Interview analysis
Interview Analysis Matrix Example: From Source Quotes to Findings
Work through a filled interview analysis matrix with three fictional participants, source quotes, a counterexample and a missing answer. Download the CSV worksheets.

The example and its sources
An interview analysis matrix puts participants in rows and research questions in columns. Each cell keeps a short answer beside the evidence behind it. That arrangement lets you compare accounts without treating a memorable quotation as the whole study.
This example asks how operations coordinators prepare a weekly handover for the next shift. Every participant, quotation and situation below is fictional. The matrix illustrates a method; it is not customer research, a test of ScribeFlash or evidence of improved performance.
We have three different fictional people, one interview each:
- P01 / I01: an evening coordinator gathering updates from chat.
- P02 / I02: a day coordinator whose team uses a shared board.
- P03 / I03: a coordinator combining emailed updates from a field team.
The three research questions are:
- RQ1: How are handover updates collected?
- RQ2: What, if anything, delays the handover?
- RQ3: What makes the handover ready for the next shift?
You can open the complete teaching excerpts: I01 source text, I02 source text and I03 source text. References such as I01:L4 mean line 4 of that file, including interviewer lines. There is no recording, so this example has line references, not invented timestamps.
For the earlier work of choosing codes and defining a codebook, use the qualitative interview coding guide. Here we start with the excerpts already available and concentrate on the comparison.
A filled interview analysis matrix

The table separates the participant's exact words from the researcher's short reading. “Not asked” is a completed record of a gap, not an answer to infer. On a narrow screen, scroll within the table to read all three questions.
| Participant | RQ1: Collecting updates | RQ2: Handover delays | RQ3: Ready to share |
|---|---|---|---|
| P01 · I01 Evening shift | Quote: “I copy open requests from chat into the weekly handover sheet.” I01:L2Reading: Updates are manually gathered from chat. | Quote: “I wait until someone is named against each open request.” I01:L4Reading: P01 reports waiting for owner assignment. | Quote: “The next shift needs a named owner and the next action.” I01:L6Reading: Ownership and next action are P01's readiness criteria. |
| P02 · I02 Shared board | Quote: “Our shared board already lists open requests and their owners.” I02:L2Reading: P02 starts from an existing shared record. | Quote: “I do not wait for owner names; I send the link after the closing check.” I02:L4Reading: This account challenges a general claim about waiting for owners. | Quote: “A current board link is enough for the next shift.” I02:L6Reading: P02 considers the current shared record sufficient. |
| P03 · I03 Field updates | Quote: “I combine the emailed updates in a handover document.” I03:L2Reading: Updates arrive by email and are assembled into a document. | Quote: “I wait for the latest status from the field team, even when the owner is named.” I03:L4Reading: P03 reports a status-information delay, not missing ownership. | Status: Not asked. Source: No answer is present in I03. Reading: Readiness criteria remain unknown. |
Read across people and questions
Read a row first to understand one person's account. P01 gathers information from chat, waits for an owner and wants both an owner and a next action before sharing. Those statements belong together; separating them into three unrelated “pain points” would lose the sequence.
Then read a column to compare answers to the same question. RQ2 contains a reported ownership delay, an account without that delay and a different wait for current status. Coding all three as “handover friction” would conceal the useful distinction.
P01 also says, “The owner check is what keeps me waiting.” at I01:L8. This is a second mention from one person, not evidence from a fourth participant. The filled CSV retains both references in P01's RQ2 cell. If you interview the same person again, keep their participant ID while giving the new session its own interview ID.
P02's answer is counter-evidence to “coordinators cannot hand over work until they find an owner.” It does not erase P01's experience. It suggests checking the conditions around each account, including the shared board and closing check, before choosing an explanation.
Turn the comparison into a bounded finding
An overbroad draft would be “Missing owners block weekly handovers.” It ignores P02 and misstates P03's difficulty.
A more defensible reading of this fictional set is:
P01 reports waiting for owner assignment, while P03 waits for current status even when an owner is named. P02 reports no comparable ownership wait and uses a shared board. These accounts describe different handover conditions rather than one common ownership problem.
Keep the references with that statement: I01:L4/L8, I02:L4 and I03:L4. They support the comparison, but do not establish how often these delays occur in real teams.
A possible follow-up is to observe how people check ownership and status during a handover. That is a proposed research action. The interviews do not prove that introducing a shared board would remove either delay, nor that a particular feature would improve completion time.
Review one candidate finding
Fictional teaching review, not customer research or measured product results. Keep this finding-level check separate from the participant-by-question matrix above.
| Review field | What to check |
|---|---|
| Candidate finding | These fictional accounts describe different waits: owner assignment for P01, current status for P03. |
| Supporting sources | I01:L4, I01:L8; I03:L4: P01 reports an ownership wait; P03 reports a status wait. |
| Counter-evidence | I02:L4: P02 does not wait for owner names. This challenges a claim that all three wait for owners. |
| Scope | Only P01–P03, one fictional interview each. The two I01 mentions are still one person; these are self-reports, not a prevalence estimate. |
| Unknowns | RQ3 was not asked in I03. P03’s readiness criteria and any measured effect remain unknown. |
| Next check | Observe ownership and status checks, then ask P03 about readiness before broadening the claim. These checks have not been done. |
Leave the unknowns visible
RQ3 was not asked in I03. Do not fill its cell from P01's criteria, treat silence as agreement or include P03 in a claim about readiness preferences. Write “not asked” and decide whether a follow-up is needed.
All three accounts are reported experiences. There was no observation of an actual handover in these excerpts. We also have no session dates, timings, failed-task counts or measured outcomes. Keep those fields unknown rather than borrowing the document's creation date or inventing a baseline.
The sample is deliberately tiny and fictional. Counts describe the worksheet, not a population: three people, three interviews and nine question cells, of which eight contain an answer. A source match makes a reading inspectable; it does not make every interpretation correct. The AI finding verification guide covers that next review.
Download and reuse the worksheet
Use the blank CSV for analysis after interviews, not as an interview script or a ScribeFlash import file.
Use the matrix to compare answers by participant and question. After that comparison, use one row per candidate finding in the finding review sheet. Record supporting and conflicting sources, scope, unknowns and the next check; keep references such as I01:L4 so you can reopen the source.
- Filled matrix CSV: nine question cells, quotations, source references, short readings and the explicit gap.
- Blank matrix CSV: the same long-form columns, ready for your own research questions and source records.
- Blank finding review CSV: six fields for reviewing one candidate finding per row, using the worked example above.
- I01, I02 and I03: the fictional source excerpts used above.
Open the CSV in a spreadsheet and filter by question or participant. Use one row per participant–question cell; keep additional mentions and session references together instead of silently turning them into extra people. Add columns when your study needs them, such as concept version or the conditions under which an observation occurred. These are manual teaching worksheets, not a ScribeFlash import format.
Take the evidence into a project
For the product workflow, explore cross-interview analysis in ScribeFlash. To inspect how source material, evidence and a report fit together, open the sample project before bringing your own authorized material.
Keep the filled matrix alongside the source excerpts and review any finding against both supporting and conflicting cells. You can then return to the complete interview analysis workflow for the wider process. The matrix's useful output is a comparison another researcher can check, including the parts you still cannot answer.
Compare interviews without losing the source
Explore the cross-interview workflow, then inspect a sample project before starting your own study.
