| Section | Time | Focus |
|---|---|---|
| Context | 5 min | Role, environment, current workflow. What tools they use today. |
| Problems | 10 min | Frustrations, bottlenecks, workarounds. "Tell me about the last time you..." |
| Behaviors | 10 min | What did they actually do? How often? What triggered the behavior? |
| Impact | 5 min | Time lost, revenue impact, stress level. "What happens if this isn't solved?" |
The Echo Probe
When a user mentions something interesting, echo their exact words back as a question:
User: "The setup was really confusing."
You: "Confusing?" (pause, let them elaborate)
This signals you're listening and invites depth without leading. Never rephrase their words.
The Specificity Ladder
When users speak in generalities, climb down to concrete specifics:
| Level | Example | What to Ask |
|---|---|---|
| General | "We have problems with onboarding" | "What kind of problems?" |
| Specific | "People get stuck in the setup" | "Where exactly do they get stuck?" |
| Concrete | "At step 3, connecting data sources" | "Tell me about the last person who got stuck there" |
| Actionable | "They don't know what an API key is" | (You now have a design problem to solve) |
The "Show Me" Request
Whenever possible, ask users to show you rather than tell you. Observed behavior always overrides self-reported behavior — users rationalize their actions after the fact.
The Timeline Walk
For complex workflows, have the user walk you through chronologically: "Walk me through what happens from the moment you decide to add a new team member to FlowScale, step by step, until they're fully set up." This reveals hidden steps, dependencies, and pain points.
The Dissent Question
At the end, always ask: "Is there anything I should have asked that I didn't?" and "Is there anything you think I'm getting wrong about your experience?"
| Mistake | Why It Happens | Fix |
|---|---|---|
| Leading questions | Wanting confirmation of your hypothesis | Replace with open questions |
| Accepting solutions as requirements | Users are helpful; they want to solve it | Dig into the problem behind the suggestion |
| Talking too much | Nervousness, wanting to explain the product | Target: you talk 20%, they talk 80% |
| Only interviewing happy users | Easier to recruit, less uncomfortable | Actively seek churned and unhappy users |
| Not recording | Thinking you'll remember | Always record (with consent) |
| Interviewing one person and generalizing | One data point feels like enough | Minimum 5 interviews before drawing conclusions |
Customers "hire" products to make progress in specific situations. JTBD shifts your thinking from features to outcomes.
| Bad Framing (Feature-Centric) | Good Framing (JTBD) |
|---|---|
| "Users want dark mode" | "Users working late need reduced eye strain to maintain productivity" |
| "Users want export" | "Ops managers need to share workflow data with leadership who don't use the platform" |
| "Users want AI" | "Ops managers spending 3+ hours/week on manual scheduling need automated workflow triggers" |
When [situation], I need to [motivation], so I can [outcome]
Example: When I hire a new team member, I need to get them set up on FlowScale without spending 45 minutes sitting with them, so I can focus on actual operations work.
Interview 1 — Sarah, Operations Manager (200 employees)
"Our biggest pain is onboarding new team members. Right now I have to sit with each person for 45 minutes
walking them through the setup. We hire 5-10 people a month, so that's 5-8 hours a month just on onboarding
walks. When someone gets stuck, they message me on Slack and I have to stop what I'm doing to help them."
Interview 2 — Marcus, VP Operations (500 employees)
"The problem isn't the tool itself — it's getting our team to actually use it properly. We pay for 200 seats
but only 60 are active. I think people create their account, get confused by the setup, and then go back to
spreadsheets. I've asked for better onboarding materials three times now."
Interview 3 — Lisa, Operations Coordinator (80 employees)
"The first time I used it, I got stuck on step 3 where it asks me to connect my data sources. I'm not
technical — I don't know what an API key is. I had to ask our IT person to help. That took two days to
schedule. Once we got past that, it was fine. But I almost gave up."
Interview 4 — James, Operations Director (350 employees)
"We need better permissions. Right now everyone can see everything. We have contractors who shouldn't see
financial workflows, and junior staff who shouldn't be able to delete things. We've had two incidents where
someone accidentally modified a production workflow. If you had role-based permissions, we'd expand immediately."
Interview 5 — Priya, Senior Ops Manager (150 employees)
"The templates are useless. They're too generic. I spent last weekend building a custom onboarding flow from
scratch — took me 6 hours. A template that was even 50% right would have saved me a full day. Also, there's
no way to share workflows across our team — everyone builds their own version."Create prompts/discovery-synthesis.md:
# Discovery Synthesis Prompt
You are a senior product discovery analyst at FlowScale.
Load context from context/company.md before responding.
## Task
Analyze the provided interview transcripts and produce:
1. **Pain Point Clusters**: Theme name, affected users, frequency, severity (1-5), evidence (direct quotes)
2. **Jobs-to-be-Done**: Job statement (When/I need/so I can), which users, current solution, satisfaction score
3. **Opportunity Map**: Ranked by user impact, strategic alignment, and current solution gap
## Quality Rules
- Every claim must be backed by a direct quote
- Do not invent insights not present in the data
- Separate observed behavior from stated preference
- Flag contradictory evidence between interviewsRun the synthesis and evaluate the output quality:
Create deliverables/W02-discovery-report.md with validated pain point clusters, JTBD statements, prioritised opportunity map, and a "gap note" for any insight you added that wasn't in the interviews.
Find a colleague or friend who uses any software tool for work. Conduct a 15-minute interview using the Lesson 2.2 framework. Then run the same synthesis prompt on your own interview notes. Compare: How does the AI handle your raw, imperfect notes vs. the polished transcripts?
prompts/discovery-synthesis.md)deliverables/W02-discovery-report.md)