Ask a hospital chief executive what keeps them awake, and the answer is rarely a new drug or a piece of equipment. It is people. Wards cannot open because there are not enough nurses. Waiting lists grow because consultants are stretched thin. Primary care practices close their books to new patients because they cannot recruit doctors. In almost every developed health system, labour has become the binding constraint on how much care can be delivered.
At the same time, the clinicians who are in post spend a startling share of their working day on something other than patients. Notes, orders, coding, inbox messages and insurance forms consume hours that were never part of the job description. That combination, too few people doing too much administration, has made healthcare one of the most promising and most scrutinised arenas for artificial intelligence.
The Size of the Gap
The World Health Organization projects a global shortfall of around 11 million health workers by 2030, concentrated most heavily in low and middle income countries. Its State of the World’s Nursing report in 2025 estimated a nursing gap of several million alone. These are not abstract numbers. They translate into longer waits, closed services and overworked staff.
Wealthy countries are not immune. The Association of American Medical Colleges projects a US shortage of up to 86,000 physicians by 2036, driven by an ageing population that needs more care and an ageing workforce heading towards retirement. In the UK, the NHS has carried tens of thousands of vacancies for years and relies heavily on international recruitment, much of it from countries such as India, Nigeria and the Philippines that face shortages of their own.
India illustrates the global tangle well. It trains large numbers of doctors and nurses, many of whom go on to work in the Gulf, the UK, Australia and North America. Rural India, meanwhile, struggles to staff basic health centres. Shortages in one country are often filled by creating shortages in another, which is why automation that stretches each clinician’s capacity matters everywhere.
The Hidden Cost of Burnout
Burnout is often discussed as a wellbeing issue, and it is one. It is also a financial problem that rarely appears clearly on a balance sheet. Research going back to a widely cited 2016 study in the Annals of Internal Medicine found that physicians in ambulatory care spent nearly two hours on electronic records and desk work for every hour of direct patient time. Much of that work spills into evenings and weekends, a pattern clinicians call pyjama time.
When exhausted clinicians cut their hours or leave, the costs mount quickly. Replacing a single physician, once recruitment, onboarding and lost revenue are counted, is commonly estimated to cost hundreds of thousands of dollars. Nurse turnover costs US hospitals tens of thousands of dollars per departure. Health systems fill the gaps with agency and locum staff at premium rates, and the NHS has spent billions of pounds a year on temporary staffing despite repeated efforts to cap it.
There is a quality cost as well. Tired clinicians make more errors, have less time to listen and are more likely to order defensive tests. For hospital boards, the link between administrative burden, burnout, turnover and patient outcomes is now well enough established that reducing documentation load has become a strategic priority rather than an IT project.
The Rise of the AI Scribe
Ambient AI scribes are the technology that has captured this moment. The idea is simple. With the patient’s consent, a device listens to the consultation, and AI produces a draft clinical note, often with suggested codes and follow-up orders, for the clinician to review and sign. The doctor can look at the patient rather than the screen.
Adoption has been remarkably fast for a sector known for caution. Kaiser Permanente rolled out Abridge’s ambient tool across its hospitals and hundreds of medical offices in 2024. Microsoft has folded its Nuance DAX technology into Dragon Copilot, and Epic, the dominant US electronic record vendor, has been building ambient features directly into its software. In England, NHS guidance on ambient voice technology and the government’s ten year health plan have both pointed to these tools as a way to free up clinical time.
What the Evidence Actually Shows
The evidence base is young but growing. A quality improvement study published in JAMA Network Open in October 2025 followed 263 physicians and advanced practice clinicians across six US health systems. Among ambulatory clinicians, reported burnout fell from 51.9% to 38.8% after 30 days of using an ambient scribe. The study had no control group, so it cannot prove cause and effect, but the direction and size of the change caught attention.
More rigorous trials are now arriving. A randomised trial published in NEJM AI compared two commercial scribes with usual care and found improvements in measures of work exhaustion and task load for users of either product. A review in npj Digital Medicine in 2026 found a consistent pattern across studies: improvements in wellbeing and cognitive load were more reliable than savings in raw documentation time.
