Key points
- Of 180 health system executives Deloitte surveyed in six countries in 2025, about 30% run generative AI at scale in select areas and 2% across the whole enterprise.1
- In a randomised trial of 66 practitioners, ambient AI cut time on notes by 0.36 hours a day without compromising note quality or billing compliance.6
- In a randomised trial of 238 physicians, one ambient scribe cut time in notes by 9.5% and a second showed no significant change.5 Evaluate on your own wards.
- Of 67 countries reporting to WHO, 56% had no functional digital health governance.4 WHO and OECD principles give the template.23
- Deploy in order, from documentation and letters through patient communication, scheduling, coding and evidence answers, each drafting for a person to approve.
Summary
Deploy ambient clinical documentation first, because it has randomised trial evidence and touches every clinician, then letters, patient communication, scheduling, coding and cited evidence answers. Each should draft for a person to sign or approve, be evaluated on your own wards, and sit under a named owner and the WHO principles for AI in health, with every output logged.
Clinicians close their notes before they leave the ward, and every AI draft is approved by the person accountable for it. Write the governance model before the first pilot, and each later deployment goes faster than the one before.
See it in your clinicThirty minutes on your own patients and doctors.
Request a demoSee an ambient note and a patient reply drafted and approved on one record, with the audit log a governance board would review.
| Order | Use | What AI drafts | Who signs or approves | Measure before scaling | Governance gate |
|---|---|---|---|---|---|
| 1 | Ambient documentation | The visit note | Treating clinician | Time in notes, use rate, corrections | Clinical owner named, consent approach agreed |
| 2 | Letters and result explanations | Referral letters, clinic letters, result explanations | Treating clinician | Letters sent the same day | Template approved by the specialty lead |
| 3 | Patient communication | Replies to messages across channels | Patient services staff, clinicians for clinical questions | Time to reply, escalation rate | Approval map signed off, clinical escalation tested |
| 4 | Scheduling and reminders | Bookings, changes, reminders, waitlist offers | Booking team | No-show rate, slot use | Rules reviewed for equity of access |
| 5 | Coding | Suggested codes from the signed note | Clinical coder | Coder agreement rate, denials | Audit sample each month |
| 6 | Evidence answers | Cited answers to clinical questions | Clinician decides | Share of answers with checkable sources | Intended use registered, no autonomous decisions |
Where hospitals stand in 2026
In a Deloitte survey of 180 health system C-suite executives in six countries, fielded in August and September 2025, about 30% reported operating generative AI at scale in select areas and only 2% had deployed AI across the whole enterprise.1 More than 90% said improving productivity would be a priority for 2026.1
WHO reports that 129 Member States now have national digital health strategies, up from 85 in 2015.4 Yet of 67 countries reporting to the Global Digital Health Monitor, 56% had no functional digital health governance.4 Most hospitals will write their own rules.
What to deploy first, and why in that order
Ambient documentation goes first, because it touches every clinician and its output is a draft the clinician already signs. In a stepped-wedge randomised trial of 66 practitioners in ambulatory clinics across two US states, ambient AI wrote 38% of 71,487 notes, cut time on notes by 0.36 hours a day, and improved diagnostic billing codes without compromising note quality.6
Letters and result explanations follow, drafted from the same encounter. Then patient replies drafted in the hospital's voice and approved by staff, then scheduling and reminders, then coding drafted from the signed note. Cited evidence answers for clinicians can run from the start, because every answer carries its sources.
Each step reuses the record the previous one built, so every draft starts from what the clinician already signed. See one clinical system across the organisation.
Evaluate on your own wards
In a randomised trial of 238 outpatient physicians across 14 specialties at one US academic health system, two ambient scribes were compared with usual care. One cut time in notes by 9.5% and the other showed no significant change.5 The scribes were used in 33.5% and 29.5% of visits, and physicians reported clinically significant inaccuracies occasionally.5
Before scaling, measure time in notes, the share of eligible visits where the tool is used and the rate of corrections before signing. A stepped rollout across units gives a fair comparison at no extra cost. The pilot plan sets out the measures.
