Quirónsalud
Quirónsalud's in-house AI scribe: emergency consultations 21.8% shorter on average across 48 Spanish hospitals
Spanish hospital group Quirónsalud built its own AI scribe, which listens to consultations and drafts the medical report inside the patient record. Over 12 months in the emergency departments of 48 hospitals it was used in 45.3% of consultations, and scribe-assisted consultations were shorter by a mean 21.8%.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
21.8%
Mean relative time saving in scribe-assisted emergency consultations (12-month study, 48 hospitals)1
more than 2.33 million
Assisted outpatient encounters, September 2024 to December 20252
What it took
Cost, team size and timeline were not disclosed in the sources.
The problem
Clinical documentation is a persistent burden for doctors. Quirónsalud, a large Spanish hospital network, built and evaluated its own AI scribe to ease it, in emergency departments and in outpatient care.
What they built
The group developed its own ambient AI scribe. It records the consultation, turns speech into text with speech recognition, and uses a large language model (an AI system that writes text) to fill a structured report from predefined templates. It works with several languages and sits inside the electronic health record, where the doctor reviews and edits the draft before validating it.
How they did it
Outpatient data in the study begins in September 2024, and use was optional. The emergency study covered February 2025 to January 2026, five emergency specialties and all level 4 and 5 emergency consultations in 48 hospitals. The team tracked adoption monthly, consultation duration, transcription accuracy, patient Net Promoter Score (a likelihood-to-recommend survey), clinician surveys and structured audits of report quality.
Results
The scribe was used in 1,032,558 emergency consultations, 45.3% of the total, with monthly adoption rising from 7.7% to 57.8%. Assisted consultations were shorter by a mean relative 21.8%. Audits scored scribe reports higher on documentation quality, and patient experience was higher in assisted consultations. In outpatient care the tool reached more than 2.33 million encounters.
What went wrong
Outpatient results were more modest than the emergency ones: consultation length was almost unchanged at 15.01 vs. 14.65 minutes, and efficiency gains converged toward parity over time. The authors suggest the saved time may be reinvested in patient interaction, but that remains a hypothesis.
All reported figures
| Measure | Figure | Citation |
|---|---|---|
| Mean relative time saving in scribe-assisted emergency consultations (12-month study, 48 hospitals) | 21.8%1 | |
| Emergency consultations in which the scribe was used | 1,032,558 (45.3%)1 | |
| Monthly scribe adoption in emergency departments, start to end of study | 7.7% to 57.8%1 | |
| Assisted outpatient encounters, September 2024 to December 2025 | more than 2.33 million2 | |
| Average outpatient consultation length, scribe vs non-scribe | 15.01 vs. 14.65 min2 | |
| Semantic agreement between AI drafts and clinician-validated outpatient notes | 87.4-89.2%2 |
Lessons for your project
Let doctors opt in and measure the pull. In outpatient care use was not mandated, and adoption rose from 2.7% to about 31% of visits.
Judge value by more than time. Outpatient consultation length barely changed, while clinicians reported better experience in five of seven areas.
Audit the output, not only the speed. In emergency care the group scored report quality through structured audits and tracked patient satisfaction with Net Promoter Score.
Expect results to mature over months. Gains in outpatient care converged toward parity as workflows settled, and the system's speech and language components were updated during the study.
Integrate with the patient record. Clinicians review and edit the draft inside the electronic health record before validating it.
Caveats
- Both papers come from Quirónsalud authors (the outpatient paper discloses that all are employees and two hold management roles), so the figures are the company's own. The journals are peer-reviewed; no outside party measured the results.
- The emergency study is retrospective and observational, so it cannot prove the scribe caused the shorter consultations. The emergency figures here come only from the published abstract; the full text is behind a paywall and was not read.
- In the outpatient paper, consultation length differed little (15.01 vs. 14.65 minutes), and the authors call the duration comparison descriptive, based on aggregated data.
- The outpatient paper says its semantic agreement score compares AI drafts with the doctor's final text, not with an independent truth. The emergency abstract also reports a mean transcription accuracy of 93.9% without saying how accuracy was measured, so it is not used as a result here.
- The model behind the scribe is not named. It is described as an in-house pipeline of speech recognition and a large language model. Deployed date is when the outpatient data begins; the emergency study ran February 2025 to January 2026.
- Quirónsalud is a private hospital group, so results may not transfer to public systems. The abstract gives no figure for patient Net Promoter Score or audit scores, only that they were higher.
Sources
- Deployment of an ambient AI scribe in emergency care: A 12-month evaluation in a large Spanish hospital network
International Journal of Medical Informatics · 2026-07-02 · Research paper · doi.org
- Real-world evaluation of an ambient AI scribe in Spanish outpatient care after 2.3 million uses: impact on clinician experience, semantic agreement, and workflow efficiency
Frontiers in Digital Health · 2026-07-06 · Research paper · frontiersin.org
Summarized in our own words from the sources above. Spotted an error? See corrections.