Manage changes to the work environment

Why does this matter?

Implementing a new AI product in healthcare requires consideration of complex work environments. Technology adoption decisions often occur at higher administrative levels and may not consider the benefits for end users and other affected staff. Inadequate support for end users and other affected staff in adapting to the new product can lead to under or over-reliance, inappropriate use (see the guide – prevent inappropriate use of AI ), user resistance, and degraded job performance. Proper institutional support via training, incentives, and feedback can increase the success of implementation by promoting product use, efficiency, healthcare delivery, and return on investment. 

How to do this?

Step 1: Plan the rollout

• Recognize work roles, practices, and routines that need to change to accommodate the new AI product.
• Create a dedicated change management team that includes managers with expertise in human behavior and complex organizations to oversee changes to the work environment. 
• Establish a communication channel that allows end users to provide feedback to those who developed the product. If working with an external vendor, ensure that there is a support contract during the initial rollout.
• Communicate project information (see the guide on disseminating information to end users ), expectations about changes that will occur, and anticipated timelines via existing communication channels and trusted personnel. 
• Undertake scenario design and/or workflow diagramming for various impacted roles prior to the rollout.
• Respect the roles and professional jurisdictions, occupational values, and autonomy of workers to the extent possible. Emphasize the importance of professionals maintaining oversight over the product. 
• Be aware that it may be necessary to establish new staff positions, at least initially, to support the adaptation of work roles to the tool.

“A huge effort in terms of making sure that we had pilot activities, stakeholder feedback, iteration of the tool, making sure that we had the right layers of awareness. So I think in terms of the clinical integration concept, we have generally taken the approach of piloting in certain divisions, making sure that we’re doing phased rollouts, making sure that we have the opportunity for stakeholder feedback in how the tool is performing, or any concerns before we’re rolling it out, then pausing and looking at something if there’s concerns, and proceeding with rollout if we’re not seeing issues.”

Operational Leader

Step 2: Execute the rollout 

• Roll out the new AI product incrementally, as problems introduced at scale have larger consequences. Start in the specific setting, unit, or clinic where the clinical champion works.
• Create two-way feedback mechanisms that allow end users and stakeholders to provide input on the product and how it is being used in the work environment. It helps ensure the appropriate use of the product, identify problems that may arise (see guide on disseminating information to end users ), and empower end users (see guide on designing and testing workflow for clinicians).  

Step 3: Respond to feedback

• Implementations of new technologies or care delivery models are rarely seamless. There are wrinkles to work through and unanticipated challenges.
• As feedback from frontline clinicians comes in, a small group consisting of a clinical lead, technical lead, and administrative lead should meet regularly to prioritize changes. Urgent changes may need to be made immediately, whereas other changes may be planned out with scheduled updates.

Step 4: Update AI product intended use, workflow, and training materials

• As you make changes to the work environment and the way the AI product is used in practice, make sure to update previously completed tasks.
• Update the AI product scope of use (see the guide on define the role of AI), clinical workflow (see guide on design and test workflow for clinicians’), and training materials distributed to frontline clinicians (see guide disseminate information to end users ). Make sure that changes are documented and communicated to affected clinicians.

Step 5: Keep senior leaders informed of changes

• Managing the AI product rollout and making appropriate changes to the product and work environment requires effort from multiple individuals and resources. There may be a need to also engage external consultants or vendors in support contracts.
• Ensuring that senior leaders have visibility into the costs associated with AI product implementation and maintenance will ensure that the organization can most effectively use the new technology.

“For any new implementation, if you want it to be successful, you build on already well known practices. Making sure you have the support, you have the training, you have your champions, you have all those kinds of things that allow for a successful implementation. So I don’t see it as being very different from any good implementation plan, be it for a software solution, or a new process improvement solution, which doesn’t require technology. [it is about] training folks on how to do something they used to do, but just in a different way”

AI Bias and Fairness Expert

References

Blomberg, Jeanette, Mark Burrell, and Greg Guest. 2003. “An Ethnographic Approach to Design.” Pp. 964–86 in The human-computer interaction handbook : fundamentals, evolving technologies, and emerging applications, edited by J. Jacko. Mahwah, NJ: Lawrence Erlbaum Associates.

Brynjolfsson and Hitt (1998) “Beyond the Productivity Paradox: computers are the catalyst for bigger changes”

Debono et al (2013) “Nurses’ workarounds in acute healthcare settings: a scoping review” BMC Health Services Research — https://link.springer.com/article/10.1186/1472-6963-13-175

Levy, Karen (2022) Data Driven: Truckers, Technology, and the New Workplace Surveillance.

Novek, Joel (2002) “IT, Gender, and Professional Practice: Or, Why an Automated Drug Distribution System WasSent Back to the Manufacturer”

Sendak et al (2020) “The Human Body is a Black Box’: supporting clinical decision-making with deep learning”