What the Scribe Leaves Behind

It is hard to think of newer clinical technology that has shifted to the default norm ahead of AI-assisted ambient documentation. AI drafts clinical notes after listening to an encounter, and clinicians regain the time spent typing. The promise is simple: clinicians spend more time with the patient and less time typing. For many, that has been the case.

But what happens next is not as simple.

The work relocates, not disappears

The inclination to consider automation in healthcare as a work removal tool has created a lot of erroneous thinking. Ambient documentation is not an exception. The typing load shifts from the clinician to the reviewer, and they need to ensure that the physician validates the truth of the record. While the typing load diminishes, the responsibility does not.

In some areas of healthcare this trade is more positive, and in others it is less so. For some, a good draft that is mostly accurate saves time, while for others a draft that is inaccurate requires checking, which shifts the burden back to writing. This is the main reason the burden of documentation was always more than just typing. The greatest burden is making sure the record of the clinical encounter is accurate, including codes to support billing, and that it is a record the next clinician can rely on. Ambient documentation can help relieve some of the burden of typing while leaving the burden of record keeping mostly unchanged.

The signed note is still yours

For ambient documentation, the main oversight question is not whether a tool can create a note. It is who owns the note once it exists, and the answer has not changed. Whoever signs the record is responsible for the content.

This is important to understand because a clinical note is not simply the summary of a visit. It influences the next decision and impacts how the visit is coded and billed. This note also becomes a permanent record that may one day be reviewed by an auditor or the court. A generated sentence that seems plausible but was never actually spoken has the same consequence as a sentence that a physician wrote, as long as it remains in the signed record.

So leadership needs to be careful in how they define success. A record that is documented in a shorter amount of time is not necessarily an improvement, and an error-free record is not necessarily more accurate. The physician who documented less may have actually spent more time determining if the machine documented the visit correctly. Trusting the machine is a large act that requires a significant amount of trust, and the signature is that act.

After the pilot, measure the right things

Most ambient documentation pilots measure what is easy to measure, and what is easy to measure is primarily time spent documenting and clinician satisfaction. While those figures are important and should be measured, they should not be the only things that are measured.

The benefits of a new tool are usually overestimated in the initial stages. A motivating pilot program is conducted to test a tool, and a positive response is largely attributable to the testers who had a vested interest in it and wanted it to succeed. After six months, and with a full patient panel, the novelty wears off, and the early positive results are usually not sustained.

Harder questions usually come later, and measuring them is much more difficult. After the initial stages, is the quality of notes still acceptable when clinicians read the notes less and less? What is the relationship between the outcomes and the notes, and is the record of what was said beginning to drift?

Note length is the variable to focus on. When the cost of producing text is almost zero, notes will grow. A note with a single finding can be buried in a long text, and that text shifts the burden of cost to the next person who has to read the note. Savings that are measured at the end of the day become a cost to someone else later on.

Drift is harder to measure. As the signed note moves further away from the original statement, the trustworthiness of the record declines, and no dashboard will show that. It takes intentional sampling to detect drift, with the reviewer comparing the generated notes to the actual encounters rather than to a satisfaction score. The healthcare sector has learned that documentation is an integral part of the safety infrastructure of care, not an administrative afterthought, and has spent decades on this. In this scenario, ambient documentation must be assessed against the quality of the record it retains to justify the time it saves.

What health system leaders really own

The technology will continue to get better. Drafts will be more accurate and easier, and speech recognition technology will be better. What cannot be easier is ownership.

Organizations that get the most sustained benefits from ambient documentation view it as an operational capability and not a software implementation. They name the owner of documentation quality, and continue to analyze drift and note length rather than focusing on the metrics of implementation, reading notes long after go-live. The clinicians on these teams come to view generated content as a draft, rather than a final product.

Ownership is a person, not a policy. It is someone with the authority to impact the tools and is responsible for the integrity of what they produce. Evidence that a note-quality problem is occurring cannot be detected with a steering committee that meets quarterly.

Review also must endure post-launch. Most of the interest occurs during the execution phase, and the project fades into the background once the funding ceases and the tool is considered complete. The risks do not arise when the tool goes live. They appear after the initial excitement wears off and the tool is treated as a final draft. That default state is the biggest risk, since the longer the tool is considered final, the less it is scrutinized, and a final draft is assumed to be ready for signing. The ability to interpret an incomplete draft and the skill of documenting final thoughts is the true challenge, and can be a more advanced skill than traditional writing.

This does not avoid answerability. Documentation remains, and so does the answerability that comes with it. The first generation of these tools told a story that they save time. I believe the more accurate story is that they require us to take more responsibility for their use. The technology will become more advanced and the drafts more rapid, and the responsibility for ensuring the accuracy of the record remains with those who sign it, the clinicians, and the health systems that choose to implement this technology.

Christopher Hutchins Founder and CEO, Hutchins Data Strategy Consultants