By Betsy Castillo, Vice President, Clinical Data Abstraction, Carta Healthcare
There is a familiar arc when AI arrives in a clinical team, and it does not begin where vendors wish it did. It begins with a quiet, well-earned skepticism. The people who do the work have seen tools arrive before with bold promises, and they have learned to wait. Some arrive at an even sharper version of the same feeling, closer to you are going to replace us than to where do I sign. Ignoring that reaction is a mistake. Understanding it is the beginning of real adoption.
Clinical abstractors are not resistant to technology out of habit. They are protective of accuracy because they are the ones held responsible for it. They know how easily a confident wrong answer can slip into a registry, and they know that they, not the software, will answer for it when an auditor asks. From that vantage point, a tool that promises speed is not automatically good news. It is a new source of error to supervise, and possibly a threat to the role itself.
This is worth saying plainly because it reframes the adoption problem. The barrier is rarely a lack of technical ability. It is a reasonable fear that the tool will be wrong in ways the abstractor will have to catch, or that it will be trusted too much by people who do not understand the work. Both fears are legitimate, and neither is overcome by a better demo.
Trust does not arrive through persuasion. It arrives through experience, and the experience that matters most is the abstractor staying in control. When the professional remains the one making the final call, the tool stops being a rival and becomes an instrument. The model surfaces candidate values, flags inconsistencies, and compresses the hours of scrolling through patient history. The abstractor verifies, corrects, and decides. The first time the tool saves real time on a case without introducing a mistake the abstractor did not catch, the relationship begins to change.
The survey data underlines what drives this shift. Sixty-seven percent of healthcare professionals say human review or validation is the single biggest factor that increases their trust in AI outputs. Sixty-four percent say AI delivers the most value when it accelerates the work and clinicians validate the results. Trust is not built by removing the human from the loop. It is built by keeping the human unmistakably in charge, with full visibility into how a result was produced.
What is striking about teams that cross this threshold is how far the pendulum swings. The same abstractors who started by guarding against the tool often end up unwilling to work without it. Once the technology has proven that it speeds the tedious parts without compromising the judgment-heavy parts, going back to fully manual work feels like working with one hand tied. The skepticism does not fade into neutral acceptance. It inverts into genuine preference.
That arc, from suspicion to reliance, is the real adoption curve, and it has a shape worth respecting. It cannot be skipped by mandate. A team told to use a tool they do not trust will comply minimally and disengage. A team that discovers, case by case, that the tool makes them faster and no less accurate will adopt it on their own and defend it.
The shift is easiest to see at the moment the tool disappears. One abstractor, a nurse with more than three decades of experience, described an evening when a network outage cut off access to both the electronic health record and the abstraction tool, leaving eight cases due that night. She had spent most of her career working manually and assumed reverting would be easy. Instead she realized how completely her workflow had come to depend on the tool's fast, source-linked answers, and how laborious it felt to go back to reading lengthy progress notes and comparing lab values across days by hand. By then her time on a routine case had fallen from more than thirty minutes to fifteen or twenty, and a complex case from five hours to roughly ninety minutes, while her inter-rater reliability scores improved rather than slipped. The dependence she felt that night was not a weakness. It was the considered judgment of an expert who had tested the tool against her own standard and found it worth relying on.
For anyone introducing AI into a clinical workflow, the implication is that change management is not a soft add-on to the technology. It is the core of whether the technology succeeds. The teams most likely to win are the ones who treat the abstractor's skepticism as information rather than obstruction, who design the workflow so the human stays at the helm, and who let trust accumulate through evidence instead of demanding it up front.
It also argues against the framing that pits people against automation. The goal is not to overcome the clinical team's resistance so the tool can take over. It is to give the clinical team a force multiplier and let them feel the difference. When that happens, the question stops being whether staff will accept the tool. It becomes whether they would ever give it back. In practice, the ones who were most skeptical at the start tend to give the most emphatic answer.
There is a detail in the data that complicates the easy story about clinician resistance. Most abstractors are, in fact, open to AI. What many of them lack is access to tools they can actually use in their daily work. The gap between willingness and adoption is often not about attitude at all. It is about deployment, about whether the organization has put a usable, trustworthy tool in front of the people who would benefit from it. Framing the challenge purely as overcoming skepticism misreads a workforce that is, in many cases, waiting to be equipped.
Where trust does have to be earned, the mechanism is visibility. People come to rely on a tool when they can see how it reached a result, check it against the source, and correct it when it is wrong. Sixty-seven percent of healthcare professionals name human review and validation as the single biggest factor that increases their trust in AI outputs, and that preference is really a preference for transparency and control. A black box that produces an answer invites suspicion. A tool that shows its reasoning and leaves the final call to the professional invites adoption.
For leaders, the practical implication is that the path to adoption runs through the people closest to the work, not around them. The teams that succeed treat the abstractor's caution as design input, build the workflow so the clinician stays at the helm, and make sure the tool is genuinely available rather than merely announced. Do that, and the workforce stops being an obstacle to manage and becomes the strongest advocate the technology has.
Technology adoption in high-stakes clinical work is ultimately a trust problem wearing a technical costume. The features matter, but they are not what carries a tool from pilot to daily reliance. What carries it is a series of small, verifiable moments in which the people responsible for the data learn that the tool makes their judgment go further without ever asking them to surrender it. Earn that, case by case, and skepticism becomes the strongest endorsement a team can give.
