Frame an AI Support Copilot Problem

Turn fictional support interviews into an evidence-backed discovery brief.

Introduction

30 Second Summary

A confident feature request can send a team straight into building. The real customer problem can stay hidden until the product misses the mark.

In this project, you will investigate a fictional request for AI-generated replies in business support software. You will practice problem-first product discovery by turning six interview excerpts into an evidence-backed brief plus a 60-second stakeholder pitch.

What You'll Build

You'll present a concise discovery brief that lets a stakeholder trace the recommended opportunity from interview evidence to the next research action.

By the end of this project, you'll have:

An evidence trail linking every finding to a fictional interview ID. A contradiction table shows exactly where your first hypothesis fails.
A ranked opportunity set with traceable scores. A solution-free user-need statement captures the strongest opportunity.
A portfolio-ready discovery brief that connects your recommendation to evidence. A 60-second stakeholder pitch explains the uncertainty before naming the next research action.
Secret Mission: Add the customer as a secondary user while keeping the support agent's primary problem in focus.

Are there any prerequisites?

No prior product management experience is required. Your existing paid Claude subscription is the only requirement.

Before We Start

This is your moment to commit to the problem you are investigating before the fictional interview evidence pulls you toward a solution. You are building an evidence-backed discovery brief because problem-first product discovery prevents the team from committing to AI-generated replies before the underlying support problem is clear.

Set Up Claude as a Neutral Recorder

Problem-first product discovery depends on a clear evidence trail. If Claude makes the product decisions for you, that trail no longer represents your reasoning.

In this step, you will set up a fresh Claude conversation as a neutral recorder. Its role is to preserve your material while leaving every product decision with you.

In this step, get ready to:

Open a fresh Claude conversation.
Configure Claude to preserve your wording without performing the analysis.
Verify that the recorder returns only the exact readiness message.

Open a fresh conversation

A fresh conversation keeps unrelated chats out of your discovery brief. It gives the recorder one clear role from the first message.

Visit claude.ai in your browser.
Complete your usual sign-in flow if Claude asks you to authenticate.
Select the control that starts a fresh conversation.

You should see a blank chat with no earlier messages. This confirms that your recorder starts without unrelated context.

Configure Claude as a recorder

A neutral recorder captures your work without improving or interpreting it. The setup prompt defines that boundary before any fictional interview evidence enters the conversation.

Paste the following setup prompt into the empty message field:

You are the neutral document recorder for a product discovery exercise.

Your role:
- Preserve my wording and decisions.
- Format only the material I provide.
- Do not analyze, infer, prioritize, rewrite, recommend, or invent evidence.
- If a field is missing, write [Not decided].
- Keep every interview reference exactly as I write it, such as I1 or I4.

Create and maintain this Markdown structure throughout our conversation:

# AI Support Discovery Brief
## Discovery goal
## Interview evidence
## Provisional hypothesis
## Contradictions and new insights
## Opportunity ranking
## Primary user need
## Recommendation
## Assumptions and unknowns
## Next research action
## 60-second pitch

Reply only: Ready for interview evidence.

How Does the Recorder Boundary Work?

The recorder role keeps authorship of every product decision with you.
The forbidden actions stop Claude from turning your evidence into its own recommendation.
The literal value [Not decided] makes missing decisions visible.
The Markdown headings create the complete structure that your later work fills.
Exact interview IDs keep each later claim traceable to its source.

Before you submit the prompt, which outcome do you predict: analysis or a wait for evidence?

Submit the message using the chat's submit button.

Claude should return only the exact readiness message: Ready for interview evidence.

Check the recorder boundary

The exact reply acts as a small contract for the rest of the project. Any added analysis shows that Claude has stepped outside the recorder role.

Compare Claude's reply with Ready for interview evidence.
Choose the tab that matches the reply you received.

✔️ I see only the readiness message

Claude returned the exact readiness response: Ready for interview evidence.

You have set the guardrail that keeps every later product decision yours.

ⓧ Claude added analysis

Extra analysis means the recorder boundary was not followed. Use the correction from the exercise to restore the role.

Paste this correction into the same conversation:

Formatting only. Remove all inferences and restore the exact recorder role. Reply only: Ready for interview evidence.

What Does This Correction Do?

This correction removes the added inferences from Claude's role. It restates the only response that belongs at this point.

Submit the correction message.

You should now see only Ready for interview evidence.

Still Seeing Extra Analysis?

Start another fresh conversation.
Paste the original setup prompt again.
Submit the original setup prompt without adding other instructions.

If Claude still adds analysis, ask for help tightening the recorder boundary.

✔️ Awesome, I've got everything!

Your fresh conversation now contains the blank discovery brief structure. Claude's reply is exactly Ready for interview evidence.

ⓧ I'd like to double check the full code

Compare your complete setup prompt with this reference:

You are the neutral document recorder for a product discovery exercise.

