From Interview Insights to Tickets: How to Stop Insight Decay
Most interview insights die between the call and the backlog. A seven-step flow, an evidence threshold and a ticket template that keeps the customer's words attached.
Share
To turn interview insights into tickets without losing them, synthesise each interview within a day into evidence rows with source timestamps, group those rows into opportunities across interviews, and create a ticket only when an opportunity passes an evidence threshold you set in advance. Every ticket then carries the problem, the user story, acceptance criteria and links back to the exact moments customers described it.
Key takeaways
Insight decay is the loss of research value between the call and the backlog: notes go unread, context fades, and the ticket that finally gets written has no customer in it.
Speed matters. Teresa Torres fills in a one-page interview snapshot straight after each call, in a 45-minute slot of 20-30 minutes interviewing and 15-25 minutes writing (Product Talk, 2024).
Do not create one ticket per quote. Aggregate evidence first, then ticket the opportunities that cross a threshold, such as three independent interviewees in the target segment.
AI can do the first pass far faster but not as reliably: in one study, generative AI took 20 minutes against 567 for human coders, yet coding agreement was only 36-47% (Prescott et al., 2024).
Tools such as Linear and Jira Product Discovery can attach customer evidence to issues, but only if someone captures it in a linkable form.
What is insight decay?
Insight decay is what happens to customer research between the end of an interview and the moment an engineer picks up a ticket. The recording sits in a video tool. The notes live in one person's document. The synthesis, if it happens, becomes a slide deck. By the time a ticket is written, it says "add bulk export" and nothing about the customer who spent two hours copying rows by hand.
Three forces drive the decay:
Context fades. Torres notes that customer understanding goes stale quickly, which is why she recommends interviewing at least weekly and synthesising each conversation while it is fresh (Product Talk).
Evidence gets buried. Atlassian's State of Teams 2025 survey of 12,000 knowledge workers found teams spend over 25% of their working week searching for information (Atlassian, 2025). Unstructured interview notes are among the hardest things to find.
Handoffs strip the "why". Each copy, from transcript to notes to deck to ticket, keeps the conclusion and drops the evidence.
Without evidence on the ticket, engineers cannot make good trade-offs, and nobody can later check whether the shipped work solved the customer's problem. (What happens before the interview is covered in the product discovery process guide.)
Where do interview insights get lost?
Handoff
What usually happens
What to do instead
Call to notes
Notes written days later from memory
Snapshot within 24 hours, with timestamps
Notes to synthesis
Synthesis skipped or done once a quarter
Evidence rows after each call; patterns every 3-4 interviews
Synthesis to backlog
A deck nobody links to
Opportunities with linked evidence rows
Backlog to ticket
"Add feature X" with no customer context
Ticket template with problem, evidence links and acceptance criteria
Ticket to release
Nobody checks the original problem
Success measure on the ticket; follow up with the interviewees
Fix the first handoff first. GOV.UK's service manual says to analyse a session as soon as possible, budgeting about 1 hour of analysis for every 2 hours of research (GOV.UK).
What should an evidence-backed ticket contain?
A ticket is actionable when a developer can pick it up without re-watching a 45-minute recording, and can still check the customer's words if a trade-off comes up. Keep the four parts from the original version of this post (pain point, user story, evidence links and priority) and add what the backlog usually lacks: acceptance criteria and open questions.
The user story format, "As a [type of user], I want [goal] so that [reason]", is a thinking tool rather than a guarantee, as Mountain Goat Software puts it, and it works best with Ron Jeffries's three parts: card, conversation, confirmation (Mountain Goat Software). Bill Wake's INVEST checklist from 2003 still applies: a story should be Independent, Negotiable, Valuable, Estimable, Small and Testable (Wake, 2003).
Copyable ticket template
Title: [verb + outcome, not a feature name]
Problem (customer's situation, in their terms):
User story: As a [segment/role], I want [goal] so that [reason].
Evidence
- [Interview ID, date, timestamp] observed: "short verbatim quote"
- [Interview ID, date, timestamp] observed: what they did
- [Support ticket / analytics link] supporting signal
Segment(s) affected:
Evidence count: [n] independent interviewees, [n] other signals
Contradicting evidence:
Acceptance criteria
- Given [context], when [action], then [observable result]
- ...
