More time with the patient.
Less time with the keyboard.
The NHS's national AVT evaluation in outpatient care — Kingston Hospital and UCLH as core sites. Direct care time rose 23.5% and appointments shortened by 8% on average, with patient satisfaction held and clinician cognitive load down.
What the trial measured.
Five core sites with full TimeCat time-motion observation. 104 clinicians, 2,095 observations. Pre-post within-subject design — each clinician compared against themselves.
Core site with full TimeCat time-motion observation. 19 clinicians per-protocol; 269 baseline observations vs 214 AVT observations. Patient surveys: 140 baseline, 91 AVT.
Largest core site in the Phase 4 trial by observation count. 38 clinicians per-protocol; 381 baseline observations vs 381 AVT observations across multiple outpatient specialties.
GLMM-predicted direct care time rose from 69.3% to 84.1% — a +14.8 percentage-point absolute increase (95% CI +13.6 to +15.7 pp, p<0.001). The shift is from the keyboard to the patient.
Patient surveys at Kingston and UCLH showed no significant reduction in satisfaction under AVT. Net Promoter Score for AVT across the trial was +38 (58% promoters, 19% detractors, n=184).
The arithmetic of one extra patient.
Median appointment time at the outpatient core sites fell from 18.4 minutes to 16.9 minutes — about 1.5 minutes per patient. Across a typical four-hour outpatient clinic of 12–14 patients, that compounds into ~18–21 minutes recovered per session: enough to add one further patient per clinic where demand allows, without lengthening the session.
~1.5 minutes median (mean reduction 5.86%, p<0.005). Verified by independent observers using the TimeCat app, second-by-second.
~18–21 minutes across a typical 12–14-patient outpatient clinic. Modelled from the median per-appointment saving; site-specific results vary by specialty.
Modelled implication where demand and template allow. Sites have used recovered time for additional capacity, earlier clinic finishes, or clinician wellbeing.
Note: the "+1 patient per clinic" figure is a modelled implication of the per-appointment saving observed at Kingston and UCLH. The 47-minute / one-extra-patient headline from the Phase 4 Executive Summary refers to the St George's Emergency Department site, where documentation time fell from 12 to 6 minutes per encounter.
I feel it takes some stress away from the additional cognitive strain of capturing everything I said. It completes notes with a high level of accuracy and safety netting — my focus is on the patient.
This trial is significant. The NHS can lead the way in safely adopting AI — by working collaboratively from hospitals to ambulances we've proven it can work at scale and make a real difference for patients and clinicians.
How the trial was run.
Non-randomised, multi-centre, within-subject, pre-post intervention trial. 12 months. 9 NHS sites. Single AVT supplier, version-locked for the duration.
- 01Core outpatient sitesMay 2024 – Apr 2025
Kingston Hospital and UCLH ran the full core-site protocol: TimeCat time-motion observation by trained observers, plus pre and post clinician and patient surveys.
- 02TimeCat observation7 categories
Second-by-second time allocation into Direct Care, Computer Notes, Computer Orders, Computer Read, Written Notes, No Care/Absent, Other Indirect. Each clinician needed ≥9 baseline + ≥9 AVT observations.
- 03Statistical analysisPer-protocol
Generalised Linear Mixed Models (GLMM, beta distribution) for direct care %; paired t-tests and Wilcoxon signed-rank tests for total appointment duration. p<0.05 threshold. Powered to detect 10% change in direct care at 90% power.
- 04GovernanceClass I device
MHRA-registered Class I medical device. CE+ certified. Bias assessed. Funded by NHS England (London) Frontline Digitisation; NIHR GOSH BRC infrastructure support; Tortus AI provided free of charge.
What the authors flag.
Service evaluation, not a randomised controlled trial. Patient survey response rates varied by site (11–71%). The companion preprint (Wray et al.) is published via SSRN / Preprints with The Lancet and is not yet peer-reviewed. GOSH has a declared collaborative partnership with Tortus AI. The capacity arithmetic above is modelled from the per-appointment saving; specialty-level results vary.
Run the same study on your clinic.
Bring your outpatient pathway, your template, your specialty. We bring the evaluation method that already passed national scrutiny.