v1-learning-from-customers - SKILL.md Agent Skill

name: v1-learning-from-customers description: Use when planning customer discovery, auditing customer interview questions, synthesizing customer conversation notes, checking whether demand evidence is real, or applying Jobs-to-Be-Done discovery. Triggers on "customer discovery", "customer interviews", "talk to customers", "mom test", "is this real demand", "jobs to be done". allowed-tools: - Bash - Read - Grep - Glob - WebFetch - WebSearch

Learning From Customers

Plan, audit, and synthesize customer conversations so product decisions rest on behavior, context, and commitments instead of compliments, opinions, or imagined futures.

Default mode is advisory. Do not contact customers, send messages, schedule meetings, or mutate CRM/research systems unless the user explicitly asks.

Quick Start

For a product idea, discovery plan, interview guide, meeting notes, transcript, or pile of customer quotes:

Identify the product decision the research must inform.
Define the customer slice narrowly enough that patterns can repeat.
Write or audit the three most important questions for that slice.
Convert bad questions into past-specific questions about real behavior, current workarounds, costs, constraints, and buying process.
Classify evidence as fact, commitment, workflow, pain, budget/process signal, idea, compliment, or fluff.
Decide the next action: keep learning, narrow the segment, ask for advancement, prototype, sell, or stop.

Input Modes

Choose one mode from the user's request:

Mode	Use When	Output
Plan conversations	The user has an idea, segment, or open research question	Customer-learning plan and conversation guide
Audit questions	The user has drafted questions or a survey/interview script	Bad-question findings and rewritten questions
Synthesize notes	The user has notes, transcripts, call summaries, or quotes	Evidence synthesis and next decision
Assess demand	The user asks whether evidence proves demand	Demand-quality review and gaps
Create job spec	The user needs Jobs-to-Be-Done framing	Job spec with progress, circumstance, competitors, anxieties, and success signals

If the request lacks the artifact needed for the selected mode, state the missing artifact and provide the smallest useful next step.

Workflow

1. Identify The Decision

Start with the decision the conversations should change:

Is this problem worth solving?
Which customer segment has the strongest pull?
What current workaround, cost, or frustration exists?
What must a new product replace, reduce, or beat?
Is the team ready to prototype, sell, or stop?

Do not run generic discovery. Name the decision before writing questions or interpreting notes.

2. Slice The Customer

Define a narrow customer slice:

Field	Questions
Role	Who experiences the problem directly?
Trigger	What event makes the problem show up?
Current behavior	What do they do today?
Context	Device, workflow, organization, budget, permissions, or environment
Stakes	What happens if the problem remains unsolved?
Access	How can conversations with this slice be found?

If evidence is inconsistent, first check whether the team is mixing several customer slices. One conversation each with many different customer types is not the same as many conversations with one type.

3. Write The Three Big Questions

Choose the three highest-value unknowns for this customer slice.

Good big questions usually concern:

The last time the problem happened.
The current workaround and its cost.
Who else is involved in the workflow or purchase.
What has already been tried and abandoned.
What would need to be fired before a new solution could be hired.
The functional, emotional, and social progress the customer is trying to make.

Do not ask every participant the same script if reliable evidence already exists. Keep filling the picture from the highest-uncertainty edges.

4. Audit Questions

Flag and rewrite bad questions before the conversation.

Bad Question Type	Why It Fails	Rewrite Pattern
Opinion	Produces politeness or taste, not evidence	Ask about the last real instance
Hypothetical	Future behavior is over-optimistic	Ask what happened last time
Leading	Reveals the desired answer	Remove the pitch and ask about current life
Feature request	Turns customers into product designers	Ask what problem the idea would solve
Price fantasy	Treats invented willingness as budget evidence	Ask about existing spend, approval path, or replacement cost
Generic habit	Produces "usually" and "always"	Ask for a concrete walkthrough

Keep the conversation about the customer's life and workflow until the learning decision requires showing the idea or product.

5. Run The Conversation

Use a light guide rather than a rigid script:

Set context briefly without pitching the solution.
Ask about the customer's current workflow, last occurrence, and workaround.
Follow concrete nouns, costs, delays, approvals, and repeated frustrations.
Deflect compliments by asking for specifics behind them.
Anchor fluff by asking when it last happened and what they did.
Dig beneath ideas by asking what problem they would solve and how the customer handles it now.
End by asking for the next useful person, artifact, or concrete advancement.

Talk less as the conversation improves. The more the agent or interviewer explains, the weaker the evidence becomes.

6. Capture Evidence

Separate what was observed from what was inferred.

