Analyzing in-depth interviews: a table and sample findings

First, formulate a research question

After five conversations, it’s easy to remember the emotional participant and build the entire report around his story. Another risk is to get five neat retellings, from which it is unclear what to change in the product. Before the analysis of the records, write down what decision the team should make.

For example: “Why do small team leaders agree on documents in correspondence, although they have a system of tasks?” This is more specific than “knowing the needs of users.” In the first case, you will look for situations of harmonization, participants, workarounds and consequences. In the second, almost any line will seem important.

Separate the initial assumption from the result. The phrase “they need a new interface” is a possible explanation. The question is, “What step do people get out of the system?” allows you to check other reasons: lack of access, incomprehensible responsibility or habit of colleagues.

Prepare the transcription and save the context

Give each participant a neutral code: R01, R02, R03. Next, specify only signs important for research, such as the role and experience of working with the product. The work table does not need personal phones and full names.

1. Check the speakers. The question of the interviewer should not be the opinion of the participant.
2. Check the semantic places. Denials, numbers, conditions, and names are especially important. “I don’t send” and “send” lead to opposite conclusions.
3. Leave the fragment address. The recording code and timecode will help the colleague quickly check the quote.
4. Keep the neighborhood replicas. The answer “yes” without a question is almost useless, and an ironic replica without context can look like an endorsement.

Do not correct live speech so that the participant becomes more confident than it was. “Probably a couple of times” cannot be turned into “regularly.” If the record is illegible, note this and do not use the phrase as an accurate proof.

The detailed order of the proofreading is in the article How to Check and Fix Transcription.

Highlight the observations and assign them short tags

Re-read the interview in its entirety, then note the fragments that answer your question. One fragment may contain several observations. A user-friendly label describes an action or obstacle: “waits for the coordination of the manager”, “duplicates the link to the chat”, “does not see the status of the document”.

In the manual GOV.UK on analysis of research sessions observations are proposed to be separated from their interpretation, grouped in meaning and only then proceed to conclusions and actions. In practice, different columns of the table are enough for this.

Don’t start with appraisal labels like a “lazy customer.” Write the observed: “did not open the instruction before the first launch.” The reason is unknown. Perhaps the person did not find the link or the instruction did not answer his question.

If you disassemble the material together, first compare a few fragments. Discuss what everyone understands by labels. The disagreement is useful to record: it shows a place where data or context is not yet enough.

Analysis matrix: a completed example

Below is a fictional example of a study of document harmonization. Quotes and timecodes are created to explain the method.

All three mentioned chat, but the reasons are different. Combining them into a conclusion “users don’t like the system” – to lose the content of the interview. A more useful result: highlight three scenarios and check each separately.

For your table, use the fields: participant → context → exact quote → timecode → observation → interpretation → open question. If the string becomes too long, leave a short fragment in it, and store the full transcription separately.

How to write a conclusion that can be verified

First compare cases within one group: for example, managers with managers. Then look at how the experience of the coordinators is different. The same words in different roles can mean different tasks.

If you consider mentions, consider the participants and determine the rule in advance. Ten replicas of one person are not equal to ten independent confirmations. The wording “three of the six respondents described such a case” refers to your group, not to half of potential buyers.

Do not prioritize only the frequency. A single obstacle can completely block the task of an important segment. In the report, share the prevalence in the sample, the consequences of the problem and the confidence in the explanation.

What to transfer to the command instead of the transcription folder

Gather a short document: a research question, who was interviewed, basic observations, supporting fragments, contradictions and the following checks. At the end, attach a matrix and references to sourcers with suitable access rights.

For each withdrawal, use one card:

Sammari helps to remember the content of the conversation, but primary replicas are needed for analysis. If you use AI for a draft grouping, ask to save the participants’ codes and references to snippets. Check every quote offered: a beautiful generalization without a source does not fall into the report.

transcription removes some of the work on the transfer of speech to the text. Interpretation, choice of priority and product solution remain the task of the researcher and the team.

Verification before discussing results

Ask a colleague to choose one take and walk from it to the record. If the chain is restored without your oral explanations, the material is convenient to check and use further.