Accident Investigation Interviews: How Existing Approaches Have Critical Blind Spots

Safety managers tell me with real frustration that they just don’t seem to be learning – that the same accidents keep repeating. That frustration is backed by data. SIF rates have remained stubbornly resistant to improvement across multiple high-hazard industries, despite sustained effort and investment in conventional safety management approaches.

The research evidence points to a specific and largely unaddressed reason for this. Studies across a range of high-risk sectors – including offshore drilling (Roberts et al., 2015), aviation (Beaty, 1995; Weick, 1991), and healthcare (Rogers et al., 2006) – consistently show that lapses in situational awareness, rather than work system glitches or technical shortfalls alone, play a significant contributory role in the large majority of SIFs.

Situational awareness is one of a cluster of capabilities now widely referred to as non-technical safety skills (NoTSS) – alongside decision-making, communication, teamwork, leadership, improvisation, and metacognition (Flin et al., 2008; Flin et al., 2016; Hutton, 2019). These accidents also share another characteristic: they routinely occur within complex work systems, rather than merely complicated ones.

Complex systems are non-linear, unbounded, and dynamic. They generate uncertainty that cannot be fully predicted or designed out. Standard operating procedures remain essential, but alone they are insufficient for managing the uncertainties that complex dynamic environments generate (Pemberton & Mooney, 2023). It is now widely recognised across high-risk sectors that equipping workers with non-technical cognitive safety skills is critical – not optional – for safe performance in these environments (Flin et al., 2008; Flin et al., 2016).

Yet in my experience, and as my research confirms, accident investigations across many sectors routinely fail to explore cognitive factors and NoTSS at all. The reason is straightforward: accident investigators are often not trained or equipped to do so. This is the core problem this blog – and the research underpinning it – addresses directly.

Most safety professionals recognise that human factors play a role in accidents. But in my experience, “human factors” is too often interpreted as the physical dimension – equipment ergonomics, the working environment, interface design. These things matter. But there is a cognitive dimension of human factors that is, almost as a matter of routine, completely overlooked in accident investigation practice.

The root of the problem lies in the mental models that accident investigators often bring to their work. Research into investigator biases demonstrates clearly how these mental models shape and constrain the entire investigation process. Lundberg, Rollenhagen and Hollnagel (2009) coined the phrase “What You Look For Is What You Find” specifically to describe this phenomenon in accident investigation.

The investigator’s underlying mental model determines what lines of questioning they pursue, what evidence they gather, and ultimately what learning they generate. If that mental model does not include cognitive human factors, then cognitive human factors will be invisible – investigation after investigation.

The empirical evidence for this is compelling. Thallapureddy et al. (2023) conducted simulated accident investigation interviews with 34 experienced investigators across multiple high-hazard sectors and identified a consistent pattern of cognitive biases shaping investigative practice.

Confirmation bias led investigators to pursue lines of questioning that confirmed their pre-existing beliefs about what had happened – often from the very first phone call after an incident. Anchoring bias caused investigators to fixate on the first salient observable workplace factor they identified, to the exclusion of other contributory factors.

Fundamental attribution error inclined investigators to attribute the cause of an incident to individual character rather than the wider situational context or indeed cognitive aspects, such as decision-making. Hindsight bias led them to conclude the incident was “obviously preventable” – which shuts down genuine inquiry before it begins. These biases were not independent; they interacted and compounded each other (Thallapureddy et al., 2023).

Critically, my own research has demonstrated that the training and mindsets of accident investigators are not equipped to encompass cognitive human factors. The majority of investigation approaches – and the training that supports them – are rooted in a workplace, equipment, and linear root-cause mental model.

The cognitive dimension of accidents falls entirely outside that frame. The result is that much of what we think of as “human factors investigation” is still firmly focused on physical factors, systems, and processes – while the headspace remains almost entirely unexplored.

In my research, I examined an accident investigation conducted following a serious incident in a well-run organisation with experienced, competent safety professionals (Pemberton & Mooney, 2023). This was not a poorly conducted investigation. It was a thorough, professional piece of work – representative of what good conventional investigation practice looks like.

