds, where hours were previously needed to aggregate information from sales teams and Salesforce. Claude can produce summaries, highlight opportunities, flag risks for an account, or detect a recent call from an unhappy customer, in order to prepare for meetings more quickly. The company has integrated these uses across a global workforce of 27,000 people. According to Kirsty Roth, leaders must also work to overcome professionals' fears: demystifying AI upfront and allowing teams to experiment with the tools has been crucial.

The diagnosis: 91% see no value, 95% failure rate according to MIT

According to a survey of 1,800 professionals, 91% believe their organization is not achieving the value that AI can offer, a gap deemed concerning by Kirsty Roth. Research from MIT suggests that 95% of AI projects fail to deliver value, while leaders fear rising IT bills due to high resource consumption. On the ground, 41% of AI users report not having access to high-quality tools. Even when an AI strategy is formalized, 35% of professionals say it does not translate into their daily work. Feedback indicates an abundance of tools without clear guidelines, making it difficult to demonstrate benefits as software costs rise.

What teams demand from their tools: security and explainability

Professionals specify clear expectations for AI tools. A vast majority demands the protection of confidential data (96%). They also want results grounded in authoritative content (94%) and reasoning that is explainable and defensible (90%).

Moving from a plethora of tools to production in a fragmented landscape

The AI market is described as hyper-fragmented and hyper-fractured by Steve Lucas, with a proliferation of concepts such as private models, open weights, frameworks, and agentic loops. New technologies, particularly agentic models and deep search tools, will continue to enter the enterprise. Kirsty Roth recommends focusing efforts on both well-founded explorations and production-level use cases, as the discussion shifts towards change management and process adaptation. Organizations that make progress are those that convert exploration into production services; when a tool is chosen and processes evolve accordingly, improvements and savings emerge. Eighteen months ago, the emphasis was more on experimentation supported by management.