⚠️ Channel adjustments for 2C products
This skill defaults to dev / B2B channels.2C consumer goods / education / apps: The main battleground for customer acquisition is vertical communities + short videos + vertical KOLs, choose the first platform based on publicly available regional data (e.g., short video in Indonesia/Thailand has already overtaken Facebook). For KOLs, prioritize nano / micro verticals—the more followers, the lower the engagement rate; micro > mega offers better cost-effectiveness. For the complete 2C channel database + public sources, see → gingiris-seo-geo/references/2c-adaptation.md。
User Interviews and Cold Start Operations Practical Handbook
Podcast firsthand cases and citation boundaries:references/podcast-evidence.md。
🌍 Language / 语言: Chinese | Interview Guide | Cold-Start Ops
Core Principles
"A product's founder will intensively conduct a large number of user interviews during the first six months. HeyGen was six months, and the founder personally did 937 sessions; Wisperflow and Higgsfield both did more than 500 sessions—just continuously doing interviews and continuously iterating." — 生姜iris
Key Insight: User interviews are not a research tool, but a PMF discovery engine. Interview frequency is positively correlated with product success rate.
Token Top10 Interview Method (2026 signature method)
The shortest path to turn interviews from a "research action" into a "daily operations loop," already reused across the board in multiple AI Agent projects:
- In the backend, pull the Top10 user listby token consumption / usage (refresh weekly)
- Aboutscreen-sharing interviews, must record the screen—seeing real operations > hearing verbal accounts
- Feed transcripts to AI, automatically extract feature requests / bugs directly into Linear (or any issue tool)
- Follow up the next day: tell them "the X you mentioned has been fixed/scheduled"—the feedback loop itself is a retention lever
- Cadence:at least 2 sessions per day; after the interview, cut 5-10 minutes of key clips and send them directly to product and engineering
Supporting actions：
- New interviewers first do mock interviews + full recording + highlight review, then go to real users
- KOL interviews combined into one: while collaborating with YouTubers, also do user interviews + usability testing, getting two kinds of value from one collaboration; 20-30 minute video interviews have the highest ROI
- Turn interviews into assets: one paid user interview can simultaneously produce a three-piece set of feature requirements + SEO material + reseller leads
Invitation outreach cadence (2026 update)
- Four touches on days 1/3/7/14: Day 1 first send → Day 3 → Day 7 → Day 14 each follow up once
- Dual channel: email template + LinkedIn DM via email reverse lookup; if there is no reply to the email, switch to LinkedIn for an extra round
- Use broad wording in the script ("the team is doing user interviews early on"),do not reveal the real screening reason, to avoid users performing with preset expectations
Benchmark anchors
Wisprflow: before launch, manually onboarded the first 500 users and interviewed them one by one via video, using the accumulated momentum to drive launch; HuggingFace: 8 months of research + 500 user interviews before scaling up.
2026-07 addendum (firsthand distillation from podcast)：
- According to a retelling (per the former COO of AFFiNE): Gamma / Notion / HeyGen all conducted about 1,000 user interviews within six months, and "whatever bad points users mentioned, they really went and solved them."
- The common path of successful products like Gamma / HeyGen / Lovable: spend half a year on 500-1,000 interviews to polish product details before you're even qualified to talk about growth—"a huge number of people overlooked the previous half a year to a year of dirty, hard work."
- Counterexample (anonymous project, reported): skipping interview-driven polishing and going straight to burning money on paid acquisition, with a cost per registration as high as dozens of US dollars and a cost per paying user of several hundred US dollars—"no logic, but money."
User Interview Execution Framework (5 Steps)
Step 1: Define Goals and Screen Users
User Priority：
| Priority | User Type | Value |
|---|---|---|
| P0 | Paying Users | Payment willingness already validated, highest value |
| P0 | High-Frequency Active Users | Deep understanding of the product, most effective feedback |
| P1 | Competitor Users | Provide a competitive perspective |
| P1 | Churned Users | Expose the product's real problems |
| P2 | Registered but Non-Paying Users | Insight into conversion obstacles |
Step 2: Invitation and Scheduling
Follow-Up Cadence: Day 1 → Day 3 → Day 7 → Day 10 (all channels: LinkedIn/Email/Telegram/Discord)
Core of the Invitation Script: explain the purpose + promise the duration (30 minutes) + explain screen recording + exchange of value
Step 3: Interview Execution (30-45 minutes)
Must: 1 person moderating + 1 person taking notes + screen recording required (including screen sharing)
Core Question Framework：
- Background Understanding: occupation, acquisition channels, length of use, core features, competitors
- Workflow Exploration: What problem does it solve? How was it done before? What has changed now?
