AI and Nurse Retention: Closing the Gap That Drives New Nurses Out

Around 50% of new nurses reportedly leave the bedside within the first two years of becoming a nurse. This statistic comes from a recording of a Signal Room conversation with nurse practitioner, nurse educator, and co-founder of Prismn Health Technologies Angela Millan. Millan argues that statistic should not be placed in the Human Resources category. Rather, it should be viewed as a tangible outcome of a problem that takes shape long before nurses leave. Nurses who leave do not fail. They graduate into a discrepancy of what they have been taught and what they are allowed to do, and are expected to navigate the discrepancy without support until they develop their own judgment.

Millan has seen that discrepancy in all roles throughout her nursing practice. At 22 years old, she had approximately 18 days of direct one-on-one preceptorship, after which she was expected to take on a full patient load. She saw that discrepancy again when she moved away from the bedside and became a nurse practitioner. Now, 15 years into her work as a clinical instructor, she is still witnessing the same students freeze in the same place. At Hutchins Data Strategy Consultants, the discussions about staff retention usually stem from a quantitative perspective about the available staff. The episode is a reminder of the root of the retention problem and that AI can be a solution or another problem.

Licensed Is Not the Same as Ready

Millan named the uncomfortable truth that passing the NCLEX means a nurse is "safe" enough to start; it does not mean they are "ready" to care for actual people. Schools teach the nurse to recognize, "This is what the patient is showing you." However, developing the next step is an undercurrent. For each symptom a nurse learns to recognize, there must be an intervention. The friction zone between seeing and doing is where new nurses stall.

Two things missing from the training of nurses are practice and feedback. Practice is the actual doing of the thing on the floor. Feedback is having someone close to you say, "This is what you missed," or, "This is another way to think about it." Lectures and content create the structure for concepts, but the structure doesn't build the reps. This is where judgment is formed. The passing of the NCLEX as a built-in system of clinical recognition, she argued, is a comforting fiction that the profession is telling itself.

The Compression That Eats Mentors

Retention problems stem from more than just nurses quitting. It also concerns the impact on the remaining staff. Millan observed a situation that silently and continuously became worse. As new nurses leave before developing into experienced clinicians, the supply of seasoned mentors thins, and hospitals are forced to have nurses with a mere 1.5 years of experience serve as mentors and train new nurses who are still wrestling with their clinical identity, who have yet to learn how to navigate a complex hospital system, and who have yet to develop the skills and deal with the responsibilities of being a clinical nurse. Additionally, the more experienced nurses who normally would support the unit are either resigning from nursing altogether or retiring to leave the bedside as well. This causes the entire workforce to be squeezed from both ends, with a staff that is overextended and has to balance training and nursing with little experience. Millan has described this unfortunate situation as "survival" instead of "thrival." The most concerning aspect is that if you learn and teach in survival mode, it causes burnout that feeds attrition into the next generation.

When the Gap Costs a Patient

The stakes are tremendously real. Millan shared an example from a colleague's student to illustrate the recognition-to-action gap. The student was responsible for a nursing-home patient with diabetes and dementia who was unable to self-administer his medications. The student was knowledgeable about the insulin as well as the dose and even checked the patient's blood sugar level prior to administration; according to Millan, she almost did everything correctly. She, however, failed to consider that the rapid-acting insulin she was administering worked in about 15 minutes and was intended to cover the blood sugar rise after meals. The student left the room and failed to set up the patient's meal tray. About 30 minutes later, a certified nursing assistant found the patient unresponsive. The patient went into hypoglycemic shock and had to be transported to the hospital.

The knowledge was all present. What failed was the connection between a recognized fact and the anticipatory action it demanded — exactly the kind of cause-and-effect foresight that practice and feedback are supposed to build, and that a compressed orientation rarely has time to. This is the example of one practitioner, not a published study, but it captures the mechanism behind the turnover number better than any chart could.

The Clipboard Problem

Millan was quite explicit in saying that part of the reason nurses leave is emotional. People who choose this profession do so for their passion for caring for patients, and then the system, in her words, meets them with a clipboard. The sorts of questions asked on the unit don't ask whether you spent time with a patient, whether they understood their discharge instructions, or how you are holding up after the patient you were caring for passed away — but rather, did you chart it, did you chart it on time. The metrics will not see that a patient who could not read the discharge paper they were handed is a failure.

That is the cost she brought to the hospital executives. Nurses are people and are not robots, as she said, and have a breaking point. On top of any task or metric you give to a nurse, something must be taken away to keep the system working. Retention comes from respecting this constraint, and won't happen if you keep piling things onto people who are already stretched thin.

Where AI Helps the Workforce — and Where It Hurts

This is the crux for those looking to implement AI in nursing contexts. Millan was not against the use of technology; she was more pointed about the direction in which it was headed. AI scribing and documentation tools, she mentioned, aid nurses to function more efficiently. However, tools that add efficiency do not close the gap of workforce readiness which pushes people out. Further, adding a new tool on top of an already burdened nurse can become an added burden of another thing to learn, instead of relief.

The use she is building toward with Prismn runs the other way. Their tutor, Hazel, is a conversational AI and is designed to teach the way a good nurse educator teaches. A good nurse educator does not give the answer; instead, they work to confirm the learner truly understands, and they adapt to the learner, using clear examples and simple analogies. A clinical reasoning lab walks a student through a patient case in real time, lets the decision be made, and then debriefs to explain where the decision and reasoning went wrong. The practice that a class of 30 students or an orientation of 18 days cannot provide is done where the stakes are zero.

Two guardrails are essential, both belonging to Millan. Hazel is purely educational and does not provide real-time clinical bedside or medical suggestions; if prompted with a live patient case, it will not advise to supply oxygen or take any clinical measure. Her deeper line is that clinical reasoning is not to be handed over to AI. The objective here is to strengthen the cognitive muscle of the nurse, and not to eliminate it and ultimately relegate the nurse to a mere button pusher. AI is justified in the workforce if it strengthens the clinician's capabilities — not when it silently thins out the thinking that made them a clinician.

Nurse Retention and AI at Hutchins

We treat workforce retention as a problem of data and design, and not just recruitment. Instead of asking a health system where its nurses are going, it is more helpful to ask where readiness, workload, and tools are failing the staff at the bedside — and whether the AI under consideration is building capacity or increasing the workload. We apply the same distinction to any clinical AI: a tool is only responsible if it strengthens the judgment of the person using it. We connect our work to AI literacy in healthcare, which is what lets a workforce use new tools well, and to responsible AI in healthcare, which sets the guardrails — like keeping clinical reasoning with the clinician — that decide whether a deployment helps or harms. These themes run throughout The Signal Room podcast, where practitioners describe what supporting a clinical workforce actually takes.

How Hutchins Approaches Nurse Retention and AI

We treat workforce retention as a data and design problem, not only a recruiting one. When a health system asks where its nurses are going, the more useful question is where readiness, workload, and tooling are failing the people at the bedside — and whether the AI being introduced is building capability or merely adding load. That distinction is the same one we apply to any clinical AI: a tool is only responsible if it strengthens the judgment of the person using it. Our work connects to AI literacy in healthcare, which is what lets a workforce use new tools well, and to responsible AI in healthcare, which sets the guardrails — like keeping clinical reasoning with the clinician — that decide whether a deployment helps or harms. These themes run throughout The Signal Room podcast, where practitioners describe what supporting a clinical workforce actually takes.

Authoritative sources

• World Health Organization — Ethics & governance of AI for health
• Agency for Healthcare Research and Quality
• NIST AI Risk Management Framework
• HIMSS
• NEJM AI