Cambridge, MA, Sept. 09, 2026 (GLOBE NEWSWIRE) -- As organizations invest heavily in generative artificial intelligence, many leaders assume that expanding employee access to AI tools will naturally lead to transformative innovation. New research from MIT Sloan School of Management suggests otherwise: the organizations that generate lasting value from AI are not necessarily those with the best technology, but those that do the most to support the people experimenting with it.
In a new working paper, Experimentalist Intensification Governance: Managing Worker Negative Consequences Associated with Generative AI Innovation Work, MIT Sloan professor Katherine C. Kellogg, together with Batia Wiesenfeld of New York University and Arvind Karunakaran of Stanford University, found that successful AI innovation depends on whether organizations recognize and support the substantial but often invisible work required to turn AI experimentation into organization-wide solutions.
The consequences of getting this wrong can be dramatic. In one organization studied, more than 80% of employees engaged in AI innovation efforts eventually disengaged from the process, leaving only three organization-wide AI solutions in use. In another, leaders invested in structures that supported experimentation and collaboration, resulting in 141 AI solutions already deployed across the organization, with many more in development.
“Many executives are focused on getting more employees to experiment with AI,” said Kellogg. “But organization-wide AI innovation isn’t an adoption problem, it’s a persistence problem. It depends on whether employees continue choosing, week after week, to experiment together, refine solutions, and adapt them for real-world use across their organizations.”
Successful AI innovation requires significant employee experimentation and coordination
The researchers found that employees developing organization-wide AI solutions performed substantial work beyond their day-to-day responsibilities. This included trial-and-error experimentation to determine what AI could and could not reliably do, reviewing and refining outputs with colleagues across departments, and continually adapting solutions as AI models rapidly evolved. The researchers describe these activities as forms of collective “experimentalist work” that are critical to innovation but often remain invisible to leadership.
“Building organization-wide AI solutions requires ongoing experimentation,” Kellogg said. “Domain experts spend a lot of time testing possibilities, refining outputs, and adapting solutions as the technology evolves, but that work often goes unrecognized.”
Lack of organizational support can cause AI innovation efforts to stall
When the additional effort required to build and scale AI solutions is not recognized or supported, employees may slowly withdraw from the innovation process. Rather than openly resisting AI, participants often disengage from the continual experimentation, review, and refinement required to create organization-wide value. In the law firm that the researchers studied, more than 80% of domain experts eventually dropped out of AI innovation efforts.
The study found that scaffolding — or structured support systems — for collective GenAI experimentation, not technological capability, was the key differentiator between success and failure. The healthcare organization studied now has 141 organization-wide AI solutions in use and many more under development. The law firm has only three. Researchers found no meaningful differences in AI readiness, access to technology, regulatory constraints, or the suitability of AI for the problems being addressed. What differed was the support infrastructure surrounding experimentation.
Study Finds Similar AI Investments Produced Dramatically Different Outcomes
To understand how organizations generate lasting value from AI, the researchers conducted a two-year field study at an academic medical center, referred to as NE Health, and a corporate law firm, LegalCo. Both organizations introduced secure AI “sandbox” environments — or isolated environments to run tests — provided training, and encouraged employees to develop AI applications with organization-wide potential. At NE Health, teams worked on patient-friendly discharge summaries. At LegalCo, teams focused on legal research.
The two organizations started from similar positions, but their outcomes diverged sharply.
At LegalCo, employees gradually stepped away from the innovation process as the burden of experimentation mounted. The organization ultimately scaled back many of its AI initiatives, leaving only three organization-wide solutions in use.
“What we saw was people quietly stepping back,” Kellogg said. “Workers didn’t refuse to engage outright. But because of the burdensome extra work, they stopped participating in the everyday trial-and-error that innovation requires. Once that happened, the organization’s AI initiatives quickly lost momentum.”
At NE Health, however, employees remained engaged in collective experimentation and additional contributors joined innovation efforts over time. The result was a growing portfolio of AI applications, including 141 organization-wide solutions currently in use and many more in development.
Researchers identify four organizational practices that support AI innovation
NE Health provided ongoing education through workshops and “promptathons,” clear documentation practices, and dedicated technical experts for high priority projects. LegalCo provided initial training but limited ongoing support.
NE Health established shared evaluation rubrics and knowledge-sharing forums that helped employees assess and improve AI solutions collectively. These structures gave teams a shared framework for measuring, discussing, and improving Ai solution performance.
Leaders at NE Health implemented a formal risk-screening process and provided technical support to help employees integrate AI solutions into existing systems and workflows. LegalCo offered little comparable support.
NE Health formally incorporated AI experimentation into job responsibilities, performance evaluations, promotions, publications, and career advancement opportunities. Employees saw their innovation efforts translated into professional recognition and growth. At LegalCo, many participants reported that AI-related work remained invisible in reviews and compensation decisions.
“Leaders often assume AI experimentation is a stretch assignment that motivated employees will absorb on top of their regular responsibilities,” Kellogg said. “Our findings suggest they will, but only temporarily. Without meaningful support, recognition, and resources, people eventually disengage, and organizations struggle to realize the value they expected to gain from their AI innovation efforts.”
Findings highlight a critical challenge facing enterprise AI investments
The findings arrive as organizations across industries continue searching for ways to convert AI adoption into measurable business value. While many executives focus primarily on technology deployment, the research suggests that sustainable AI innovation is fundamentally an organizational challenge rather than a technical one.
According to the researchers, organization-wide AI innovation depends on employees’ willingness to continue experimenting, evaluating, refining, and implementing new solutions over time. Organizations that fail to support this process risk losing the very expertise necessary to identify and scale high-value solutions.
Ultimately, the study concludes that competitive advantage from generative AI will not come from access to technology alone. It will come from an organization’s ability to build the structures, incentives, and support systems that enable employees to keep experimenting, learning, and innovating together.
“The real driver of AI innovation is not the technology itself,” Kellogg said. “It’s whether organizations invest in supporting the people who make that innovation possible.”
Attachment
Matthew Aliberti MIT Sloan School of Management 7815583436 malib@mit.edu