AI in Healthcare-: Practical Use Cases, Benefits and Challenges for Healthcare Organizations

Learn where AI in healthcare delivers value today, what the evidence shows, the risks to plan for and how to start with one clear use case.

Manpreet

11/16/20266 min read

Most healthcare leaders no longer ask whether AI is coming. Their staff are already using it. The harder questions are where it helps, where it creates risk and what to do first. AI in healthcare means software that spots patterns, predicts outcomes or writes and summarizes content to support clinical and administrative work. Today the clearest value is in documentation and information tasks, where a person still reviews the result. Tools that influence diagnosis or treatment need more evidence and tighter oversight. The pace of adoption is fast. In the American Medical Association’s 2026 survey, 72% of physicians said they use at least one AI use case in their practice, up from 38% in 2023. That figure is self reported, but the direction is clear.

Where AI is being used in healthcare today

The AMA asked physicians which of 17 AI use cases they actually use. The table shows the most common ones and what to watch for in each. According to the AMA's 2026 Physician AI Sentiment Report, physicians are using AI across several clinical and administrative tasks. The most common use is summarizing medical research and standards of care (39%), followed by preparing discharge instructions, care plans, and progress notes (30%). Around 28% use AI for billing codes, medical charts, and visit notes, while another 28% use it for chart summaries. Other applications include drafting replies to patient portal messages (19%), translation (18%), and assistive diagnosis (17%). However, each use requires appropriate oversight. Physicians should verify research summaries against original sources, review clinical documentation before filing, check coding accuracy, watch for missing chart details, protect patient privacy, validate clinical translations, and continuously monitor AI-assisted diagnostic outputs. Notice the pattern. The most common uses sit in documentation and summarization. The use cases closest to diagnosis are the least adopted.

Ambient documentation

Ambient AI scribes listen to a visit and draft the note. They are the best studied tool in this list. A randomized trial of 238 outpatient physicians compared two scribes with usual care. Both scribe groups improved on burnout, cognitive load and work exhaustion, but users also reported clinically significant inaccuracies, including one adverse patient safety event. A multi site study from Mass General Brigham found gains that were real but modest. Clinicians who used scribes in more than half their visits saw about three times the drop in documentation time, yet only 32% of users reached that level of use. The lesson is that results depend on how deeply the tool is built into daily work.

Summaries, discharge paperwork and chart review

Much of the rest of the list is about turning long records into short, usable text. Examples include summarizing a chart before a visit, drafting discharge instructions and pulling the key points from a research article. These tasks are time consuming and easy to check, so a human can review the output quickly.

Patient communication and engagement

Draft portal replies, translation and chatbots all fall here. This area also fits the broader push toward easier access, such as teleconsultation and remote monitoring. AI works best when it connects to the workflows you already run, rather than adding another standalone tool.

Decision support and risk prediction

Assistive diagnosis and risk scoring have the highest potential and the highest stakes. If you move here, expect to ask for validation evidence and to monitor performance after launch.

Benefits the evidence supports

Less documentation work. Documentation is a leading driver of physician burnout and it is where the strongest trial evidence sits. Just as important, physicians expect AI to ease burnout, with 70% expecting it to automate clinical tasks and 73% expecting it to automate administrative duties.

Better access to information. Summarizing research and charts helps busy clinicians find what matters faster. That is likely why it leads the adoption list.

A more confident workforce, when it is done well. In the AMA survey, 76% of physicians said AI gives them an advantage in caring for patients, up from 65% in 2023

For community hospitals and clinics with thin margins, the appeal is simple. If AI gives staff time back, that time can go to patients. But the benefit only appears when the tool fits the workflow and people actually use it.

Challenges to plan for

Accuracy and safety

Generative tools can produce confident mistakes. The scribe trial above is a good reminder. Every output that touches a patient record needs a clear review step and staff need to know that review is part of the job.