- Competitor Comparison: Which competitors have they used? For which scenarios? What are our strengths and weaknesses?
- Pain Point Mining: What's frustrating? Bugs? Features that are hard to use? The "magic wand" question
- Payment Willingness: What have they paid for before? How much are they willing to pay? What prompts an upgrade?
Key Technique: Observing via screen sharing > user verbal description. Have users demonstrate real operations and observe where they get stuck.
Step 4: Wrap-Up and Follow-Up
- Ambassador Program Invitation
- Referral Request
- Complete the interview debrief template within 24 hours
Don't Stall When Customer Materials Are Missing: Before the meeting, use the official website, public case studies, and existing screenshots to build a rough draft that can be discussed; during the meeting, have the user directly demonstrate the real workflow and request the materials they already have. Record missing materials as items to be verified, and do not make "the customer will go back and organize complete materials" a prerequisite for the next learning session.
Step 5: Data Aggregation (every 5-10 interviews)
Aggregate common needs/pain points → update feature priorities → refine user personas → adjust product direction
How to Tell Effective vs Ineffective Interviews (added 2026-07)
"An ineffective user interview is like giving an investor pitch—you talk 80% of the time, and the user has already drifted off into another world." — 生姜iris
| Dimension | Ineffective Interview | Effective Interview |
|---|---|---|
| Who Is Talking | You talk 80% of the time, and the user has drifted off into another world | Listen to the user talk about how they actually use it |
| What to Look At | Only listening to verbal accounts and watching demo scripts | Have users share their screens to see real operations |
| Scenario | Hypothetical scenarios / talking about the product yourself | Real scenarios, real user cases |
| Mindset | Treating users as pure numbers | Keep your EGO in check and befriend users |
Basis for judgment: The user journey you construct is completely different from the user's actual user journey—only screen sharing can expose this gap.
Dual logic of willingness to pay: "Saying you're willing to pay in an interview and actually paying in the end are two different logics"—once it comes to actually paying, everyone becomes extremely rational. Payment statements in interviews serve only as directional reference; validation is always based on real payment behavior (corresponding to the "four timestamps of payment attribution" below).
Interview-driven iteration loop (practical path, added 2026-07)
- The first interview can run a full 2 hours: Deep users often understand the scenario, industry, and all competitors better than you do, and can help you distill the true value proposition and the biggest comparative advantage at the moment.
- Sincere operations earn deep feedback: An early overseas power user later voluntarily wrote a 2,000-word long review with screenshots and GIFs, pointing out item by item that "too much time was spent on feature scenarios users don't need" plus specific bugs—"if it weren't for how sincere you are, I wouldn't have spent two hours writing this."
- Whole-team breakdown → fix within a deadline: After receiving the long review, the whole team met to break it down item by item, and took 1-2 months to fix all the bugs in it.
- Close the loop with a report back: After fixing/scheduling, follow up with the user that "the X you raised has been fixed"—closing the feedback loop is itself a retention lever (echoing Step 4 of the Token Top 10).
- Pitfall warning: If, after interview feedback is uploaded, the No. 1 position still makes trade-off decisions alone, a large amount of user input will be ignored—product, engineering, and operations must jointly set goals and jointly digest the essence of the interviews.
Benchmark for speed of trust building (self-reported account): No matter how senior the title, stay on the front line chatting directly with users every day; trust built with users in 2 months can exceed that of colleagues who don't get in the field in over 1 year—users will open up voluntarily and vouch for you across departments. High-frequency direct frontline chats are the prerequisite for interview quality.
Beta test design framework
Three-layer structure of test objectives
| Layer | Dimension | Data collection method |
|---|---|---|
| Value validation layer | Whether it solves the core pain point | Interview |
| Experience perception layer | Speed, ease of use, effectiveness | Observation method (screen recording/live call) |
| Gap identification layer | Missing features, experience lag | Structured questionnaire + Bug feedback |
Test task design principles
- Tasks simulate a complete workflow, not isolated feature points
- Task length triggers "waiting" scenarios (exposing performance issues)
- Provide prompt examples to reduce cognitive load
- Clearly state that "Beta instability is normal"
Core of live-call real-time testing
- Emphasize "observation" rather than "testing the user"
- Think Aloud Protocol (speak while doing)
- Use loading wait time to collect user background (an underrated technique)
Five Whys (structured wrap-up)
- What kind of product do you think this is? (Have the user apply a label)
- What is the most satisfying feature?
- Which part is the most sluggish or confusing?
- If it officially launched tomorrow, would you be willing to pay? Why?
- What would you most like to see added in the next version?