Governance that follows WHO and OECD
WHO sets six principles for AI in health. Humans remain in control of health care systems and medical decisions, designers meet safety and accuracy requirements for well-defined uses, information is documented before deployment, responsibility and redress are clear, use is inclusive and equitable, and AI is assessed continuously during actual use.2
The OECD AI Principles, adopted in 2019, updated in May 2024 and now with 47 adherents, add transparency, security, safety and accountability as values every AI actor should hold.3 On a ward these become a named clinical owner for each use, human approval on every output, a full audit log, a register of each tool's intended use and a quarterly review of accuracy and equity.
Data, hosting and integration
wai hosts each clinic's records in its own country, under that country's rules. We set this up with your clinic. Staff access is role-based and least-privilege, enforced in the database, data is encrypted with AES-256 at rest and TLS 1.3 in transit, and every message, draft, approval and signature is written to an event log the hospital can export.8 Every subprocessor is named, with thirty days notice before anything changes.8
wai is built to connect to US record systems over FHIR R4. We set up your clinic's connection with you. Ask any AI supplier for current certifications and open work in writing, using the procurement checklist.
How wai works with hospitals
wai works with hospitals and health systems on Enterprise terms, with volume pricing, a security review, custom integrations and a dedicated success manager.7 Work starts with one clinical pathway with measurable outcomes and a named owner, validated against safety and governance gates before live use, and each further workflow is assessed on its own risks. See the organisations page and custom healthcare AI for hospitals and governments.
Where to start
Name an executive sponsor and a clinical owner, and write the governance model on one page, covering intended use, human approval, audit, evaluation and redress. Choose one outpatient service for ambient documentation and record its baseline time in notes and same-day note closure for two weeks.
Run a stepped rollout across two or three units for six to eight weeks and report adoption, time saved and corrections to the governance board. Then add letters and patient communication on the same record, and review every deployment against the WHO principles each quarter.
Bring one pathway and your governance board's questions, and see the drafts and the event log on that pathway.
Frequently asked questions
Which AI should a hospital deploy first?
Ambient clinical documentation. It touches every clinician, its output is a draft the clinician already signs, and it has randomised trial evidence of time saved.6
Are AI agents for hospitals safe to let act on their own?
The safe design is drafting with human approval. WHO's first principle is that humans remain in control of health care systems and medical decisions.2 wai drafts and a clinician or member of staff approves every output.
How should a hospital evaluate an AI scribe?
On its own wards, against usual care, measuring time in notes, use rate and corrections before signing. Products differ, and one randomised trial found a significant effect for one scribe and none for another.5
What governance framework should we use?
Can wai work with our existing record system?
wai is built to connect to US record systems over FHIR R4. We set up your clinic's connection with you. Custom integrations are part of Enterprise terms.7
How is a hospital deployment priced?
On Enterprise terms with volume pricing, agreed after a demo and scoping of the first pathway.7 Published prices for clinics are on the pricing page.
Sources
- 1Deloitte, 20252026 Global Health Care OutlookRead 14 October 2026Read source
- 2World Health Organization, 2021WHO issues first global report on Artificial Intelligence in health and six guiding principles for its design and useRead 14 October 2026Read source
- 3OECD, 2024OECD AI Principles overviewRead 14 October 2026Read source
- 4World Health Organization, 2024Global strategy on digital health 2020-2025, report by the Director-General (EB156/35)Read 14 October 2026Read source
- 5NEJM AI, Lukac and colleagues, 2025Ambient AI Scribes in Clinical Practice, A Randomized TrialRead 14 October 2026Read source
- 6NEJM AI, Afshar and colleagues, 2025A Pragmatic Randomized Controlled Trial of Ambient Artificial Intelligence to Improve Health Practitioner Well-BeingRead 14 October 2026Read source
- 7wai, 2026wai for clinicians, plans and EnterpriseRead 14 October 2026Read source
- 8wai, 2026Security at waiRead 14 October 2026Read source
Notes closed on the ward, and every AI output signed by a clinician.
Fifteen minutes with a clinician who uses wai every day, on the pathway you would start with. We reply within one working day.
Partner with us
See it on your own patients, then build it with us
Thirty minutes on your own patients, doctors and rules, in your market. We reply within one working day, and the clinics that partner with us early shape what we build next.
See what wai does between the visits