"The answers are not in the building. They never were." — Alex's journal, Week 3
Alex had 19 email addresses from her smoke test. She opened them in a spreadsheet. Then she stared at the spreadsheet for two days and did nothing.
The fear was specific: these were real people who had expressed real interest. What if she talked to them and they told her the idea was terrible? Or worse, what if they were polite about it? She'd read enough startup books to know that politeness was the enemy of useful feedback.
On day three she wrote a message and posted it on Twitter: "Hi! I'm a researcher building a tool for managing research papers. Would you have 20 minutes to tell me about your workflow? No pitch, just questions. ☕ on me (virtual)." She sent it as a reply to 30 PhD students she found by searching "literature review" and "PhD" in the past week's posts. Then she closed her laptop and went for a walk.
When she came back four hours later: 7 replies. Four agreed to a call. She formatted a question guide from her notebook, scheduled all four interviews for the same week, and told her roommate: "I think I'm about to find out if I'm delusional."
Most customer conversations fail not because founders ask too many questions, but because they ask the wrong kinds. Questions designed to comfort you produce comfortable lies. Questions designed to reveal reality produce uncomfortable truth, which is the only kind worth having.
Three principles every good discovery interview is built on:
| Type | Question | Why it fails or works |
|---|---|---|
| ❌ Bad | "Would you use an app that helps manage research papers?" | Hypothetical and leading, they'll say yes to be polite |
| ❌ Bad | "Don't you find it frustrating to manage citations?" | Leading question primes a "yes" because you've told them the answer |
| ✅ Good | "Walk me through how you managed your last literature review." | Past behavior, open-ended, reveals the real process |
| ✅ Good | "When did you last lose track of a paper you'd already read?" | Specific past event with emotional context |
| ✅ Good | "What did you do when that happened?" | Reveals workarounds, the strongest validation signal |
Interview number seven. Marcus, a third-year materials science PhD student at MIT with the specific tired eyes of someone who had been awake since 5am doing something that wasn't fun. Alex had gotten better at staying quiet. She asked: "Walk me through your last week of research."
Marcus laughed nervously. He described downloading 40 papers over the weekend, reading abstracts in Preview, maintaining 14 browser tabs "permanently," and keeping a running Google Doc called "notes_FINAL_v3_REAL.docx." He said the document was 47 pages long and he avoided opening it because it made him anxious.
When Alex asked "What happens when you realize two papers you read months apart are actually about the same thing?" he paused for a long beat. "Honestly? I usually lose a day. I have to go back, re-read both papers, rebuild the mental model. It's like..." He searched for the word. "It's like my brain's search function is broken."
Alex wrote in the margin of her notepad in capital letters: LOST A DAY. THIS IS THE JOB. She asked three more questions. She thanked him. She hung up and immediately read her notes back. She'd now heard variations of "lost time re-establishing context" from six of seven people. She texted her friend: "I think this is actually real."
The structure Alex used for all 20 of her discovery interviews. The key is moving from broad context to specific pain, and never revealing your solution until you're completely done listening.
Fifteen interviews done. Alex bought a pack of index cards from a nearby pharmacy and wrote one insight per card. She covered her kitchen table completely, then moved to the floor. 47 cards. She put on a playlist she used for coding and started grouping them intuitively.
Three clusters emerged immediately: "lost time reconnecting context" (11 cards), "too many tools, no single source of truth" (9 cards), "citation chaos during writing" (8 cards). A smaller cluster of 5 cards said some version of "actually pretty happy with Zotero," an important signal that not everyone felt the pain equally.
She stepped back, took a photo, and looked at it. The dominant cluster was unmistakable. People weren't losing time because they had too many papers. They were losing time because their context didn't persist, the understanding they'd built about a paper evaporated between sessions, and they had to rebuild it from scratch.
That was the job. Not "paper organization." Context restoration. She rewrote her problem hypothesis on a new card, pinned it above her desk, and called her mentor. "I finally know what I'm actually building."
Not all customers are equal. Early adopters are the narrow group most likely to try a new, imperfect product, give honest feedback, and tell others about it. They're not your average customer. They're your best customers, found early.
The five criteria for identifying true early adopters:
The JTBD "switch interview" focuses on a specific moment: when someone last changed how they solved a problem. That moment reveals the emotional triggers, context, and switching costs you'll need to understand and overcome.
"When did you first realize you needed a new solution?"
Why it mattersIdentifies the triggering event and the emotional context that drives change. The trigger is usually a moment of failure or frustration, the thing that finally broke the camel's back.
"What were you doing right before you decided to search for something new?"
Why it mattersReveals the specific situation, not a general attitude. Context gives you the precise moment to target with messaging. "When you've just lost 3 hours re-reading a paper" is more powerful than "when you're frustrated."
"What almost stopped you from making a change?"
Why it mattersUncovers the anxieties and switching costs you'll need to overcome. These are the objections that live in your customers' heads before they even talk to you. Address them proactively in your messaging.
Be ruthlessly specific. "Small businesses" is not a profile. "Ops managers at 10–50 person SaaS companies who manually create weekly reports in Google Sheets" is a profile. Use the checklist below.
Get out of the building. No amount of desk research replaces 20 direct conversations with people in your target segment. The answers live in their workflows, not your spreadsheet. Schedule one interview today.
Listen for workarounds. When someone builds a workaround, such as a spreadsheet, a naming convention ritual, or a custom script, they're investing their own time to solve the problem. That's one of the strongest validation signals you can find.
Affinity mapping reveals truth. After 15+ interviews, write insights on cards and cluster them. The biggest cluster is your product's core job. Trust the data over your instincts. The market is smarter than any hypothesis.