Focusing on Employee Roles and Status Will Favor a Successful Tech Implementation

By now, leaders are savvy enough to know that implementing new technology requires buy-in from powerful stakeholders, fit with the current workflow and training of employees. Yet, even if a company checks all these boxes, an implementation can be a dismal failure. Professor Kate Kellogg’s recent healthcare research uncovers other crucial considerations for successful technology change. Companies must account for challenges to employees’ roles and status, Kellogg says. Paying attention to these non-technical, non-operational characteristics is necessary for a successful implementation.  

Kellogg's research addresses four mistakes not related to technology that managers make and what to do instead. Kellogg recommends that managers rotate the role of trainer, introduce new technological roles with care, and create a central group to incorporate input from frontline staff. And if the technology is a machine learning tool, managers must also provide the time and resources to encourage give and take between users and developers of the new technology.

Trainers need to be attuned to the status of the trainees

Using tech-savvy employees to teach their colleagues may seem like an efficient way to roll out a new tool. However, this strategy can backfire. Kellogg and her team studied the rollout of a new Electronic Medical Record (EMR) technology across multiple medical clinics. At most of the clinics, leaders chose younger, tech-savvy employees to be trainers, because they thought these employees would be quick to learn the new technology. Patient Service Representatives (PSR) needed to be trained on the new system and learn new ways of completing some of their tasks. The PSRs generally had long tenures and possessed tacit institutional knowledge that they used daily to be effective at their work. They were not particularly comfortable with the technology or with changing their ways of working. They viewed the junior trainers as challenging their status, as if they were being told that what they had been doing was no longer valid or valued. As such, they were not particularly open to being trained.

Kellogg recounted one trainer saying, “I work with veterans. They’ve been here forever. They have created their own standard work, and they own it. It is a challenge to try to train them. There is a lot of swearing… They say, “Oh are you kiddin’ me, I’m doin’ this and now you’re tellin’ me I need to do that?!”  A trainee echoed: “There’s been a little bit of backlash about it...Some trainees are saying, ‘Why are you the one who’s telling us this?’”

Other clinics introduced the technology differently. There, managers rotated the role of trainer. Some of the long-time employees were given the resources, help and time to learn the new system well enough to train others. Rotating the trainer role facilitated openness and acceptance by the longer-tenured trainees. While it did cost both time and resources to develop a training course that allowed less digitally-savvy employees to do the teaching, in the end managers who made this upfront investment garnered the benefits of the technology more quickly and completely. This may not seem like the most efficient way to do things, but in practice, it was far and away more successful.

Be mindful when integrating new technology roles

New technology sometimes requires new positions. If managers hire data science experts to work with the current staff, they must manage the introduction and integration of the new staff carefully to minimize disruptions. Introducing specialists or superusers to provide technical expertise or take on new tasks may spare current employees from having to learn new technical skills, but it can be seen as a challenging the current staff’s ability and know-how. Kellogg learned through her research that managers must mitigate expected conflict when they add roles to help with technology implementation.

If the data scientists are going to create new tools with the technology, they must do so with the input of those they are purporting to help. The object is to make the staff more effective, so a conversation about what that looks like and what is needed is important. The data experts can talk about what is possible; there may be capabilities that the current staff wouldn’t have considered. The current staff can highlight limitations, restrictions, needs or idiosyncratic elements that would not be obvious to developers.

New technology positions may not only be filled to provide expertise. These new employees may also take on additional tasks that are required by the new technology. Here too, there is ample opportunity for missteps. Kellogg discusses a study of the implementation of a new machine learning tool at Duke University Hospital to help ED clinicians better detect and manage sepsis. Clinical leaders knew that ED clinicians focus on triage care and already have many alarms and decision-support tools vying for their attention. Rather than tasking ED clinicians with the additional steps required for tool use, clinical leaders asked some of the existing nurses on the Rapid Response Team (RRT) to help with some of this extra work. These RRT nurses use the tool to monitor patients remotely. They contact the clinician if they think a patient is at risk for sepsis, and if the clinician agrees, they make sure that the appropriate sepsis care is implemented and completed.