Privacy and security

Privacy was the only area in the AMA survey where more physicians expected harm than help. Concern was also much higher for tools not provided by the physician’s own institution, at 71%, compared with 42% for institutional tools. That gap is a strong argument for giving staff approved tools so they don’t reach for unapproved ones.

Existing HIPAA rules still apply to AI tools that handle patient data. The proposed overhaul of the HIPAA Security Rule has been pushed to a long term agenda, with final action estimated for July 2027, but the current rule stays fully enforceable. OCR’s enforcement continues to focus on thorough risk analyses, vendor risk management, staff training and fixing known vulnerabilities. Any AI vendor you add belongs in that risk analysis.

Less federal transparency for AI built into EHRs

This is a recent change that many organizations have not yet absorbed. On October 28, 2026, ONC finalized a rule called HTI 5. It removes the requirement that certified EHRs provide “model card” information about embedded predictive tools, including training data, performance and limitations. The changes take effect on December 28, 2026 and a large majority of commenters had opposed them, warning that evaluation and liability would shift to providers and health systems.

In practice, you can no longer assume a federal baseline of information about AI inside your EHR. Health systems will need to seek that information through contracts and vendors that keep publishing model cards voluntarily may stand out in procurement. Before you switch on any predictive tool, ask the vendor.

  • What data was the model trained on?

  • How does it perform across different patient groups?

  • How often is it updated and how will you be told?

  • Who monitors it after launch and what happens when performance drops?

Skills, training and trust

Nearly nine in ten physicians (88%) report at least mild concern about losing skills to AI and 92% want more training. About a quarter have had no AI training from any source and 85% want to be consulted on or responsible for adoption decisions. Tools rolled out without staff input tend to be ignored or misused.

Accountability

Physicians ranked clear liability frameworks as the regulatory step most likely to build their trust. Until those frameworks exist, your organization needs its own answer to a simple question who is responsible when an AI assisted decision goes wrong?

Data readiness and integration

AI is only as good as the data and systems behind it. Seamless EHR integration was among the top requirements physicians named for adopting a tool. Messy data and disconnected systems undermine a tool before it starts.

How to start with AI in healthcare

  1. Pick a problem, not a tool. Name one pain point, such as note writing time or discharge paperwork and describe what better looks like.

  2. Start where a human reviews the output. Documentation and summarization are lower risk starting points than decision support.

  3. Check data and workflow readiness. Look at data quality, EHR integration and who will use the tool day to day.

  4. Set simple governance. Decide who approves new tools, who reviews outputs and how problems get reported.

  5. Bring clinicians in early and train them. Involve the people who will use the tool before you buy it.

  6. Measure, then review. Record a baseline first, such as minutes per note or share of visits where the tool is used. The Mass General Brigham result shows why usage rate belongs on that list.

When AI consulting for healthcare makes sense

A good starting point is an outside view when AI questions are piling up and no one owns the answers. AI consulting for healthcare typically covers readiness assessments, picking and ranking use cases, strategy, solution design, implementation support and governance.

TransformativeHLTH offers this as part of its healthcare digital transformation services. Its AI work starts with readiness, covering data, infrastructure, workflows, people and governance. It then moves to high value use cases and a roadmap, with attention to privacy, security and change management. The practice leads with a plain question where can AI create meaningful value for our organization? Its Transformation Table is a focused session where leaders can work through that question, along with vendor choices and legacy system constraints.

Technology can be evaluated by others, but accountability stays with your leaders. A consultant should help you ask better questions and weigh trade offs, not make the call for you. TransformativeHLTH also provides healthcare advisory services for independent vendor and investment decisions and engagement support for ongoing guidance after planning. To talk through where AI fits your organization, book a 30 minute call with Paul McRae or contact the team.

Innovation

Empowering health systems through strategic digital transformation solutions.

Support

© 2026. All rights reserved.

TransformativeHLTH  logo
TransformativeHLTH  logo