Turn Interview Notes Into Product Findings Without Losing Disagreements

Keep the question, speaker and situation beside a statement before turning it into a product finding. Then compare what each person describes doing, what troubled them and how they handled it. A repeated word can be useful for locating material without proving that everyone has the same need.

One public writer described an interviewer's contribution being attributed to a participant during AI-assisted synthesis. Another asked how to find common themes in a large amount of interview material. These are individual accounts, with exact posting dates unverified. They do not establish a tool's current reliability or validate the approach below.

The worked example is entirely fictional. The excerpts, people, project, extractions and summaries were authored for teaching; no interviews were conducted and no synthesis tool was tested. This is one practical sorting exercise, not a certified research procedure.

Start with three complete excerpts

The fictional Juniper team uses a shared rota sheet and messages to arrange shifts. The three excerpts are separate conversations. I is the interviewer. P1 and P2 are team members; P3 coordinates changes. The turn numbers preserve who spoke first within each excerpt. They do not describe a sample of real customers.

E1: a member checks a changed arrangement

E1-T01 · I:
Would automatic reminders help you avoid missing a shift?

E1-T02 · P1:
The Sunday reminder already arrives. On my last swap, the other person
and I agreed in chat after that reminder, but the sheet still showed
our old shifts.

E1-T03 · I:
What did you do when you saw that?

E1-T04 · P1:
I reopened the chat and the sheet, then called the coordinator before
leaving. They said the swap was confirmed and told me which start time
to follow. I used that answer. I do not know who was meant to update
the sheet.

E1-T05 · I:
Would another reminder have fixed that moment?

E1-T06 · P1:
Not if it repeated the old sheet. I wanted to know which arrangement
I could act on.

E1-T07 · I:
What do you do in a week without a swap?

E1-T08 · P1:
I check the sheet once and go. I would not want extra alerts for those
weeks.

The interviewer introduced automatic reminders before P1 described the incident. P1's account concerns different cues in chat and the sheet, followed by a call to the coordinator. It does not say that a shift was missed. Nor can this excerpt tell us whether P1 would have raised reminders without the interviewer's suggestion.

E2: a member describes an ordinary week

E2-T01 · I:
Tell me how you used the rota for your most recent ordinary week.

E2-T02 · P2:
On Monday I opened the published sheet, checked my shifts and wrote
them in my diary. My hours are normally the same.

E2-T03 · I:
What would a reminder before each shift add?

E2-T04 · P2:
I keep routine notifications muted. Once I have written the week down,
I do not need another reminder for that routine.

E2-T05 · I:
What happened the last time a shift changed after you had written it down?

E2-T06 · P2:
That has not happened to me recently, so I do not have an example to
give you.

P2 supplies an ordinary-week practice and a reason that extra reminders are not wanted in that routine. There is no recent change-handling story here. Do not turn that absence into a claim that P2 never experiences changes or would always handle them easily.

E3: a coordinator distinguishes confirmation from a nudge

E3-T01 · I:
Walk me through the most recent shift swap you coordinated.

E3-T02 · P3:
I marked the request pending, asked both people to confirm, then updated
the shared sheet and posted a confirmed message with the sheet link.
Nobody came back with a question about that swap.

E3-T03 · I:
Where do reminders fit in your work?

E3-T04 · P3:
If I ask for a reminder, I mean a nudge for me to finish the update after
the confirmations. I do not mean another alert for every person on the rota.

E3-T05 · I:
What happens when a swap starts while you are away?

E3-T06 · P3:
Sometimes people agree something in chat without telling me whether both
have confirmed. I ask both people before I change the sheet. One person's
message is not enough for me to treat it as confirmed.

E3-T07 · I:
Would automating the messages remove that uncertainty?

E3-T08 · P3:
I cannot say. A message can still describe a request rather than an agreed
change. The example I just gave went smoothly when we followed the
confirmation steps.

P3 uses “reminder” for their own follow-through after confirmation. That is different from reminding members when to attend a shift. The reported smooth example is also important. P3's account that nobody returned with a question is not a direct observation of every member's understanding, but it is a case the summary must not erase.

Extract every turn before writing the theme

The following is a manual extraction of the complete excerpts. Short descriptions point back to the full wording above. Interviewer questions stay in the table because they help explain where a feature suggestion came from; they are not participant requests.

Some turns contain more than one kind of statement. P1's action in T04 is reported behavior; not knowing who should update the sheet is a reported information gap. Neither establishes what the application's implementation did. Keeping those limits visible makes a shorter finding easier to defend later.

