PSIRF, Human Factors and the Resource to Turn Patient Safety Learning into Action

A governance challenge for NHS and independent healthcare providers

For healthcare leaders, the central argument is straightforward: the value of a learning response lies not in the volume of analysis it produces, but in whether it enables timely, accountable and measurable improvement. Boards and executive teams therefore need visibility of the trade-off between further investigation and implementation, the capability required at each stage, and the effect on patient outcomes, workforce capacity and organisational risk.

A patient safety incident occurs. A learning response is commissioned. Staff and patients may be interviewed, records reviewed, timelines reconstructed and contributory factors identified. The final report makes recommendations, which enter an action plan and begin their journey through committees, assurance meetings and performance reporting.

At what point would the next hour spent investigating create less safety value than an hour spent implementing what we already know?

This is not an argument against investigation, nor a case for closing legitimate questions prematurely. It is a leadership question for boards, executives, clinical leaders and those overseeing patient safety: how should finite organisational capacity be allocated when further inquiry competes directly with implementation, evaluation and attention to other unresolved risks? More specifically, it is a question about the fungibility of organisational capacity

The Patient Safety Incident Response Framework (PSIRF) was designed to support a more considered and proportionate response to patient safety incidents. It applies across NHS providers and independent healthcare organisations delivering NHS-funded care. NHS England is explicit that PSIRF is not simply an investigation framework. It supports compassionate engagement, systems-based learning, proportionate responses and supportive oversight. Its guidance connects learning to improvement, safety action development and safety improvement plans [1, 2].

These are matters of patient safety, clinical governance and corporate governance: what the organisation learns, who is accountable for acting, and how boards gain assurance that care is becoming safer.

Yet even where organisations have adopted PSIRF terminology and learning-response tools, they can remain trapped in an older mental model: every serious outcome must generate a lengthy investigation, every investigation must generate new recommendations, and completion of the report is treated as evidence that the organisation has responded.

PSIRF asks us to think differently. It asks not simply, “How serious was the outcome?” but, “What is the potential for new learning, and what response is likely to improve patient safety?” [1, 2]

What is a fungible resource?

In economics, an asset is fungible when one unit can be substituted for another unit of the same kind. Money is the familiar example. One £10 note may be exchanged for another without altering its economic value.

Richard H. Thaler’s work on mental accounting challenged the assumption that people consistently behave as though money is completely fungible. People divide resources into psychological accounts and apply different rules depending on how money was obtained, labelled or intended to be used. Consequently, resources which are economically interchangeable may not be treated as interchangeable in practice [3].

Thaler’s argument offers a useful organisational analogy. Healthcare organisations also create implicit accounts, even when those accounts do not appear in the ledger:

·         investigation capacity

·         clinical time

·         governance time

·         quality-improvement capability

·         data and analytical support

·         patient-safety expertise

·         executive attention

·         project-management capacity

·         workforce goodwill

·         the attention of patients, families and staff affected by an incident

An organisation may behave as though these resources belong to separate, non-transferable accounts. An investigator’s time is seen as “investigation resource”. A clinician’s attendance at a learning-response meeting is recorded as participation rather than as time unavailable for implementation. Committee time is treated as governance rather than as a scarce safety resource.

In reality, these resources are at least partly fungible. An experienced patient-safety practitioner who spends another week refining a report cannot spend that same week facilitating implementation, evaluating an existing safety intervention or examining a different area of emerging risk.

Fully fungible, partly fungible and non-fungible resources

The distinction matters because not every resource can be exchanged on equal terms.

Fully or highly fungible resources

Financial resources may be relatively fungible where budgets permit their reallocation. Generic administrative or project-support capacity may also be transferable between activities, subject to contractual and practical constraints.

Partly fungible resources

Most specialist healthcare capacity is only partly fungible. A patient-safety specialist may be able to investigate, facilitate a multidisciplinary review, support action design or evaluate an intervention, but these activities are not identical. Moving the individual between them has consequences for quality, continuity and opportunity cost.

Clinical time is also partly fungible. A consultant, nurse, pharmacist or allied health professional may contribute to an investigation, improvement programme, audit or training activity. However, their knowledge of a particular patient, pathway or service may make them difficult to substitute.