And yet it had two significant blind spots. The first was its exclusive focus on workplace factors – environment, equipment, systems, and processes. The second was its focus on errors: what went wrong. Neither blind spot is unusual. Both are, in my experience, characteristic of conventional investigation practice.

The investigation examined the workplace environment, equipment states, systems, and processes. It documented the sequence of events and identified procedural failures. It made recommendations around workplace controls. Such insights are helpful. By conventional standards, it was a solid investigation. But it had not explored the cognitive work at all.

It had not asked what the people involved were perceiving, understanding, or deciding at critical junctures. It had not examined how situational awareness had developed – or failed to develop – in the lead-up to the incident. It had not explored the non-technical safety skills – communication, teamwork, decision-making under pressure – that were subsequently found to have played a significant contributory role. The focus was exclusively on errors: what went wrong. There was no attempt to understand what needs to go right.

This is what I call the workspace versus headspace distinction. The workspace – physical environment, equipment, systems – was reasonably well examined. The headspace – the cognitive work of the people involved – was unexplored. And because the cognitive dimension was not explored in the original investigation, the most important learning from this accident was never generated.

This is not an isolated finding. I have replicated this published research across more than 19 other serious accidents – and found the same results. The two blind spots – an exclusive workspace and error focus – appear consistently, across sectors and organisation types.

Why is it so? To begin with, the large majority of accident investigators are simply not equipped with the expertise needed to undertake a cognitive human factors-oriented accident interview (Pemberton & Mooney, 2023). Interview approaches are often too short, too superficial, and too poorly recorded to generate the quality of data that meaningful learning requires.

In the worst cases, witnesses are asked to write a police-style statement. In better cases, an investigator takes notes during a conversation of perhaps thirty minutes. As Sherlock Holmes observed: “Data! Data! Data! I can’t make bricks without clay.” Existing accident investigation approaches are literally not collecting the data required for effective learning – because they are not looking in the right place, and do not have the tools to dig.

Cognitive task analysis (CTA) is a structured approach to understanding the cognitive work that underlies human performance – how people perceive situations, build understanding, make decisions, and take action (Pemberton & Mooney, 2023).

In the context of accident investigation, CTA interview techniques are designed to explore the headspace: not just what happened, but how people were thinking at each critical decision point – a more complete form of accident investigation interview techniques.

I applied a CTA approach in the above research project and demonstrated that this approach can generate qualitatively different and substantially richer learnings following a serious accident (Pemberton & Mooney, 2023).

Where the conventional investigation had gathered surface-level workplace accounts of events, the CTA interviews revealed the cognitive work underlying those events: what information was being used, what cues were being perceived or missed, what the individuals understood the situation to be at each critical juncture, and what would have needed to be different for different decisions to have been made.

The contrast was stark. CTA interviews are substantially longer and more intensive than conventional accident investigation interviews – and that depth is not incidental, it is the point. But more importantly, what they deliver is qualitatively different: cognitive insights and positive capacities that conventional interviews simply cannot surface.

To give a concrete example: a conventional investigation interview might record a workplace account such as “I checked pressure gauge before proceeding,” a CTA interview would go further and might include comments such as “the pressure gauge was telling me was probably safe to proceed – but it does not tell you the whole picture – you have to make a call in the information you have at that point – I never imagined it could have affected the process as it did.

This type of richer narrative explores three cues the operative was sensing: the goals they had, what their mental model of the bigger picture was, and how these formed their decision-making.

Critically, as we will see, the richest learnings come from comparing the cognitive approach of this involved with resident experts. This shines a light on the difference between knowing what went wrong and understanding what needs to go right.

It is now also my firm view that recording interviews and using AI-based transcription and analysis is becoming fundamental to getting the depth of insight that CTA interviews can generate. Properly recorded and transcribed CTA interviews provide the raw material – the clay – from which a meaningful learning structure can be built.

In my research, the CTA interview approach identified seven difficult decisions in the accident sequence. A difficult decision is one that a less experienced practitioner would be likely to get wrong – situations of genuine cognitive complexity, where the right course of action is not obvious and where expert pattern recognition makes a critical difference to outcomes.