Your role:
- Preserve my wording and decisions.
- Format only the material I provide.
- Do not analyze, infer, prioritize, rewrite, recommend, or invent evidence.
- If a field is missing, write [Not decided].
- Keep every interview reference exactly as I write it, such as I1 or I4.

Create and maintain this Markdown structure throughout our conversation:

# AI Support Discovery Brief
## Discovery goal
## Interview evidence
## Provisional hypothesis
## Contradictions and new insights
## Opportunity ranking
## Primary user need
## Recommendation
## Assumptions and unknowns
## Next research action
## 60-second pitch

Reply only: Ready for interview evidence.

What Should This Create?

The prompt gives Claude a formatting role with strict limits. It also creates the complete blank heading structure that your later evidence fills.

Your conversation should contain the blank AI Support Discovery Brief structure shown in the prompt. Claude's reply should be exactly Ready for interview evidence.

Your neutral recorder is ready to preserve the evidence trail without taking over the discovery work. Next, you will form a fast hypothesis from the first two fictional interviews.

Form Your First Hypothesis

You have the boundary in place: your neutral Claude recorder is ready. Your reasoning stays yours while Claude handles the formatting.

Product teams often anchor on the first compelling pain point. You will deliberately capture a fast, fragile provisional hypothesis from the first two interviews.

In this step, get ready to:

Review the fictional case with Interviews I1 and I2.
Draft a five-part first pass from the early evidence.
Add your exact wording to the discovery brief.

Review the fictional case and early evidence

Product discovery starts by connecting a user problem to evidence. This simulation asks how a B2B support-software company could improve correct first-response resolutions without adding agent effort.

Review the fictional case file below.

Fictional Case File

The company has asked its product team to explore AI-generated replies. The discovery goal is to improve the number of support tickets resolved correctly on the first response without adding agent effort.

I1, Maya, Tier 1 support agent: "On password resets and billing-address changes, I mostly repeat the same steps. I spend more time assembling the response than deciding what to do. A starting draft would help on these simple tickets."

I2, Jonah, support team lead: "Monday backlogs are rough, and response time gets attention. But agents slow down because refund rules differ by region. They often ask in chat before replying."

Draft your first pass manually

A provisional hypothesis records what the current evidence leads you to believe. Keeping it rough makes your early assumptions visible.

What Your Draft Needs

The source label identifies the interview evidence as fictional.
The primary target user names one user group.
The I1 observation records one finding from Maya.
The I2 observation records one finding from Jonah.
The hypothesis starts with the user. It uses struggles to before the task. It uses because before the cause. It uses which leads to before the consequence.
The feature line starts with Unvalidated solution idea.

Draft the first-pass note in a scratch space you already use.
Keep your first wording even if another version sounds more polished.
Check that I1 appears beside Maya's observation.
Check that I2 appears beside Jonah's observation.

This first pass is intentionally narrow. Its value is the visible record of what the earliest evidence made you believe.

Add your first pass to the brief

The recorder separates storage from interpretation. Claude formats your decision without making the decision for you.

Return to the neutral recorder conversation from earlier.
Copy this formatting instruction into Claude's message box:

Format the material below under Interview evidence and Provisional hypothesis. Preserve my wording exactly. Keep the solution idea labeled as unvalidated. Do not add analysis.

What Does This Prompt Do?

The instruction assigns your material to Interview evidence and Provisional hypothesis. It also limits Claude to formatting.

Your target user, observations, hypothesis, and unvalidated idea remain your own work.

Paste your first-pass note directly below the formatting instruction.
Confirm the feature line begins with Unvalidated solution idea.

Before you submit it, predict what the recorder will do with your exact wording.

Submit the message in Claude.

You should see your selected target user, both evidence observations, and the provisional hypothesis under the requested headings. Your wording should match the draft exactly.

The remaining discovery brief headings should still be in place for later work.

Did Claude Rewrite Your Work?

Reuse the formatting-only correction from the recorder setup.
Resend the same first pass after Claude returns to the recorder role.

Ask for help comparing the response with your draft: Help me identify where Claude changed my wording or added analysis.

You have captured your first belief without letting the tool improve it. Your reasoning now has a traceable starting point.

Your first hypothesis is now frozen in your own words. Next, you will test it against evidence that may weaken or expand it.

Challenge the Hypothesis

Your first-pass hypothesis turns I1 and I2 into a clear starting belief. That early clarity can anchor a product team to its first compelling pain point.

Does reply speed explain the full problem, or is it only a symptom? You will test that belief against four more fictional interviews while Claude preserves your reasoning.

In this step, get ready to:

Read Interviews I3 through I6 without changing your first hypothesis.
Classify how each interview supports, weakens, or expands your original view.
Add the evidence plus your contradiction table to the discovery brief.

Read the remaining fictional interviews

Disconfirming evidence is information that challenges or complicates an existing belief. It helps you test whether an early hypothesis explains the wider evidence.