Non-goals:
Open questions:
Success measure (and when we will check):
Priority and why:
Linked opportunity / decision record:
Two rules keep it honest. Quotes stay verbatim, short and timestamped, so anyone can check them. And the evidence count is of independent people: one customer saying something four times is one data point.
How do you get from interview to ticket, step by step?
1. Capture within 24 hours
Right after the call, write a snapshot: who the person is, the story they told, a memorable quote, and the opportunities you heard (Product Talk, 2024). Torres reports it takes 15-20 minutes once practised. A customer interview template and well-chosen questions keep snapshots comparable.
2. Extract evidence rows
Turn the transcript into one row per meaningful moment: situation, what they did or said, workaround, consequence, and a timestamp. Keep observation and interpretation in separate columns. The interview synthesis template has the full row format.
3. Attach rows to opportunities
Map each row to an existing opportunity or create a new one. An opportunity is a customer need or pain point, not a feature. After every three to four interviews, review the groupings, as Torres recommends for opportunity solution trees.
4. Apply the evidence threshold
An opportunity becomes ticket-worthy only when it passes the threshold you agreed in advance (see the next section). Below the threshold it stays in discovery, visible and growing, not forgotten.
5. Draft the ticket from the evidence
Write the problem and story from the linked rows, then the acceptance criteria. AI can draft this step well if it is given the rows, not the raw transcript, and told to cite each claim.
6. Review with engineering
Walk through the evidence with the engineer who will build it. This is the "conversation" in card, conversation, confirmation. Expect the scope to change; that is the point.
7. Close the loop
When the change ships, tell the interviewees and ask whether it solved the problem they described. Their answer is the first test of whether the insight was right. Microsoft's experimentation team found that only about one-third of well-designed experiments improved the metric they targeted (Kohavi, Crook & Longbotham, 2009), so a well-evidenced ticket can still miss.
What is an evidence threshold, and how do you set one?
An evidence threshold, the "evidence density" idea from the original post, is the minimum amount of independent evidence an opportunity needs before it reaches the delivery backlog. It stops two failure modes at once: the loudest recent customer jumping the queue, and real patterns sitting unticketed because nobody counted them.
Set it per segment and write it down. Starting points, not benchmarks:
Situation
Example threshold
Small UX fix, easily reversible
2 independent interviewees, or 1 interviewee plus support tickets
New capability for an existing segment
3 independent interviewees in that segment plus 1 behavioural signal (usage data, tickets, churn reasons)
New segment or pricing change
Interviews plus a quantitative check, such as a survey or a fake-door test
Contract-critical request from one account
Allowed below threshold, but labelled as a single-account exception
The threshold is about confidence that a problem is real, not about how common it is. Interviews show what happens and why; they do not measure prevalence. When the decision depends on frequency, check analytics or run a survey before building.
Can AI automate the flow from interviews to tickets?
Partly, and the evidence says where to draw the line.
User Interviews' 2025 survey of 485 researchers found 80% now use AI in their work, but 91% worry about output accuracy and hallucinations (User Interviews, 2025). A peer-reviewed comparison shows why both numbers make sense. Prescott and colleagues gave the same 40 text messages to human coders and to ChatGPT and Bard. The AI matched 71% (5 of 7) of the human inductive themes and finished in about 20 minutes against 567 minutes for humans. But agreement on how individual items were coded was only 36-47%, and the authors concluded hybrid human-AI approaches are necessary (Prescott et al., JMIR AI, 2024).
Step
Automate?
Why
Transcription and timestamps
Yes
Mechanical, easy to check
Extracting candidate evidence rows
Yes, with review
Fast first pass; check each row against the transcript
Grouping rows into opportunities
Draft only
Themes broadly match; item-level coding often does not
Applying the evidence threshold
Yes
It is a count, once rows are verified
Drafting the ticket from linked rows
Yes, with review
Good at structure; must cite every claim
Deciding priority
No
A judgement about strategy and trade-offs
Automation also does not remove bias by itself. A model counts what it is given; if you interviewed only enthusiastic early adopters, the clusters will reflect them. For prompts and a verification table, see AI customer research synthesis.
Which tools link customer evidence to tickets?