For each of these difficult decisions, I built a Decision Requirements Table (DRT): a structured document that captures the cognitive demands of each decision point, the cues experts use to read the situation, and the mental models – so-called cognitive diamonds – that underpin expert judgement.

The DRT is the bridge between investigation insight and practical application: it provides the data needed both for simulation training design and for improving system and task design to better support people when facing these decisions in critical operational contexts.

Both technical cognitive skills and non-technical safety skills – particularly situational awareness, communication, and teamwork – were implicated across these decision points (Pemberton & Mooney, 2023).

The most powerful – and most neglected – element of the CTA approach in accident investigation is the use of resident expert interviews. In my research, having completed the CTA interviews with the accident participants, I went further. I identified experienced practitioners who had not been involved in the accident, but who had deep operational knowledge of the same work context, and took them through simulation interviews (Pemberton & Mooney, 2023).

The simulation interview approach presents the expert with the scenario as it unfolded at each critical decision point, and asks: ‘What would you have done here, and why? What would you be looking for? What would be telling you something was wrong? How would you check your understanding?’

What these types of accident investigation questions (and there are lots of them) generated was something the conventional investigation had no mechanism to produce: a systematic account of what needs to go right. The resident expert interviews revealed the cognitive diamonds – the mental models that expert practitioners use to make safe, reliable decisions in complex and dynamic situations.

These are the pattern-matching capabilities described by Klein’s (1992) Recognition-Primed decision-making (RPD) model – experts matching mental models to environmental cues across all three stages of situational awareness: perceiving what is present (Stage 1), understanding what it means (Stage 2), and projecting what will happen next (Stage 3).

The Decision Requirements Table built from the CTA process detailed the cognitive diamonds that experts deployed at each of the seven difficult decision points – covering both technical knowledge and non-technical safety skills. This is learning that exists nowhere in the conventional investigation record.

In my experience, organisations consistently underestimate the value of the tacit expertise their resident experts hold. CTA-based investigation is a practical way of surfacing and preserving that expertise – which is also, as Klein (1992) highlighted, a critical mechanism for preserving corporate memory before experienced practitioners retire.

The cognitive diamonds identified through the CTA process are not just analytically interesting – they are directly actionable in two important ways. First, as training content: the difficult decisions and associated cognitive diamonds provide the blueprint for simulation-based training that builds the tacit skills that matter.

Second, as the basis for improved system and task design – what is sometimes called cognitive task design – to better support people when facing difficult decisions in critical operational contexts. This is not a “fix the worker” approach. It is about redesigning the joint cognitive system so that safe, expert-level decisions become more consistent and more likely.

In my research, I used a Decision Requirements Table to design a simulation-based training intervention using a Decision-Making Exercise (DMXs) approach (Pemberton & Mooney, 2023).

DMXs are short scenario-based simulations – and can be delivered via face-to-face paper-and-pencil exercises to more sophisticated online, AI-supported versions.

DMXs place participants in the kinds of complex, ambiguous decision situations that characterise their real work environment. They are explicitly designed to build the tacit cognitive skills that matter: situational awareness, pattern recognition, decision-making under uncertainty, and the communication and teamwork behaviours that make safe performance consistent.

I benchmarked a cohort of existing employees before the training intervention and assessed them afterwards with a DMX intervention. These were experienced workers – averaging nearly ten years of operational experience – not new starters. And yet, pre-DMX training, they demonstrated less than a quarter of the tacit knowledge documented in the resident expert interviews.

That finding alone tells you something important: when tacit knowledge transfer is left to chance, it is slow, inconsistent, and deeply incomplete. Decades of experience do not guarantee that the right cognitive skills have been developed, particularly in complex, dynamic environments where the learning opportunities are unpredictable.

The DMX results were striking. Through approximately two hours of simulation training, I was able to significantly improve tacit knowledge across the key cognitive skills implicated in the accident (Pemberton & Mooney, 2023). This is consistent with a growing body of research on accelerated tacit expertise development.