Some ED clinicians initially resisted. They didn’t like the idea of other people, who didn’t have eyes on the patient, reaching into their area and trying to shape their decision making. Fortunately, the clinical leaders took specific steps to allay clinicians’ fears that the nurses wouldn’t be trying to dictate care from afar, and instead positioned them as an extra resource for busy ED clinicians. They chose nurses with high emotional intelligence who could finesse the interactions with clinicians and learn the best way to intervene. In the end, ED clinicians saw the remote nurses as an extra resource to help with this potentially lifesaving task, so the clinicians could focus on other complex, challenging issues.

Employees at all levels need to be on board

Leaders need to secure the cooperation from powerful stakeholders for a successful technology implementation. However, just as important is to win over the less senior, less powerful workers. Often these are the ones whose jobs will be most directly affected by the technology. They, not the powerful stakeholders, will have to become expert at the new tool. Often their jobs become more difficult, not less, especially in the short term. Kellogg has found that frontline staff often resist new technology or leave the company altogether.

Kellogg studied a set of primary care departments in a medical center during the introduction of a technology to flag patients who needed vaccinations and specific medical tests. The clinicians embraced the tool even though it didn’t provide all the information they wanted. To make up for the gap, the clinicians tasked frontline staff, Medical Assistants (MA), with digging through patient files to read the free-text portions and find information that would make the technology most useful for the clinicians. Doctors saw this as a good fix for a shortfall in the technology’s functionality. However, the unintended result was MAs reporting to their managers that they did not have enough time. Several of the best MAs quit. This is a case of implementing a new technology without considering the effect on people with less status. Without those employees on board, the technology is not going to deliver the best results.

Managers addressed this challenge by creating a central group that approached the problem on two fronts. First, the group revised the decision support rules so clinicians would receive more complete information. With more pertinent information, clinicians were less likely to ask the MA to dig into patient charts. The second change was to add several options that would help the MAs with their extra work. Rather than having to explain vaccinations to patients, a task given to them by the clinicians, MAs could print out fact sheets to give to patients. So, by removing some of the new tasks and making others easier to complete, the central group made the MAs’ workload more manageable and the clinicians more satisfied with the new technology.

Implementing a machine learning tool requires back-and-forth dialogue between developers and users

When Kellogg and her team studied the rollout of two machine learning (ML) tools at Westchester Medical Center Health Network (WMCHN), they learned that introducing a machine learning tool requires an iterative conversation with developers that is not common with a traditional technology deployment. Nobody – developers or users – may be aware beforehand of the time and patience that are required for this cooperative implementation.

Unlike traditional technology, ML technologies require a large amount of training data. Collecting, curating, reconciling, and assuring that the data reflect the target population requires input and collaboration from both developers and users. Another characteristic of ML technology is that its reasoning is often opaque, and users might be reluctant to adopt recommendations from a “black box.” Developers and users must work together to validate the tool so that the result is something that is trustworthy and useful.

At WMCHN, Kellogg and her team studied how the two groups worked together to develop a tool for identifying those patients most at risk for readmission. Care managers took the time to review charts and recommendations to help the developers fine-tune the model. During this back and forth, care managers realized that crucial data was in the free text portion of the medical record. Passing along this insight allowed the developers to address this problem to create a better predictive tool.

Kellogg says ML implementations require this additional work to be successful. Managers need to prepare staff for the conversations with developers and the extra work required to assess the accuracy of the tool’s recommendations. Without this groundwork done well, the organization will not get the best value from an ML decision support tool.

Professor Kellogg’s research on the less obvious, less technical requirements for technology success highlights the importance of the impact of roles and status. It is not enough to choose a technology system or product based on technological function or even on what will fit in best with the current workflow. A successful implementation requires more nuanced emotionally intelligent considerations. Managers must consider more “soft” attributes and consequences of their implementation choices for those who are most affected. Often these are employees with the least power in an organization, but whose buy-in is crucial. Managers who hope to successfully bring on board new technologies need to focus on issues of employee status and roles, and the amount of new work that will need to be done.