Healthcare systems are turning to artificial intelligence to help manage growing demand, reduce administrative work and support overstretched clinicians. But a new industry white paper warns that the technology may struggle to make a meaningful difference unless hospitals, regulators and insurers change the way healthcare is organised and funded.
The paper, produced following a roundtable hosted by AstraZeneca, NVIDIA and Nebius during BioTechX Europe 2025, says the global healthcare workforce shortage is expected to exceed 11 million practitioners by 2030.
An ageing population and the rising prevalence of noncommunicable diseases, including cancer, cardiovascular disease, respiratory conditions and diabetes, are expected to increase pressure on already stretched services.
The report argues that AI could help healthcare workers manage more patients without simply increasing the size of the workforce. Potential applications include automated consultation notes, patient monitoring, clinical decision support, risk-based triage and remote follow-up.
“Global healthcare workforce shortage is expected to exceed 11 million practitioners by 2030”
However, its central message is that technology alone will not solve the problem. The greatest obstacles may be found in hospital procurement departments, reimbursement rules, fragmented information systems and uncertainty over regulation.
From administrative support to clinical capacity
The white paper describes a range of AI applications intended to reduce pressure on clinicians.
Generative AI tools can summarise consultations, prepare medical notes and create patient overviews. Medical scribes can reduce the amount of time doctors and nurses spend documenting appointments. Analytical systems can help clinicians identify relevant research and apply clinical guidelines to individual patients.
Other tools are designed to monitor patients outside hospitals. Voice agents, for example, can conduct follow-up assessments and escalate patients who appear to be at higher risk. The report also refers to systems that support at-home anaesthesia evaluations and help reduce surgery waiting times.
One example discussed by participants involves a system that allows nurses to monitor as many as ten times more patients by reducing the need for manual file reviews. Another involves an AI pipeline that tracks tumour volumes over time, potentially allowing clinicians to assess treatment response more quickly than conventional methods.
These examples point to a shift in how AI is being presented in healthcare. The argument is no longer simply that machines can perform isolated tasks. Instead, AI is being positioned as a way to reorganise clinical work, allowing professionals to focus on complex decisions, prevention and direct patient care.
Handling more patients does not necessarily mean delivering better care. Increased productivity could bring benefits if it reduces paperwork and allows earlier intervention. It could also create new risks if staff are expected to supervise larger patient groups, verify automated outputs or respond to a higher volume of alerts.
Hospitals are not ready to adopt technology at the same speed
The report identifies a gap between the speed of innovation and the ability of healthcare organisations to adopt new tools.
Hospitals often operate with fragmented systems, legacy software and electronic health records that do not communicate easily with one another. Clinicians may be required to enter the same information into several systems, undermining the efficiency gains promised by automation.
The paper says contracting and procurement can take between eight and 12 months. By the time a hospital has assessed a product, completed security checks, negotiated a contract and secured approval, the technology may already have changed. Smaller companies can struggle to survive through such lengthy sales cycles.
This creates a particular problem for healthcare AI. Unlike a conventional software purchase, an AI system may require continuing evaluation, model updates, staff training and monitoring for changes in performance. Hospitals must also assess how the system handles sensitive health data and whether it produces different results for different groups of patients.
The report calls for closer cooperation between hospitals and technology companies, as well as simpler integration with existing electronic health record systems. It also proposes the use of machine-learning operations support to help healthcare organisations deploy and monitor AI more reliably.
Payment systems may not reward efficiency
One of the paper’s most significant arguments concerns reimbursement.
Healthcare payment systems often reward the volume of services delivered or the tasks performed by professionals. They may not recognise the value created when technology allows a clinician to monitor more patients, prevents an admission or reduces the time needed for documentation.
This can leave hospitals facing the cost of implementation while the financial benefits are distributed elsewhere. An investment in primary-care triage, for example, may reduce pressure on emergency departments, but the department funding the system may not receive the resulting savings.
The same problem can apply to staff productivity. If an AI tool enables a nurse to monitor more patients, the improvement may not translate into additional income under existing payment arrangements. Hospitals may therefore have little financial incentive to invest, even if patients receive faster or more coordinated care.
The white paper argues that insurers and policymakers should modernise reimbursement frameworks to recognise better outcomes and new models of care. It recommends using health economics and outcomes research to measure whether AI improves clinical results, reduces costs or improves patients’ quality of life.
That approach would move healthcare away from paying mainly for activity and towards rewarding prevention, efficiency and outcomes. It would also give organisations a clearer basis for deciding whether an AI system is worth adopting.
Clinicians need training and reassurance
The report also highlights resistance among clinicians and nurses, many of whom fear that AI could replace their jobs.
Its proposed response is a human-centred approach in which AI is used to support professional judgement rather than remove it. The paper recommends role-specific training that explains what systems can and cannot do, as well as greater involvement of clinicians in the design and testing of tools.
This is important because poorly integrated AI can increase rather than reduce workload. A system that produces unreliable alerts, requires duplicate data entry or offers recommendations that cannot be explained may quickly lose the confidence of staff.
Early pilots with measurable benefits could help build support. The paper cites an asthma-monitoring application that identified likely exacerbations in four out of ten patients, allowing earlier medication adjustments. But the document provides limited detail about the study and does not explain whether the result has been independently replicated.
For healthcare workers, the practical question will be whether AI gives time back to patient care or simply adds another layer of monitoring and checking.
Regulation and accountability remain unresolved
The white paper calls for clearer and more adaptable regulation. Participants argued that uncertainty can make hospitals reluctant to test new tools, while overly rigid rules may slow useful innovation.
A more flexible approach could help, but it would still need firm safeguards. These include requirements for data protection, cybersecurity, clinical oversight, performance monitoring and reporting of adverse events.
The report does not fully address who should be responsible when an AI-supported decision causes harm. It also says little about how patients should be told that AI is being used, how they can challenge an automated assessment or how systems should be tested for bias.
These issues will become more important as AI moves from administrative work into triage, treatment support and remote monitoring.
Technology cannot substitute for workforce planning
The white paper makes a persuasive case that AI could relieve some pressure on healthcare staff. It may reduce repetitive documentation, help clinicians prioritise patients and allow more consistent use of medical evidence.
But its proposals are not a substitute for training more healthcare workers, improving retention or making working conditions safer. AI may also create new responsibilities in data governance, system supervision and clinical verification.
The report’s most realistic conclusion is that AI can ease the impact of workforce shortages, rather than eliminate the shortages themselves. Whether it does so will depend on decisions made outside the technology sector.
Hospitals will need interoperable systems and faster procurement. Clinicians will need training and meaningful involvement in implementation. Regulators will need to provide clear safety requirements. Insurers and governments will need to design payment models that reward improved outcomes rather than simply counting services.
Until those changes are made, healthcare AI may remain caught between strong promises and slow adoption. The technology may be developing rapidly, but the institutions expected to use it are still operating according to older rules.
To note: PP Health Malaysia (PPHM) team has evaluated this industry white paper. The paper does not provide detailed information about the studies behind these examples, including sample sizes, error rates, patient outcomes or independent validation. The white paper is based on a round table discussion by AstraZeneca, NVIDIA and Nebius.
The post Healthcare AI Promises to Ease Staff Shortages, but Adoption Slows first appeared on PP Health Malaysia.
