Research

Critical incident technique: learn from the moments that matter

Dylan de Heer, Co-Founder & CPO

Dylan de Heer

Critical incident technique: learn from the moments that matter

The critical incident technique is a research method for examining specific events that made a meaningful difference to an outcome. Instead of asking only what someone generally thinks of a process, you ask them to describe a particular occasion, reconstruct what happened and explain why it mattered.

A useful account contains a situation, an action or event, and a sufficiently clear consequence. “The handover process is confusing” is an opinion worth exploring. “On Tuesday, the recipient checked the attachment, noticed the wrong version and stopped it reaching the client” gives you an episode you can investigate.

This guide explains what makes an incident critical, how to collect usable accounts, and what you can reasonably conclude from them. The emphasis is practical research into work and product experiences, rather than emergency response or an investigation into individual blame.

What makes an incident critical?

“Critical” refers to significance for the outcome you are studying. The event need not be a crisis, a spectacular failure or the worst thing a participant can remember. A small action that prevented a problem can be just as useful as a visible breakdown.

In the original formulation, John C. Flanagan set out a method for collecting sufficiently complete observations of behaviour against defined aims and criteria. The method therefore starts with a purpose and an evidence boundary, rather than a collection of striking stories. Flanagan’s original 1954 paper explains that foundation.

For a study of project handovers, your focus might be whether a receiving team can begin the agreed work with the information it needs. An episode about a late train may be memorable but irrelevant unless it affected that handover. An unremarkable clarification that exposed a missing approval could fit the study directly.

Before treating an account as an incident, establish:

  • The intended result: what the person was trying to achieve.

  • The circumstances: enough context to understand the situation.

  • The consequential action or event: what actually happened, as far as the available account establishes.

  • The result: what changed, and how the participant knows.

  • The evidence limits: what remains uncertain, inferred or known only second-hand.

These are practical review questions for your study. They are not a scoring system or a guarantee that a participant’s explanation establishes causation.

How it differs from an ordinary interview question

An interview can explore opinions, expectations, routines, preferences and specific experiences. A critical incident question deliberately selects an experience because of its effect on a defined outcome.

Compare three original questions about a handover:

Broad opinion: “How well do handovers work here?” This can reveal the participant’s overall view, but it may leave you without an event to inspect.

Recent example: “Walk me through your most recent handover.” This gives a concrete starting point. The latest occasion may have been routine and may contain no especially consequential event.

Critical incident: “Think of a handover when something someone did made a clear difference to whether the receiving team could begin. What happened?” This connects the recalled event to the research purpose while leaving room for a helpful or harmful effect.

NN/G describes the technique’s use in interviews and other research settings, including the distinction between requesting a general example and requesting a consequential episode. Its explanation of CIT in UX research is a useful methodological reference.

Do not make every interview question an incident question. Context about a person’s role or normal workflow can help you understand the episode. The technique gives a particular kind of evidence within the wider study.

When the technique is useful

Consider it when you need to understand how an outcome came about, especially when the relevant event may not occur during a short observation session. For example, a researcher might investigate an unusual recovery, a successful workaround or a handover that stalled at a particular point.

It can also help turn an abstract complaint into a question that is easier to investigate. If someone says “we never know who owns the next step”, ask for a particular occasion and the sequence that followed. You may find an unclear rule, conflicting expectations, missing information, or a different explanation altogether.

The method has important limits. Recalled incidents can lose detail, and memorable episodes may give a poor picture of ordinary usage. NN/G explicitly discusses both limitations. Use those cautions when deciding whether you also need observation, usability testing, documents or other evidence. Read its discussion of the technique’s strengths and limits.

Do not use a few accounts to estimate how often a problem occurs across a population. Nor should you assume that the most dramatic story deserves the highest product priority. That decision needs additional evidence about the affected task, consequences, scope and alternatives.

Define the study before collecting stories

Write a short research purpose. “Understand handovers” is broad. “Understand events that helped or prevented the receiving team from starting an agreed task” gives the interview a clearer boundary.

Then decide who can offer first-hand accounts, which kinds of events fit, and which context you need. Include a way to examine successful outcomes as well as unsuccessful ones. A process may work because somebody notices and repairs a weakness; studying only failures can hide that contribution.

Keep the incident boundary consistent enough for comparison. If one account concerns a whole six-month project and another concerns a single email, you will need to clarify which episode in the longer account is relevant. Do not force everything into an identical duration. Instead, establish a beginning, the consequential event and enough of the aftermath to understand the result.

Pilot the questions before relying on them. Check whether people understand the outcome you mean and can describe an event without being steered towards your preferred explanation.

Ask for an account before asking for an explanation

Give the participant time to recall an occasion. If they cannot identify one, record that limitation rather than manufacturing an example with them. An invented scenario can be useful for a different purpose, but it is not a recalled incident.

