By: The Capella University Editorial Team with Owen M. Borda, PhD, Assistant Professor/Lead Faculty, Doctor of Health Administration, Capella University
Reading time: 7 minutes
Most people don’t go into healthcare for the paperwork. They want to spend their days helping patients. That’s why AI is an important conversation in healthcare right now. Tools that take over notetaking, billing codes and scheduling could give clinicians and staff have more time for the people they’re caring for.
Here are four steps toward adopting AI in a healthcare organization: choose one problem and the right people, tell patients how you use it, run a pilot and prepare staff before you scale.
How is AI used in healthcare today?
AI in healthcare generally falls into two groups: tools that handle routine administrative work and tools that support clinical decisions. Common examples include:
- Clinical documentation. Ambient software records a visit with the patient’s permission and drafts the note for the clinician to review and sign.
- Medical coding and billing. AI suggests codes from the record. Coders still confirm each one, because the organization is responsible for every claim it submits.
- Scheduling and patient messages. AI drafts replies for staff to review and helps fill open appointments.
- Medical imaging. AI flags areas on scans for a radiologist to review.
Any AI that handles patient information must be used in line with HIPAA and federal or state confidentiality or privacy statutes, so make sure your team knows the rules before you start.
Step 1: Choose one problem, bring in the right people and set the rules
Organizations that adopt AI well don’t start by asking, “Where can we use AI?” They ask the people doing the work what slows them down. Nurses, medical assistants, schedulers and billers know which tasks take the most time.
Pick one problem you can describe in a sentence, such as “clinicians spend too long on documentation after visits.” A well-defined problem gives you something concrete to measure later. Measure how things work today, before the AI arrives, so you have a baseline to compare against.
Then look at a few things before choosing a product:
- Risk. Administrative tasks such as scheduling or drafting notes carry less risk than AI that shapes diagnosis or treatment. Many organizations start there.
- Fit. Does it work inside your electronic health record, or does it add another login and another screen?
- Full cost. Count licenses, IT support, training and the staff time it takes to review the AI’s work.
- Ownership. Name one leader who is responsible for it. If no one owns it, don’t start.
Gather input beyond the frontline team, too. Clinical, compliance, technology and patient perspectives each catch risks the others miss. An AI oversight committee, or an existing quality or compliance committee, can bring them together.
Set the rules before anyone uses AI
A pilot without rules puts patient information and patient care at risk. Guardrails tell staff what it can be used for, who reviews its output and who answers for mistakes. Settle these questions first:
- Privacy. Where does patient data go, who can see it and how long is it kept? Will the vendor sign a business associate agreement, as HIPAA requires, and does it use your data to train its products?
- Accuracy. Who checks the AI’s output and how often?
- Bias. Does it work equally well for different patient groups? The Patient Protection and Affordable Care Act prohibits discrimination through the use of patient care decision support tools.
- Accountability. Who is responsible when the AI is wrong? The clinician who signs a note or the coder who submits a claim still owns that work.
Step 2: Tell patients how you use AI
Patients want to know. In a Pew Research Center survey published in August 2026, 72% of U.S. adults said it is extremely or very important that a provider tell them if AI is used in their care. Yet 46% weren’t sure whether their providers had used AI at all.
Telling patients isn’t the same as helping them understand. Among adults who said their providers had used AI, only 22% said they understood extremely or very well how it was used. Good communication includes:
- Plain language. Explain what the AI does, that a clinician reviews its work and what happens to the patient’s information.
- The point of care. A line in a privacy notice isn’t enough. A clinician might say, “I use software that helps draft my notes. I review everything. Is that OK with you today?”
- A real choice. Let patients decline where you can, especially when a visit is being recorded.
- Prepared staff. Front desk and nursing staff should be ready to answer, “Is a computer making decisions about me?”
- Clear materials. Offer explanations at different reading levels and in the languages your patients speak.
Check your state’s rules, too. Some states require consent from everyone being recorded. California, for example, requires a disclaimer on AI-generated clinical messages to patients unless a licensed provider has reviewed them, and Texas now requires providers to tell patients when AI is used in their care.
Step 3: Pilot with the people who will use it and measure the results
Include the staff who do the work and ask them what works and what doesn’t. A well-run pilot has:
- A set scope and timeline. One team, one workflow and enough weeks to see how use changes once the novelty wears off.
- Measures chosen in advance. Track efficiency, such as time per task or wait times, alongside safety, such as errors found when staff review the AI’s work.
- A fair look across patients. Check whether results hold for different patient groups, not just on average.
- A decision rule. Agree before you start what results will lead you to scale up, adjust or stop.
Watch for these mistakes when judging a pilot:
- Trusting the vendor’s numbers. Results from a vendor’s testing may not hold up with your patients and your workflows. Before you rely on them, test the AI on your own data.
- Measuring only speed or satisfaction. Staff may like a system that saves time while errors slip through.
- Ignoring review time. If clinicians spend 10 minutes fixing a note that saved them 12, the gain is small.
- Choosing only enthusiasts. A pilot team of early adopters can make the AI look better than it will for everyone else.
- Stopping too soon. Over time, people may start skimming the AI’s work instead of checking it. A short pilot won’t show that.
Step 4: Prepare staff, build trust, then scale
Before staff use the AI, they should understand:
- What it’s for. The tasks it was designed for and what it shouldn’t be used for.
- How it fails. Common errors, such as details in a note that no one said or a code that doesn’t match the visit.
- Who owns the result. The person who signs or submits the work is responsible for it, not the AI.
- How to push back. How to report a problem, and that overriding the AI is expected, not penalized.
Trust comes from how leaders handle the rollout, not from training alone. Leaders can:
- Share pilot results openly, including what didn’t work.
- Let respected frontline staff help lead training.
- Show that problem reports lead to real changes.
- Be clear about the goal. If time saved simply becomes more appointments, staff will notice.
- Avoid using the AI to monitor staff performance without telling them.
As you scale, share the pilot results with each new team so they know what to expect and keep measuring after launch. Tools and workflows change, and what worked in the pilot may not hold up.