Non-fungible resources

Some resources cannot meaningfully be replaced. The lived experience of a patient or family is not interchangeable with a written record. Neither is the contextual knowledge held by a member of staff who was present when an incident occurred. Trust, psychological safety and organisational memory are also non-fungible. Once lost, they cannot simply be purchased back or replaced by another committee.

Therefore, proportionate incident response is not about doing less. It is about recognising the characteristics and value of different resources and deploying them where they can produce the greatest improvement in safety.

The hidden cost of further investigation

Investigations have both direct costs and opportunity costs.

The direct costs include investigator time, clinical participation, data collection, administrative support, engagement with those affected, report drafting and oversight. The opportunity cost is the safety work that those people cannot undertake while the response continues.

This creates a difficult possibility: a learning response may be technically thorough but strategically inefficient.

A further round of interviews may add nuance to the narrative without changing the emerging system analysis. Another report revision may improve readability without altering any potential safety action. An additional committee may provide reassurance without materially strengthening control of the risk.

The relevant leadership question is not whether further learning is possible. In complex systems, further learning is almost always possible. The question is whether the marginal value of additional learning exceeds the value that could be achieved by reallocating the same capacity to implementation, evaluation or another unresolved risk.

That is the meaning of proportionality in practice. It is not proportionality as shorthand for a cheaper process. It is a reasoned judgement about the depth, method and duration of a response, taking account of learning potential, existing knowledge and the likelihood that additional inquiry will influence improvement. NHS England’s guidance deliberately supports multiple system-based learning methods rather than positioning a full patient safety incident investigation as the automatic response to every event [1, 2].

Knowing more is not the same as becoming safer

Healthcare organisations can accumulate learning faster than they can convert it into improvement.

The result is familiar: multiple incidents identify similar contributory factors; recommendations are restated in slightly different language; and action trackers expand while the underlying conditions of work remain largely unchanged.

This creates what might be called a learning-to-action deficit.

Reports and recommendations are outputs. Reduced exposure to risk, stronger controls and more resilient systems are outcomes. Good healthcare governance must distinguish between the two.

The NHS Patient Safety Strategy is built on patient-safety culture and patient-safety systems, with insight, involvement and improvement as its three principal aims. Insight is not the destination. It creates value only when it informs and enables improvement [4].

The 2026 Quality Strategy for NHS-funded care reinforces this wider responsibility. It seeks to make quality the organising principle for NHS-funded care and presents safety, effectiveness and experience as interdependent domains. It also emphasises clearer accountability, value, transparency, data and the consistent application of improvement approaches [5].

If quality is genuinely the organising principle, boards must ask not only whether learning responses have been completed but whether organisational resources are being deployed where they will create the most value for patients.

Why Human Factors matters to PSIRF and implementation

Human Factors in healthcare examines how people interact with tasks, equipment, technology, information, teams, environments and organisational conditions. It moves the conversation beyond individual error and asks how work was designed, how it was actually carried out and what made an action reasonable at the time.

That makes Human Factors central to the Patient Safety Incident Response Framework. A systems-based learning response should identify the conditions which shaped performance. Good action design should then change those conditions rather than rely on reminders, retraining or greater vigilance as the default response.

It also changes how improvement is resourced. The question is not simply which senior professional has authority. It is what knowledge the next stage requires: clinical judgement, Human Factors expertise, governance oversight, project delivery, measurement, education or operational ownership. Treating all of that as “clinical leadership” can obscure the capability the work actually needs.

The minimum effective dose of expertise in patient safety improvement

The same argument applies once the investigation is complete. Organisations may be disciplined about the proportionality of a learning response, then become curiously imprecise about the resource used to put its findings into practice.

Medicine offers a useful discipline: use the minimum effective dose. This is not an instruction to under-resource change. It means matching the intervention to the need, increasing it when risk or evidence requires, and avoiding cost and organisational side effects which add no equivalent benefit.

In improvement work, the expensive dose is often senior clinical time. Doctors, senior nurses and other experienced professionals are placed on project groups, drawn into recurring meetings and retained throughout delivery because their presence signals importance. Sometimes their expertise is indispensable. They may hold non-fungible knowledge about clinical risk, professional practice or whether a proposed control is workable. But many implementation tasks require capable facilitation, project management, measurement, testing or coordination rather than continuous senior clinical input.