Cold-Start User Operations Methodology
Cold Outreach Funnel
触达 → 回复邀请(~47%) → 深度追踪(~47%) → 合格在测(~24%) → 持续活跃
Key: 50 people matching the ICP > 500 people reached broadly
Messaging Iteration Method
| Script Type | Effect |
|---|---|
| Feature-driven (what it does) | Low conversion, attracts the wrong profile |
| Scenario-driven (what problem it solves) | Conversion rate up ~83% |
| Pain-point comparison (how it was done before) | Highest conversion |
Pattern: The more specific the pain-point description, the higher the conversion rate.
Two-Tier Beta User System
| Tier | Profile | Threshold | Benefits |
|---|---|---|---|
| Tier 1 Creative Partners | KOL/industry expert | Continuous use >7 days + high-quality feedback | Pro permissions + direct line to the founder + Case Study |
| Tier 2 Early Beta Users | Professional users matching the ICP | Submit specific improvement suggestions | Early access + roadmap participation |
Activation and Payment Benchmarks (2026 measured)
Interview Recruitment Priority and First-Round Experiment
Priority order: real paying/high-frequency users → activated but not paying → newly registered but not activated → churned users → broad target profile. In the first round, send 10 individually customized messages, record replies, bookings, attendance, and valid insights, then decide whether to scale up. During interviews, prioritize having users share their screens to complete real tasks; opinions must be cross-validated with behavior logs, activation, and payment timestamps.
The definition of activation must be a behavioral threshold, not "registered / logged in":
| Product Type | Behavioral Threshold Example |
|---|---|
| Agent type | Complete ≥2 tasks |
| Conversational type | ≥20 rounds of conversation per day for 3 consecutive days |
| Tool type | Start running a task, or use for more than 5 minutes in a single session |
| DAU definition | Count by unique message send, excluding internal users |
Conversion Qualification Line：
- Registration → Payment >5% is the qualification line: Pro-C 5-8%, SMB 10-15% (large C-type 1.5-3%)
- Stage goals must be "within reach": for a product with a 0.1% payment rate, first set 1%/3 months, do not directly set 5%
Churn Can Be Made Operational: no return visit within 24h after two conversations = churn, automatically triggering a win-back email—once churn is defined behaviorally, win-back can be automated.
Four Timestamps for Payment Attribution: registration time × activation time × payment time × most recent login, with the four points cross-referenced to determine the true payment driver (which feature, which interview, which outreach brought the payment).
User Grading Standards
| Level | Characteristics | Operations Strategy |
|---|---|---|
| S-tier | High frequency + paid + willing to promote | Ambassador priority, direct CEO engagement |
| A-tier | High frequency + paid potential | Beta priority access, regular follow-ups |
| B-tier | Occasional use + no clear payment intent | Guide to core features, drive conversion |
| C-tier | Low activity + no payment intent | Reduce investment, focus on organic retention |
Churn Analysis Framework
| Churn Type | Typical Cause |
|---|---|
| Performance-related | Device overheating/lag, loading timeout, crashes |
| Workflow gap-related | Lack of downstream integration, insufficient feature completeness |
| ICP mismatch-related | No data sovereignty, project-based with no long-term need |
| Messaging mismatch-related | Communicated value proposition ≠ actual product value |
Three Questions for Churned User Interviews：
- When did you start not using it? What were you doing at the time?
- Did you run into a problem, or did you find an alternative?
- If we solved problem X, would you come back?
PMF Value Assessment Matrix
Assess by "content sovereignty × usage frequency":
| User Type | Content Sovereignty | PMF Value | Churn Risk |
|---|---|---|---|
| Independent Creator | Full sovereignty | ⭐⭐⭐⭐⭐ | Low |
| Institutional Decision-Maker | Full sovereignty | ⭐⭐⭐⭐ | Price sensitive |
| Freelancer | Temporary sovereignty | ⭐⭐⭐ | Churns when project ends |
| Institutional Executor | No sovereignty | ⭐⭐ | Data authorization concerns |
Toolchain
| Category | Tool | Purpose |
|---|---|---|
| Survey | Typeform | User research surveys |
| Recording | Grain, Fathom | Interview recording transcription |
| Database | Airtable | User database management |
| Calendar | Calendly | Interview scheduling |
| Outreach | LinkedIn, Discord | User contact and community |
Full Reference Documentation
| Document | Content |
|---|---|
| references/interview-guide.md | Complete user interview guide (HeyGen 937-session methodology, question framework, templates, FAQ) |
| references/cold-start-ops.md | Beta test design framework, cold-start user operations, churn analysis, reusable templates |
| references/templates.md | Interview retrospective template, user persona table, Feature Request Tracker |