Compare situations, behavior, difficulty and existing responses

Now combine turns within each conversation without merging the people:

This table retains different roles. P1 needs to decide what to follow. P3 needs enough confirmation to update the shared plan. Their stories can inform the same product question while still describing different actions and responsibilities.

Try two competing ways to group the material

A feature-oriented grouping can be useful as an index:

Those headings are readable, but “reminders” contains a suggested feature, an existing input, a refusal and a coordinator's own request. A statement may appear under several headings. That makes the grouping an index, not a set of independent votes or a common requirement.

A situation-oriented grouping exposes a different pattern:

For the question “what becomes difficult when a plan changes?”, choose B for the main discussion. Keep A as an index to the feature talk. B brings the action and the situation closer together, but it could still overemphasize confirmation and hide P3's personal reminder request. Preserve that last row rather than discarding it to make one story neater.

These are two editorial arrangements of the same small fictional material. Another question could justify a different arrangement. Neither grouping is a certified method or an automatic discovery result.

Repair a summary that invents agreement

Here is a deliberately flawed summary, also authored for this exercise rather than captured from a model:

All three participants asked for automatic reminders because late alerts made them miss shifts. They all use the same mobile app, so the product should send every change to everyone.

It sounds like a decisive product direction. The excerpts do not support it:

A more useful provisional finding is:

F1 — Recognizing a changed plan. In these fictional excerpts, P1 receives the routine reminder yet still checks with a coordinator when a swap discussion and the sheet give different cues. P3 describes both confirmation gaps and a reportedly smooth sequence that distinguishes pending requests from confirmed changes. P2 reports a stable routine, does not want another reminder for it and has no recent swap example. The material points toward a question about confirmation and update handoffs, rather than a shared request for more reminders. P3's personal follow-through nudge remains a separate idea.

P2 challenges the universal “everyone” finding. Because P2 has no recent swap example, it does not disprove a question specifically about changes. P3's smooth case is a contrasting change case, although we only have the coordinator's report. It suggests asking what made that case clear; it does not prove that a particular sequence caused the outcome.

The revised paragraph is longer because it preserves distinctions that affect the next decision. You can shorten it for a slide while keeping the source-linked table available. Do not remove the counterexamples just to make the headline stronger.

Turn the finding into one product question

A candidate problem to investigate is:

After a shift-swap discussion, a member may be unsure which arrangement to follow when the discussion and shared plan differ.

That is a proposed problem framing from the exercise. It is not a statement about how often real people experience it. The product question is:

How do members and coordinators recognize that a proposed change is confirmed and reflected in the plan they should follow?

Keep an alternative explanation beside it: unclear responsibility or an incomplete handoff may create the conflicting cues even if the existing interface is adequate. P1 does not know who should update the sheet; P3 describes needing confirmation from both people. The excerpts cannot establish whether the main change should be to the tool, the working agreement or something else.

For a next conversation, one concrete question would be:

Think of the most recent shift swap you handled. From the first request
to the point someone acted on it, what messages or sheet entries did you
use, who changed them, and how did you decide whether it was confirmed?

Allow a smooth example as well as a difficult one. If someone has no such recent event, keep that answer; do not supply an imagined story for them. No next interview or product experiment has been carried out in this example.

Reuse a bounded synthesis request

When using an assistant to help organize material you already have, a request can make the desired distinctions explicit:

Use the complete supplied interview excerpts and retain their turn order
and speaker labels. Produce a product-finding discussion, not a coding
brief or a claim that research has already validated a feature.

First extract every turn. Keep interviewer questions and suggestions
separate from participants' reported actions, preferences and uncertainty.
Do not turn the interviewer's automatic-reminder question into a
participant request. Do not add incidents, app names or missing answers.

Compare each participant's situation, behavior, difficulty and existing
response. Show two plausible groupings of the same material, explain
what each reveals or hides, and state which fits the current question.
Keep different meanings of the word reminder distinguishable.

Write a provisional finding with exact source-turn references. Include
P2's stable routine and absent recent swap example, plus P3's reportedly
smooth swap. Preserve the coordinator's own nudge as a separate idea.
Do not treat repeated mentions or multiple turns as independent votes.

Return one candidate product problem, one alternative explanation and
one concrete question for a future conversation. State what the material
cannot establish. Keep interpretations distinguishable from the wording
of the original excerpts; preserve those originals unchanged.

A compact finding card can then hold the proposed interpretation, supporting turns, contrary or limiting turns, an alternative explanation and the next question. The finished artifact is a discussion people can challenge by returning to the excerpts. It does not need to end with a feature decision.