The question is not whether clinical leadership matters. Evidence reviewed by the Health Foundation associates clinical leadership and management with stronger organisational performance and operational efficiency [8]. The question is where that leadership is needed, at what intensity and for how long. A consultant may need to define the clinical boundary conditions, approve a test and review emerging risk. That does not automatically mean the consultant must administer the action plan. A senior nurse may need to shape the standard and judge whether it is safe. That does not mean every implementation meeting requires senior nursing attendance.

Proportionate implementation means using the lowest level of resource that can safely and reliably deliver the required change, with a clear route to escalate when the risk or complexity demands more.

This is not substitution by default. It is deliberate role design. The NHS Long Term Workforce Plan describes the need for the right number of people, with the right skills and support [9]. The same test should be applied to improvement portfolios: distinguish sponsorship from delivery, technical authority from coordination, and decision points from routine attendance.

Patient safety improvement as a route to better NHS financial management

This has a direct bearing on NHS finances. Staff time does not become free simply because it is already in the establishment. Using an hour of scarce clinical or executive capacity for work that another role could undertake safely still carries an opportunity cost. It may displace patient care, defer another decision or require cover elsewhere.

The financial opportunity is not simply to replace expensive people with cheaper people. That would be crude and, in safety-critical work, potentially dangerous. The opportunity is to remove avoidable over-specification: more people than the task needs, more seniority than the decision requires, more meetings than delivery can justify, or senior involvement continuing long after its value has reduced.

The Health Foundation identifies skill-mix change, better flow, technology and stronger implementation capability as routes to sustainable productivity, while warning that benefits do not materialise merely because an opportunity has been identified [10]. Its later work argues that productivity should be framed as a route to maintaining or improving quality, not as a finance-only exercise [11]. That distinction matters. Poorly designed cost reduction transfers work, weakens controls and creates failure demand. Well-designed resource allocation removes work which adds little safety value and protects the expertise needed at the points where it does.

NHS England’s 2026 productivity plan requires annual productivity improvement of 2% over the Spending Review period and links productivity to better use of operational, clinical and technological resources [12]. A disciplined approach to improvement resource can contribute, but only if the benefit is made visible. Boards should ask what senior time was released, where it went, whether implementation moved more quickly and whether the expected safety outcome was achieved.

The unit of value is therefore not a cheaper meeting. It is a safer outcome delivered with no more resource than was necessary, while preserving capacity for the next priority.

AI in healthcare can widen the learning-to-action deficit

AI sharpens the problem because it lowers the cost of producing information far more quickly than it increases the capacity to make accountable decisions.

The attraction is obvious. NHS England’s large Microsoft 365 Copilot trial reported average time savings of 43 minutes per user per day, and the national rollout announced in June 2026 extended access to more than 500,000 staff [13]. Drafting, summarising, searching and analysing may all become quicker. Those are credible potential gains.

But faster production is not the same as faster improvement. If AI helps ten people create papers, summaries, options and action lists more quickly, it does not follow that the smaller number of people authorised to judge risk, resolve competing priorities, commit money and approve change have gained any capacity. The queue has moved; it has not necessarily shortened.

The doer saves twenty minutes. The reviewer receives three more documents.

This is the less comfortable side of the learning-to-action deficit. When the marginal cost of producing another plausible analysis falls, organisations are likely to produce more analysis. The assurance burden may then rise because AI-assisted output still needs verification, contextual judgement and clear ownership. The cross-government Copilot evaluation found useful time savings but also limitations in complex, nuanced and data-heavy work [14]. Reviews of healthcare AI implementation identify familiar barriers: workflow integration, leadership, staff engagement, change management, finance, workforce capacity, evaluation and maintenance [15, 16]. AI does not remove those constraints; it can expose them more quickly.

There is also a risk of false abundance. A polished summary can make uncertain evidence look settled. A longer list of actions can look like stronger control. More frequent reporting can create the appearance of grip while the same unresolved decisions circulate through the system. WHO’s guidance is clear that health AI requires human autonomy, transparency, accountability and sustainable governance [17]. Human oversight is not a ceremonial final check; it is work, and it needs capacity.

The right productivity question is not simply, “How much time did the tool save at the task?” It is, “What happened to the time released, what new demand appeared downstream, and did the route from insight to safer care become shorter?” Without that end-to-end measure, a local efficiency can become a system bottleneck.

