When Tech Tools Get Adopted or Ignored at Work

In fast-paced and hectic work environments—like a hospital, where staff must multi-task constantly—AI and algorithms can help employees improve decision-making. But workplace fatigue and competing priorities often slow down the use of promising technology and can prevent its adoption altogether. 

These are the findings of Duke researchers, who conducted an interdisciplinary study on “unplanned” patient readmission rates at Duke University Hospital. 

“Readmission rates are a very big deal,” said Professor Scott Rockart of Duke University’s Fuqua School of Business, who was part of the team. 

“Not only is readmission costly and problematic for the patient, but the regulator–the Centers for Medicare & Medicaid Services (CMS) —even applies penalties to hospitals who are above certain thresholds on readmission rates.”

Specifically, the Duke team investigated the declining use of an algorithm-based tool that analyzes health records and generates a score predicting the chance of readmission.

In a new paper published in JMIR Human Factors, the researchers found that although health regulator pressure may initially drive the use of technology, long-term adoption often falters in the face of conflicting pressures—especially when the tools’ value is hard to observe.

A powerful tool left unused

Research has found that tech tools—or “Computerized Clinical Decision Support” tools (CDS), as they are called in healthcare—improve diagnosis accuracy, treatment selection and overall patient outcomes.

“These tools have been shown to have a very high predictive ability, more so than the predictions made without technology,” Rockart said. 

And yet, adoption is fragmented and varies widely among healthcare professionals.

At Duke University Hospital, the researchers observed declining use of a tool called Unplanned Readmission Model. The tool analyzes electronic health records—everything from diagnoses to medications—to generate a risk score indicating whether a patient is likely to be readmitted after discharge.

In theory, the tool fits directly into clinical workflows. A patient flagged as “high risk” might receive more follow-up care, additional services, or a delayed discharge.

“The tool flags issues for the case manager to take into account,” Rockart said.

In hospital settings, where doctors and staff juggle multiple priorities simultaneously, such tools should be expected to enhance real-time decision-making, Rockart said.

For the researchers, this raised the question: what determines whether people use technology that’s been proven effective in predicting outcomes?

Adoption as a dynamic process

To answer that question, the Duke team focused on behavior and system dynamics.

They brought together clinicians, case managers, and other hospital staff in structured workshops—a process known as “group model building.”

Participants were asked to map out what was happening: Why did use of the tool rise at first and then fall? What factors influenced whether someone paid attention to it or ignored it?

“We were trying to move beyond ‘it’s a good tool and it should be used’ to ‘why isn’t it being used?’” Rockart said.

These sessions produced detailed insights on how different factors—training, workload, incentives, team dynamics—interact over time.

Their conclusion: adoption rises, then plateaus, and sometimes fades depending on how the tools fit into workflows, how visible their benefits are, and how much attention the organization continues to give them. It’s a dynamic process shaped by the system around it— and one that can be managed.

Turning behavior into a model

Rockart and colleagues translated these insights into a mathematical model.

That meant identifying key variables—such as training levels, user interest, and institutional pressure—and modeling how they change over time.

The result was a model that could simulate different scenarios: What happens if training increases? What if competing priorities are particularly strong? What if users can see clearer evidence of success or hold a lower bar for acceptance?

“We are trying to understand internal feedback loops that could be managed by those rolling out a new technology. If you change a parameter, it leads to very different outcomes,” Rockart said. “The mathematical model can help test the logic of your explanations and identify more effective interventions.” 

For practitioners and researchers, such a model offers a way to explore how adoption evolves under different conditions and how the odds of adoption can be improved.

Why adoption rises and then falls

The model shows that external pressure—such as the financial penalties from the Centers for Medicare & Medicaid Services—can be a powerful driver of initial adoption. In the case of the readmission tool, CMS penalties pushed hospitals to reduce readmissions, leading to training and early use.

But that momentum doesn’t always last.

“As you succeed, that issue becomes less pressing,” Rockart said. “And the institutional interest wanes.”

As organizational attention shifts elsewhere, training may decline, new staff are likely to be less familiar with the tool, and competing priorities pull attention away from the tool.

For example, the COVID-19 pandemic introduced new and urgent demands that appear to have diverted attention from the technology, the researchers noted. 

At the same time, users often struggle to see the tool’s impact directly.

“It may not be obvious that it actually works,” Rockart noted, especially when the outcomes affected by the tool are themselves rare or delayed.

The research describes this process as the result of two forces:

• Balancing loops, where success reduces urgency and effort;
• Reinforcing loops, where visible benefits encourage continued use.

In the hospital context, internal motivation is sustained when clinicians experience a certain threshold of visible benefits and become the vehicles of positive feedback loops.

When reinforcing loops are weak—as they are likely to be for rare events such as readmissions—adoption of the tool is at a higher risk of falling short of its potential. 

Building a tool for adoption 

Despite focusing on readmissions, the team’s goal is to build a model of system behavior that others can use, Rockart said.

Rather than offering a single recommendation, the model acts as a decision simulator—allowing organizations to test how different choices might play out over time. What happens if training drops off? If competing priorities increase? If users need clearer evidence to believe the tool works?

The vision is similar to interactive tools used in other fields, such as climate modeling—where decision-makers can test scenarios and see the consequences, he said.

By simulating these dynamics, healthcare organizations can design better implementation strategies from the start.

But the lesson applies beyond hospitals. When leaders invest in new technology, they should pay at least as much attention to the human and institutional context as they do to metrics of tool performance.

Ultimately, the success of a technology depends on whether people experience it as useful, visible, and worth their continued attention over time, Rockart said.

*

Nina Rachel Sperber, Department of Population Health Sciences

Sarah Elizabeth Haas, Department of Population Health Sciences

Jiaxin Gao, Interdisciplinary Data Science Program, Social Science Research Institute

Samantha Hamelsky, Trinity College of Arts and Sciences, Department of Statistical Science

Theresa Kiki-Teboum, Trinity College of Arts and Sciences, Department of Chemistry

Afraaz Malick, Trinity College of Arts and Sciences, Department of Computer Science

Rishab Pulugurta, Trinity College of Arts and Sciences, Department of Computer Science

Jacqueline Rodriguez, Trinity College of Arts and Sciences, Department of Public Policy

Hana Shafique, Duke University School of Medicine

Eden Singh, B.A., Duke University School of Medicine 

Kriti Vasudevan, B.S., Trinity College of Arts and Sciences, Department of Computer Science, Department of Public Policy

Shiling Zheng, M.S., Pratt School of Engineering

Scott Rockart, Fuqua School of Business 

David Gallagher, MD, Department of Medicine, Division of Hospital Medicine

Adam Johnson, MD, Duke University School of Medicine