Start with an open invitation, then use neutral follow-ups:

  1. Locate the situation. “What was happening just before this?”

  2. Establish the goal. “What were you trying to get done?”

  3. Reconstruct the sequence. “What happened next?”

  4. Clarify actions. “What did you do, and what did the other person do?”

  5. Explore the consequence. “What changed after that? What tells you it changed?”

  6. Separate knowledge from inference. “Did you observe that yourself, or hear about it later?”

  7. Check the account. “Is there anything important missing from the way I have understood this?”

These suggested questions are original examples, not a validated interview instrument. Adapt them to your research design and avoid reading them as a compulsory script.

Avoid questions that build in the answer: “How did poor communication cause the delay?” already names both a cause and a consequence. “What information was available at that point?” leaves more room for the participant’s account.

A fictional example: a handover that nearly went wrong

The following scenario is invented to demonstrate the method. It is not a Weeve customer story or a research finding.

A receiving team member describes opening a project folder before sending a document to a client. The file name looked correct, but its contents referred to an earlier scope. They asked the project lead which version had been approved. The lead supplied a newer document, and the team used that one instead.

An initial note might say: “Careful staff prevent version mistakes.” That jumps from one episode to a broad explanation about people.

A better incident account preserves the sequence: the team member inspected the contents, detected a mismatch, sought clarification and obtained a replacement. It also preserves uncertainty: the researcher does not yet know why the earlier file remained in the folder or whether the client would otherwise have received it.

Useful next questions include whether both files had similar names, what counted as approval, how the replacement was identified, and whether another person would have had access to the same information. A proposed interface change can wait until you understand those conditions.

The incident may suggest a research question about version visibility. It does not yet prove that every handover needs a new approval tool.

Turn accounts into comparable evidence

Keep a distinct record for each episode, even when one participant describes several. Give it a stable identifier and retain a route to the supporting notes, transcript or permitted recording. Mark whether the account is first-hand, retrospective, incomplete or supported by another source.

Separate the account from your interpretation. For the fictional example, “the person requested a newer file after noticing an old scope” describes the account. “The folder made approval status hard to identify” is an interpretation to test.

Compare incidents around a defined question. Which conditions recur? Where does the same action have a different result? What happens when a different person takes over? Our guide to cross-case analysis explains how to compare accounts while keeping their context.


Several cream cloth parcels sit on a dark shelf, with one orange parcel set apart for examination, representing selection of an incident without treating it as every case.

Do not flatten a successful recovery and an unsuccessful handover into the same “communication problem” label without checking what differs. Also distinguish no evidence of a problem from evidence that the problem did not occur.

When reporting, explain how incidents were selected and what the material can support. An important episode can reveal a possible failure mechanism without showing its frequency. Keep those two conclusions separate.

Use recordings to check the account, with appropriate permission

A recording can help you revisit what was said in the interview. It does not independently verify the event being described. A clear transcript of a participant’s recollection remains a record of that recollection.

With appropriate permission, Weeve can help record research interviews and revisit saved material. In Chat with recordings, ask a focused question about the relevant account, then inspect the supporting passage and its surrounding context.


A genuine Weeve transcript reader showing the synthetic Pilot planning conversation and timestamped speaker turns.

Genuine Weeve interface with synthetic demonstration content. The Pilot planning passage demonstrates checking a transcript source; it is not an example from a critical incident study.

A useful request might be: “Find the passage describing what happened before the handover was paused. Separate the participant’s stated actions from their explanation of why it happened. Keep a source reference and flag missing detail.” Review the result before adding it to your analysis.

Check names, sequence, speaker attribution and any uncertainty that affects interpretation. Follow the study’s information-handling rules. If you choose an optional external ChatGPT or Claude connection, relevant text can go to that service; choose a route suitable for the material.

Common mistakes to catch before reporting

Calling any vivid story critical. Return to the research purpose and the event’s relationship to the outcome. Include routine context when it changes the explanation.

Collecting only failures. Investigate helpful actions and recoveries where they fit your question. A successful result can contain work that the formal process does not acknowledge.

Replacing the sequence with a personality judgement. Write what the account says happened before labelling someone careless, resistant or capable.

Confusing a participant’s explanation with a demonstrated cause. Preserve their interpretation as an attributed account and look for other evidence where the conclusion requires it.

Turning incidents into prevalence figures. Report the scope and selection of your material. Do not present a handful of recalled episodes as a measured rate of failure.

Finish with the next evidence check: a document to inspect, an alternative account to explore, a routine task to observe or a question for another interview. The technique is most useful when a concrete event leads to a clearer investigation, rather than a more confident version of an assumption.