Summary
We measured note draft latency for a year. Seconds from the consultation ending to a draft on screen. It improved every quarter and we put it in decks.
It measured the fastest part of the job. We now measure consultation-to-closure. It is the elapsed time from the patient standing up to the last artefact from that encounter being signed rather than drafted. We think every vendor in this category should publish it, including us.
The metric that matters is when you are finished, not when the draft appears.
Book a demoThe metric we reported
Note draft latency. A clinician on a customer call asked how long it took her to finish with a patient, and our dashboard could not answer. We had latency. We had draft quality scores, adoption curves, weekly actives, a churn cohort chart. Not one of them could tell her when she would be done with the person who had just walked out.
What we count now
Consultation-to-closure covers the clinical note, the referral letter, the clinic letter, the ICD-10 coding, the results acknowledgement including the eGFR nobody has opened, and any callback the encounter generated. Signed, not drafted.
No vendor in this category publishes it, including us until now.
The method
The method sits in the margin so it can be argued with.11Method, for sign-off. Consecutive consultations, de-identified at source, no exclusions. Measure is elapsed time from consultation end to each artefact being signed. Comparator is the same clinicians 30 days pre-onboarding. Replace with real figures or do not publish. Consecutive consultations rather than a selected sample, de-identified at source, and the same clinicians as their own controls in the 30 days before onboarding.
We have not published our own figures on this page. Writing a number here that we have not yet measured across consecutive encounters, in an article arguing that vendors stop publishing numbers they cannot stand behind, would be the exact thing the article objects to.
What we will do instead is report it to you. Every 90-day pilot returns consultation-to-closure for your own clinic, by artefact type, against your own 30 days before onboarding, using the method in the margin. That number is yours whether or not you buy anything, and no version of this figure from anyone else should persuade you as much.
Why nobody publishes it
Note draft latency makes every product in this category look good, so every product in this category reports it. Consultation-to-closure does not. A scribe produces a draft in nine seconds. The two-week-wait referral from the same encounter is still unsigned at seven in the evening.
The UCLA trial measured both. Note-writing time improved by 41 seconds. After-hours record time did not change.22Lukac PJ, et al. NEJM AI 2025;2(12). NCT06792890. One of those two numbers appears in the marketing of the product that won.
The number we should be judged on
WA\ is built to move consultation-to-closure, so we should be held to it, including when it embarrasses us. Our 90-day pilots report it back to the clinic alongside revenue, hours saved and patients cared for. If you run a practice, the number worth having is your own rather than ours.
What we are asking for
Every vendor in this category should publish consultation-to-closure, with the method attached, on consecutive encounters rather than a chosen sample.
Sinsky put the desk-to-patient ratio at close to two to one in 2016.31Sinsky C, et al. Ann Intern Med 2016;165:753-760. Ten years and roughly $5 billion of annual recurring revenue later, the metric this industry reports is how fast a draft appears on a screen.
Frequently asked questions
What is consultation-to-closure?
The elapsed time from the patient standing up to the last artefact from that encounter being signed. Not drafted, signed. It includes the note, the referral letter, the clinic letter, the ICD-10 coding, the results acknowledgement and any callback the encounter generated. It is the only metric that corresponds to whether a clinician has actually finished with a patient, and no vendor in this category publishes it, including us until now.
Why is note draft latency a misleading metric?
Because it measures the fastest part of the job and calls it the job. A scribe can produce a draft in nine seconds while the referral letter sits unsigned at seven in the evening, and note latency will report that day as a triumph. It makes every product look good on the market, and that explains why every product on the market reports it. The randomised trial got closest to honesty by measuring both note time, which improved by 41 seconds, and after-hours record time, which did not move at all.
What metrics should I ask an AI scribe vendor for?
Ask for elapsed time from consultation end to last signature, on consecutive encounters rather than a selected sample, with the method attached and the same clinicians as their own controls before onboarding. Then ask what they measured that did not improve, because a vendor who has never published an unflattering number has not been measuring, they have been marketing. If they can only give you note draft latency, they have told you what their product reaches.
Get the number for your own clinic.
A 90-day pilot that reports consultation-to-closure alongside revenue, hours saved and patients cared for.
About this article. Written and published by WA\. The method is stated in the margin so it can be challenged. We have not published a figure we cannot yet stand behind, in an article arguing that vendors should stop doing exactly that. The randomised trial cited is independent of us and its results are not ours. WA\ has no peer-reviewed clinical trials published to date and does not claim any.