Exploration and exploitation

James G. March’s influential work on organisational learning provides another helpful lens. March distinguishes between:

1. exploration, including search, experimentation, variation, discovery and risk-taking; and

2. exploitation, including implementation, refinement, efficiency, execution and the use of established knowledge.

Both are essential, but they compete for scarce resources. An organisation that only explores generates ideas and learning without developing sufficient practical competence. An organisation that only exploits existing knowledge risks becoming efficient at yesterday’s solution and trapped in a suboptimal position [6].

Patient-safety learning responses are predominantly exploratory. They help organisations understand how work is performed, how system conditions influence behaviour and where opportunities for improvement may exist.

Safety-action design, implementation and evaluation are predominantly exploitative. They turn existing knowledge into changed processes, environments, technologies or controls.

The danger is not that healthcare organisations fail to explore. It is that investigation can become institutionally easier than implementation. Investigation has recognisable processes, templates, deadlines and reports. Implementation crosses boundaries, competes with operational priorities, requires sustained leadership and may expose difficult decisions about workforce, technology, finance and service design.

As a result, an organisation may repeatedly commission further exploration because exploitation is harder.

Risk appetite is an appetite for decisions, not harm

Discussions about patient safety and risk appetite require care. No responsible organisation has an appetite for avoidable patient harm or regulatory non-compliance.

However, an undifferentiated statement that the organisation is “risk averse” can conceal important trade-offs. Maintaining an unsafe status quo is itself a risk-bearing decision. Delaying implementation until every uncertainty has been resolved may allow known hazards to continue. Refusing to test an improvement because it may have unintended effects can preserve the unintended effects already present in the existing system.

Risk appetite is commonly understood as the amount and type of risk an organisation is prepared to pursue, retain or accept in achieving its objectives. NHS organisations increasingly differentiate appetites across areas such as safety, finance, workforce, innovation and infrastructure, rather than treating risk appetite as a single organisational setting [7].

Leadership therefore requires a more mature formulation:

We should have a low appetite for unmanaged patient-safety risk, but an appropriate appetite for controlled experimentation intended to reduce that risk.

This means creating bounded conditions for improvement: clear aims, appropriate patient and staff involvement, prospective risk assessment, defined safeguards, rapid feedback and explicit stopping or adaptation criteria.

A practical leadership test

Before extending a learning response, leaders and oversight groups should ask:

1. What decision would further learning enable? If the answer is unclear, the scope may no longer be proportionate.

2. What uncertainty remains material? Not every unanswered question would alter the choice of safety action.

3. Is the issue genuinely unknown, or repeatedly rediscovered? Recurring findings may indicate an implementation failure rather than a learning deficit.

4. What would we stop or delay by continuing this response? Opportunity cost should be visible in governance decisions.

5. Which resources are fungible, partly fungible or non-fungible? The answer will influence who needs to be involved and for how long.

6. What is the minimum effective dose of expertise? Specify which decisions require senior clinical or executive judgement, and which delivery tasks can be safely undertaken by other roles.

7. Are we counting opportunity cost as well as cash cost? Established staff time remains a scarce resource even when no new invoice is raised.

8. Has AI reduced end-to-end time, or only production time? Measure review queues, decision latency, implementation and benefits, not simply the speed of drafting or summarising.

9. Are we balancing exploration with exploitation? Boards should be able to see the resources committed to learning, implementation and evaluation.

10. How will we know that the resulting action has improved safety? Completion is not the same as effectiveness.

These questions support stronger clinical governance and corporate governance because they connect incident response to strategy, risk appetite, resource allocation, benefits realisation and board assurance.

From learning response to resource response

The success of PSIRF should not be measured by whether an organisation has replaced one investigation template with several new learning-response templates. Its potential lies in changing how organisations think.

A proportionate response is not necessarily a smaller response. It is one designed around the needs of those affected, the learning opportunity, the current level of knowledge and the realistic prospect of improvement.

Sometimes that will justify a detailed patient safety incident investigation. Sometimes it will require a rapid multidisciplinary conversation, a thematic review or the synthesis of existing evidence. Sometimes the most responsible decision will be to stop exploring a problem that is already sufficiently understood and redirect scarce capacity towards implementing and evaluating change.