Evidence Type	Counts As	Does Not Count As
Fact	Past behavior, current workflow, actual spend, named tool, concrete event	General preference
Pain	Repeated cost, delay, risk, embarrassment, lost money, wasted time	Mild annoyance without action
Workaround	Spreadsheet, manual process, hired service, internal tool, habit	"We should probably fix this"
Commitment	Time, money, intro, pilot, data, procurement step, scheduled next meeting	Compliment or vague enthusiasm
Job signal	Desired progress in a specific circumstance with functional, emotional, and social dimensions	Demographic label alone
Bad data	Compliment, future promise, generic claim, unsupported idea	Evidence strong enough to change a decision

When evidence is thin, say so. Do not inflate friendly conversations into validation.

7. Build A Job Spec When Useful

Use Jobs-to-Be-Done framing when the product decision depends on what customers are trying to accomplish rather than what feature they requested.

## Job Spec

Customer slice:
Triggering circumstance:
Desired progress:
Functional dimensions:
Emotional dimensions:
Social dimensions:
Current competitors / workarounds:
What must be fired:
Obstacles and anxieties:
Tradeoffs accepted:
Purchase experience needed:
Use experience needed:
Little Hire signals:
Success metrics:

Treat competitors broadly: products, services, spreadsheets, people, doing nothing, delaying, or cobbled-together workflows can all compete for the job.

8. Ask For Commitment Or Advancement

Once enough learning exists to show an idea, prototype, or offer, look for advancement:

Schedule a concrete next meeting with the right person.
Get an intro to a buyer, user, expert, or adjacent segment.
Ask for time, data, pilot access, procurement steps, budget owner, or payment.
Ask what would block adoption and who would need to approve.

If the customer remains friendly but never advances, classify the lead as unvalidated. A polite ongoing conversation is not demand.

9. Synthesize And Decide

End with a decision, not a pile of quotes.

Decision options:

Keep learning: the segment is plausible, but evidence is incomplete.
Narrow the slice: signals conflict because the customer group is too broad.
Prototype: the problem and workflow are real, but the solution is untested.
Sell or pilot: commitments show urgency and adoption path.
Stop: no meaningful pain, no workaround, no budget/process, or no repeated pattern.

State what evidence would change the recommendation.

Output Formats

Conversation Plan

## Customer Learning Plan

Decision:
Customer slice:
Top 3 questions:

## Conversation Guide

| Question | What It Tests | Bad Data To Watch For | Follow-Up |
| --- | --- | --- | --- |

## Evidence To Capture

## Advancement Ask

## Stop / Continue Gate

Question Audit

## Question Audit

| Question | Finding | Risk | Rewrite |
| --- | --- | --- | --- |

## Missing Questions

## Best 3 Questions

Notes Synthesis

## Customer Evidence Synthesis

Decision:
Customer slice:
Verdict:

## Evidence

| Claim | Evidence | Type | Strength | Notes |
| --- | --- | --- | --- | --- |

## Bad Data

## Job Spec

## Next Action

Evidence strength:

Strong: repeated behavior, costly workaround, real spend, or concrete commitment.
Medium: credible past behavior from a fitting customer slice.
Weak: opinion, unverified claim, noncommittal pain, or single anecdote.
Bad data: compliment, hypothetical, generic future promise, or unsupported idea.

Anti-Patterns

Asking whether the idea is good.
Asking what customers would do in the future.
Treating compliments as validation.
Letting feature requests become requirements without finding the underlying problem.
Interviewing a broad market and expecting coherent patterns.
Pitching before learning the current workflow.
Keeping customer learning in one person's head.
Reporting quotes without separating observation, inference, and team opinion.

Chaining

Use v1-interview-me first when the user needs to clarify the product idea or learning decision.
Use v1-testing-prototypes after a prototype exists and the team needs observed task/value evidence.
Use v1-prd after customer evidence is strong enough to write build requirements.
Use v1-strategy-review when customer learning changes the wedge, scope, or product direction.

When Not To Use This Skill

Do not use for general feature specification once demand is already clear; use v1-prd.
Do not use for live prototype-test planning or synthesis; use v1-testing-prototypes.
Do not use for broad strategy critique unless the immediate question is customer evidence quality; use v1-strategy-review.
Do not use for operational bottleneck or throughput diagnosis; use v1-diagnosing-constraints.

Examples

Bad question rewrite

Input: "Would you pay for a dashboard that automates this?"

Output: Ask when the workflow last happened, what they used, how long it took, who cared about the output, and what budget or approval path existing tools go through.

Weak evidence

Input: "Everyone loved the idea and said they would use it."

Output: Classify as bad data unless notes include past behavior, current workaround, cost, budget owner, or concrete next step.

Segment problem

Input: Ten conversations produce ten unrelated requested features.

Output: Check whether the team is mixing several customer slices. Narrow by trigger, role, workflow, or current workaround before drawing product conclusions.