Sell AI controls, not magic

The pitch for enterprise artificial intelligence is shifting from flashy demos to control panels: buyers want proof that outputs can be checked, flagged, corrected, and tied to results. (cio.com) In healthcare, that shift is showing up in purchasing data. Qventus said on April 9 that it surveyed and interviewed more than 60 senior health system technology leaders, and only 4% said they had scaled artificial intelligence with measurable outcomes. (qventus.com) The same report said 45% of respondents were struggling to move beyond pilot programs, 74% cited dependence on electronic health record vendors as a top barrier, and 94% said delays would create a competitive disadvantage. (healthcareitnews.com) Hospital technology leaders are describing the problem in operational terms, not science-fiction terms. Parkview Health executive Mark Mabus told CIO that before deployment his team asks where audio is processed, whether protected health information is retained, and who validates the output. (cio.com) That language now matches formal guidance. The Joint Commission and the Coalition for Health AI said on September 17, 2025 that their first national guidance for hospitals centers on policies, local validation, monitoring, and use, with governance playbooks and a voluntary certification program slated to follow. (jointcommission.org) The broader standards world is moving the same way. The National Institute of Standards and Technology says its Artificial Intelligence Risk Management Framework is organized around four functions — govern, map, measure, and manage — and the World Health Organization says health systems need governance and accountability for artificial intelligence used in care. (nist.gov; who.int) The reason buyers are asking harder questions is that impressive tools can produce messy economics. Futurism reported on April 11, citing Stat and healthcare executives, that ambient documentation tools have raised costs in some settings by supporting higher-complexity billing and letting clinicians see more patients. (futurism.com) At one health system cited by Futurism, FMOL Health said clinicians using artificial intelligence scribes saw 22% more patients overall. Caroline Pearson of the Peterson Health Technology Institute told Stat, as quoted by Futurism, that “ambient scribes are inflationary.” (futurism.com) That is pushing vendors to sell audit trails instead of awe. In healthcare, “human in the loop” means a clinician can review, edit, and sign off before an output becomes part of the record, and Parkview told CIO that higher-risk use cases face stricter review based on clinical impact and automation level. (cio.com) The Food and Drug Administration is also leaning on transparency. Its artificial intelligence-enabled medical device list is meant to show which products are authorized for marketing in the United States and to give providers and patients a clearer view of where artificial intelligence is being used. (fda.gov) The sales message is getting simpler as deployments get bigger: show the exception queue, show the validation step, show the override button, and show the numbers. Health systems are still buying artificial intelligence, but the evidence they are asking for now looks a lot more like governance than magic. (healthcareitnews.com; jointcommission.org)