The defining leadership mindset is therefore not “investigate everything” or “act quickly”. It is the disciplined ability to decide:

·         what we need to understand

·         what we already know

·         what resource the next stage will consume

·         what alternative use exists for that resource

·         where the next unit of effort is most likely to make care safer

Patient-safety resources are neither wholly fungible nor wholly fixed. Treating them intelligently requires judgement, explicit trade-offs and governance that values outcomes more highly than activity.

The same discipline must continue after learning. Implementation should not automatically attract the most senior or expensive resource available. It should attract the capability the work actually requires, with clinical and executive expertise concentrated at the points where judgement, authority or non-fungible knowledge are essential.

Applied well, this is not a choice between financial control and patient safety. It is a route to both. It releases scarce capacity without pretending that all roles are interchangeable, and it tests value by what changed rather than by how much activity surrounded the change.

AI makes that discipline more urgent. If information production accelerates while review, decision and implementation capacity remain fixed, the organisation will generate learning faster and act no sooner. The test of technology is not whether it helps us produce more. It is whether it helps the whole system decide and improve more effectively.

The aim is not to learn less. It is to ensure that learning has somewhere to go.

References

1. NHS England. (2025). Guide to responding proportionately to patient safety incidents, Version 1.3. Available online

2. NHS England. (2025). Patient safety learning response toolkit. Available online

3. Thaler, R. H. (1990). Anomalies: Saving, fungibility, and mental accounts. Journal of Economic Perspectives, 4(1), 193-205. https://doi.org/10.1257/jep.4.1.193 Available online

4. NHS England. (2019). The NHS Patient Safety Strategy. Available online

5. National Quality Board and NHS England. (2026). Quality strategy for NHS-funded care in England. Available online

6. March, J. G. (1991). Exploration and exploitation in organizational learning. Organization Science, 2(1), 71-87. https://doi.org/10.1287/orsc.2.1.71 Available online

7. NHS North East and North Cumbria Integrated Care Board. (2025). Risk Appetite Statement 2025/26. Available online

8. Jones, B., Horne, J. and Jones, D. (2024). Strengthening clinical leadership and management. The Health Foundation. https://www.health.org.uk/reports-and-analysis/briefings/strengthening-clinical-leadership-and-management

9. NHS England. (2023, updated 2024). NHS Long Term Workforce Plan. https://www.england.nhs.uk/long-read/nhs-long-term-workforce-plan-2/

10. Horton, T., Mehay, A. and Warburton, W. (2021). Agility: the missing ingredient for NHS productivity. The Health Foundation. https://www.health.org.uk/reports-and-analysis/briefings/agility-the-missing-ingredient-for-nhs-productivity

11. Jones, B. and Pereira, P. (2024). How improvement can help NHS productivity. The Health Foundation. https://www.health.org.uk/reports-and-analysis/analysis/how-improvement-can-help-nhs-productivity

12. NHS England. (2026). Productivity plan – update. https://www.england.nhs.uk/long-read/productivity-plan-update/

13. NHS England. (2026). 500,000 NHS staff to get new artificial intelligence tools to help free up more time for patients. https://www.england.nhs.uk/2026/06/500000-nhs-staff-to-get-new-artificial-intelligence-tools-to-help-free-up-more-time-for-patients/

14. Government Digital Service. (2025). Microsoft 365 Copilot Experiment: Cross-Government Findings Report. https://www.gov.uk/government/publications/microsoft-365-copilot-experiment-cross-government-findings-report/microsoft-365-copilot-experiment-cross-government-findings-report-html

15. Chomutare, T. et al. (2022). Artificial Intelligence Implementation in Healthcare: A Theory-Based Scoping Review of Barriers and Facilitators. International Journal of Environmental Research and Public Health, 19(23), 16359. https://doi.org/10.3390/ijerph192316359

16. Nair, M., Svedberg, P., Larsson, I. and Nygren, J. M. (2024). A comprehensive overview of barriers and strategies for AI implementation in healthcare: mixed-method design. PLoS ONE, 19(8), e0305949. https://doi.org/10.1371/journal.pone.0305949

17. World Health Organization. (2021). Ethics and governance of artificial intelligence for health: WHO guidance. https://www.who.int/publications/i/item/9789240029200