Several issue trackers now support evidence on the ticket, which is the part of the pipeline that matters most:
Linear Customer Requests link requests from Intercom, Zendesk, Front, Salesforce and Slack to issues and projects. Customer attributes such as revenue and tier sync in real time from Intercom and every 12 hours from other sources, and views can be ordered by customer count (Linear docs).
Jira Product Discovery insights attach evidence to ideas with a source link, a rating and labels, and can roll ratings up into a prioritisation field (Atlassian docs).
Both suit support and sales feedback. Interviews are harder, because the evidence is a moment inside a recording; the link must go to that moment, not just the file.
A worked example
The team and numbers are illustrative.
A four-person team runs 12 customer interviews over three weeks about reporting. Snapshots are done within a day for 11 of them. Extraction produces 47 evidence rows, which map to 9 opportunities. Their threshold for a new capability is three independent interviewees in the target segment plus one behavioural signal.
Two opportunities pass. "Finance leads rebuild the monthly report by hand because exports drop custom fields" has 5 interviewees and 23 related support tickets. "Managers cannot see which reports are stale" has 3 interviewees and a usage signal: 60% of saved reports had not been opened in 30 days. Each becomes a ticket with verbatim quotes, timestamps and acceptance criteria. A third opportunity, with 2 interviewees, stays in discovery with a note on what evidence would promote it.
Six weeks after release, 4 of the 5 finance leads confirm the manual rebuild is gone; the fifth uses a field the fix missed, which becomes the next ticket.
Common mistakes
One ticket per quote. The backlog fills with near-duplicates and no ticket has enough evidence to prioritise.
Dropping the source. A quote without a timestamp cannot be checked, so it cannot settle an argument.
Ticketing solutions customers proposed. "Add a dashboard" is an idea; the evidence is what they were trying to do.
Accepting AI clusters unchecked. Verify any theme that would change the roadmap against the transcript.
Never closing the loop. If nobody asks interviewees whether the change worked, the team never learns how good its synthesis was.
Where Specky fits
Specky is built for this flow. It reads customer calls, feedback, Slack and Jira into a Product Graph, so evidence from interviews and support sits in one place. It drafts specs and tickets with the customer evidence attached as inline citations, and the drafts land in the PM Inbox for human review before anything reaches the backlog. Alex, Specky's AI research agent, can run interviews from a shareable link, which helps keep a weekly cadence. More on the mechanics: one-click ticket generation.
Frequently asked questions
How do you turn customer interview notes into user stories?
First turn the notes into evidence rows: situation, what the person did or said, workaround and consequence, each with a timestamp. Group rows from several interviews into an opportunity. Then write the story from the opportunity, not from one quote: the segment becomes the "as a", the goal the "I want", and the consequence the "so that". Add acceptance criteria and links to the rows.
How soon after an interview should you synthesise it?
Within 24 hours, ideally straight after the call. Teresa Torres books a 45-minute slot with 15-25 minutes reserved for the snapshot, and GOV.UK advises analysing as soon as possible while it is fresh. Cross-interview synthesis can wait until you have three to four interviews, but the single-interview record should not.
Should every customer request become a ticket?
No. Most requests should become evidence attached to an opportunity, not tickets. Ticket an opportunity once it passes an evidence threshold you set in advance, such as three independent interviewees in the target segment plus a behavioural signal. Below that, keep it visible in discovery. Contract-critical single-account requests can be exceptions if they are labelled as such.
How do you link customer feedback to Jira or Linear issues?
Use the built-in features where they exist: Linear's Customer Requests links feedback from tools such as Intercom, Zendesk and Slack to issues, and Jira Product Discovery lets you attach insights with a source link and rating to ideas. For interviews, link to the timestamp in the recording or transcript, not just the file, and quote the key moment verbatim.
Can AI write tickets from interview transcripts?
AI can draft them, but a person should check them. Give the model verified evidence rows rather than raw transcripts and require a citation for every claim. Research comparing AI and human coders found broad theme agreement but only 36-47% agreement on item-level coding, so review each cited quote and every priority call yourself.
GDPR-aligned. EU workspace data is stored in Frankfurt, Germany. GDPR and EU data Built in Vienna, Austria. A product of influence solutions fm GmbH (FN 675374y). Imprint 14-day trial. Card required, and nothing is charged until day 14. 30-day refund after subscribing. See pricing