That last point matters for anyone building a business case. The time saved per note is often a matter of minutes, and it varies widely by specialty and by how much each clinician edits the draft. Health systems hoping that scribes will instantly let every doctor see several extra patients a day may be disappointed. The stronger argument is retention: a tool that makes the job more bearable may keep an experienced clinician in post for years longer, which is worth far more than a few extra appointments.
| Task | Who does it today | Automation now available | Maturity in 2026 |
|---|---|---|---|
| Consultation notes | Doctors, nurse practitioners | Ambient AI scribes | Widely deployed, evidence growing |
| Patient inbox and messages | Doctors, nurses | AI drafted replies for review | Early, mixed results |
| Coding and billing | Clinicians, coding teams | AI suggested codes from notes | Maturing quickly |
| Prior authorisation and forms | Clinicians, admin staff | Automated form filling and submission | Early |
| Discharge summaries and referrals | Junior doctors | AI drafted summaries | Piloting in many systems |
| Rostering and bed management | Managers, nurses | Predictive scheduling and flow tools | Established |
| Triage and symptom checking | Nurses, call handlers | AI assisted triage | Used with strict oversight |
Beyond the Scribe
Documentation is only one slice of the administrative load. The bigger opportunity lies in automating the work around the consultation. AI tools now draft replies to patient messages, pre-fill insurance prior authorisation requests, summarise long records before a clinic, and produce discharge letters that junior doctors used to write late into the night.
Operational AI is also changing how scarce staff are deployed. Predictive models that forecast admissions can help managers roster nurses where they will be needed, rather than reacting to a crisis on the day. Better patient flow tools free up beds and reduce the time clinicians spend chasing them. None of this replaces a nurse at the bedside, but all of it gives that nurse more time to be there.
These changes connect directly to the value-based care models we discussed earlier in this series. When providers are paid for outcomes rather than activity, freeing clinical time for prevention, follow-up and complex patients has a clear financial return.
The Limits and the Risks
None of this comes without risk. AI-generated notes can contain errors, including details that were never said in the room. Clinicians remain legally responsible for what they sign, which means a careless review can turn a time saver into a liability. Health systems need clear policies on review, audit and correction, and they need to monitor error rates rather than assuming the technology is accurate.
Consent and privacy matter too. Recording consultations raises understandable concerns for patients, particularly in sensitive specialties such as mental health or sexual health. The data flows involved, often through cloud services operated by third parties, add to the cybersecurity exposure we highlighted in our article on attacks against hospitals.
There is also a subtler risk. If time saved is immediately filled with more appointments, the burnout benefit can evaporate. Several health systems have found that clinicians welcome the tools precisely because they give back a little breathing room. Turning every minute saved into extra throughput may look efficient on a spreadsheet and still drive people out of the profession.
A Playbook for Health System Leaders
Start with the clinicians who are struggling most. Primary care, emergency medicine and specialties with heavy documentation usually show the biggest benefits. Let clinicians choose whether to use the tools at first, measure what happens, and expand on evidence.
Measure the right things. Time per note is useful, but retention, sick leave, vacancy rates, agency spend and patient experience tell a more complete story. Agree in advance how the time saved will be used, and be honest with staff about it.
Buy with integration in mind. A scribe that does not write cleanly into the electronic record, or that forces clinicians to copy and paste between systems, will be abandoned within weeks. Ask vendors for evidence from organisations that use the same record system as yours, check how pricing scales from a pilot of 50 clinicians to a rollout of 5,000, and make sure the contract allows you to switch suppliers as the market matures. With the large record vendors building their own ambient tools, today’s specialist start-up may face very different economics in two years.
Pay attention to who the tools work for. Speech recognition has historically performed less well with strong regional accents, non-native speakers and noisy environments such as emergency departments. In a workforce as international as the NHS, and with patients from every background, accuracy needs to be tested across the people who will actually use it, not only in a quiet pilot clinic.
Treat governance as part of the deployment. That means clear accountability for note accuracy, regular audits, patient consent processes and contracts that protect data. Finally, think of AI as one part of a wider workforce strategy that still includes training more clinicians, improving pay and conditions, and redesigning roles so that each professional works at the top of their licence.
Technology Will Not Train a Single Nurse
It is worth being clear about what automation cannot do. It cannot train a single new nurse or doctor, and it cannot provide the human presence that sits at the heart of care. Countries that treat AI as a substitute for workforce planning will be disappointed.
What it can do is give back something healthcare has been losing for two decades: clinician time and attention. Used well, the current generation of tools could make clinical work more sustainable and help health systems get more care from the people they already have. In a sector where labour is the scarcest resource of all, that may be the most valuable return on any technology investment this decade.