Hoffman et al. (2014) have laid out the evidence and approach for accelerated learning and training for high proficiency in complex domains, demonstrating that structured cognitive skill development can dramatically compress the time taken to reach expert-level performance. Militello et al. (2023) have similarly demonstrated a range of cognitive training methodologies effective for accelerated tacit skills development across safety-critical environments.

The case for a cognitive human factors approach to accident investigation rests on three interconnected arguments – each of which is becoming more urgent.

First, as we have seen, cognitive factors are central to SIF causation. Situational awareness, decision-making, improvisation, and metacognition play a fundamental role both in making safety-critical situations go right 99.999% of the time, and in triggering serious and fatal incidents when they go wrong (Flin et al., 2008; Hutton, 2019). Understanding what was going on inside the heads of those involved – and comparing this with what resident experts would have done – is the key to deeper learning and the development of the positive capacities that prevent recurrence.

Second, traditional mechanisms for developing these cognitive competencies are no longer adequate. It has historically taken decades for workers to develop tacit safety skills through real-world experience and informal on-the-job learning.

But workforce tenure is falling across many industries, and new workers are entering complex, high-risk environments without the accumulated cognitive experience that safe performance requires. CTA-based investigation and DMX simulation training offer an evidence-based, practical way of accelerating that development (Pemberton & Mooney, 2023).

Third, the brain drain risk is real and growing. As resident experts with decades of operational experience retire, they take their cognitive diamonds with them – unless those diamonds have been systematically captured. Klein (1992) highlighted that CTA provides a practical mechanism for preserving corporate memory before it walks out the door.

The tacit knowledge held by resident experts is an extraordinarily valuable safety asset. CTA-based learning reviews are a concrete way of ensuring that the asset is identified, documented, and transferred – and the DMX training results from my research demonstrate that this transfer can be achieved with remarkable speed.

Approximately two hours of targeted simulation training produced significant improvements in tacit knowledge that years of unstructured experience had not delivered. This is the promise of accelerated expertise development: compressing decades of informal learning into a few months, perhaps weeks, via a structured, measurable, and scalable training approach (Hoffman et al., 2014; Militello et al., 2023).

The cost of continuing to leave this accident learning on the table is not abstract. It is measured in repeat accidents, in serious injuries, and in fatalities that a fuller understanding of cognitive human factors could have helped prevent.

The evidence is clear. Conventional accident investigation interviews have critical blindspots. By focusing exclusively on the workspace – physical environment, equipment, systems, and processes – and on errors rather than expertise, they generate only partial learning.

The cognitive dimension of accidents: how people were perceiving, understanding, and deciding at critical junctures, and what distinguishes expert performance from less experienced performance, remains almost entirely unexplored. Exploring what normally goes right via resident experts is also routinely unexplored.

My research, applied to a real, serious accident investigation and replicated across more than 19 additional cases, demonstrates that cognitive task analysis interview techniques deliver substantially deeper learning – surfacing the cognitive diamonds that explain both what went wrong and what needs to go right.

Resident expert interviews reveal the tacit knowledge that makes safe performance reliable. Decision-Making Exercises translate that knowledge into measurable competency gains, even in experienced workers, through short and focused simulation-based training. And, the insights from CTA extend beyond training into the design of the joint cognitive system itself – making safe decisions more consistent at a systemic level.

At Human Focus International, we provide specialist Human Factors in Accident Investigation training that covers the approach described throughout this blog – including how to apply cognitive task analysis in accident investigation interviews, how to structure and record CTA interviews to the standard required for meaningful learning, and how to leverage AI-assisted transcription and analysis to maximise the depth of insight generated.

Our services extend beyond investigation methodology. We help organisations leverage resident expert knowledge through the design and delivery of DMX simulation training – building the tacit cognitive skills that CTA identifies as safety-critical. We also support improvement of the joint cognitive system through consultancy and coaching, as well as through an online competency development system that makes evidence-based cognitive skills training accessible at scale.

If you are serious about getting more learning from accident investigation – and building the cognitive capabilities that prevent recurrence – we would welcome a conversation.

[Contact Human Focus International] | [Download the Research Paper: Pemberton & Mooney, 2023]