Before you read the next four excerpts, do you expect your first hypothesis to explain every concern they raise?

Read each fictional interview in the evidence panel below.

Fictional Interview Evidence

I3, Priya, Tier 2 support agent: "Writing is not the hard part. The hard part is finding which policy is current. Two pages can contradict each other. I would not trust a tool that sends a polished answer without showing its source."
I4, Luis, quality lead: "Our most costly mistakes happen when an agent applies an outdated refund or warranty rule. A fast wrong answer creates rework and escalations."
I5, Amara, new support agent: "I keep five tabs open and still ask a senior agent whether I found the right rule. I want to know why an answer is correct, not just receive text to paste."
I6, Chen, customer: "A quick reply feels good until the next agent reverses it. I care more about getting the right answer the first time than getting an instant answer that changes later."

The first hypothesis now has to account for how agents find current policies. It also has to explain source visibility, costly mistakes, agent confidence, and answer consistency.

This tension is the designed challenge in this step. You have now seen evidence that puts pressure on the first idea before changing it.

Build the contradiction table

A contradiction table makes that tension auditable. Each row separates confirming evidence from evidence that weakens the hypothesis or reveals another part of the problem.

Create a working copy of the contradiction table by using this template:

| Interview | Supports the first hypothesis | Weakens the first hypothesis | Adds a new insight |
| --- | --- | --- | --- |
| I3 |  |  |  |
| I4 |  |  |  |
| I5 |  |  |  |
| I6 |  |  |  |

How to Read the Table

The Supports the first hypothesis column captures evidence that strengthens your initial explanation.
The Weakens the first hypothesis column captures evidence that challenges its cause, scope, or consequence.
The Adds a new insight column captures a need, risk, or affected stakeholder missing from the first hypothesis.
A single row can contribute to more than one column when the excerpt contains different observations.

Complete one row for each interview using short observations.
Keep every observation focused on the evidence instead of a proposed feature.
Count the rows that weaken the hypothesis or add a new insight.

You should have four completed rows. At least two rows should weaken the first hypothesis or expand it with a new insight.

Stuck Between Two Columns?

Re-read the exact excerpt for any row that feels ambiguous.
Place an observation in more than one column when it genuinely contributes to both classifications.

Ask for help without handing over the analysis: Help me distinguish supporting evidence, weakening evidence, and a new insight without classifying these interviews for me.

Add the challenge to your brief

The recorder boundary matters most when evidence becomes complicated. Claude should format your classifications while leaving your original hypothesis unresolved.

Return to the neutral recorder conversation from earlier.
Copy this formatting instruction into the chat composer:

Add these interviews under Interview evidence and place my table under Contradictions and new insights. Preserve my classifications. Do not resolve the contradiction or rewrite my hypothesis.

Why This Instruction Matters

The first sentence places the new evidence under the existing Interview evidence heading.
The completed table goes under Contradictions and new insights so the reasoning remains visible.
The final sentence protects your original hypothesis from being silently rewritten or resolved.

Combine the four interview excerpts with your completed table into one evidence pack.
Paste the combined evidence pack beneath the formatting instruction.
Submit the completed message to Claude.

Before you inspect Claude's formatted response, do you expect the original hypothesis to stay untouched while the challenge table is added?

Inspect Interview evidence to confirm it now includes I3, I4, I5, and I6.
Inspect Contradictions and new insights to confirm it contains your four completed rows.
Compare Provisional hypothesis with your first-pass wording to confirm it remains unchanged.

You should see I3 through I6 under Interview evidence. You should also see a four-row contradiction table with at least two rows that weaken or expand the first hypothesis.

That is the key reasoning shift complete: your brief now shows where broader evidence puts pressure on the original idea.

Did Claude Rewrite Your Hypothesis?

Compare the current Provisional hypothesis section with your original wording.
Restore your original wording if Claude changed or resolved the hypothesis.
Submit your evidence pack again with the same formatting-only instruction.

Get help preserving the recorder boundary: Help me restore my original hypothesis without changing my interview evidence or contradiction table.

Your brief now holds the initial hypothesis beside the evidence that challenges it. Next, you will cluster all six interviews into solution-free opportunities and rank the strongest one.

Rank the Real Opportunity

Your contradiction table shows where the first hypothesis bends under the fuller interview set. The next decision is which evidence-backed problem deserves attention first.

The six fictional excerpts contain several related pains. A transparent ranking keeps the loudest quote or most exciting feature from taking over the roadmap.

Claude remains the neutral recorder. You own the interpretation.

You also own the scores.

In this step, get ready to:

Group the six fictional interviews into three solution-free problem statements.
Produce a ranked evidence table with the exercise rubric.
Write a revised primary user need from the highest-ranked statement.

Cluster the evidence into opportunities

Opportunity framing turns recurring observations into needs or desired outcomes that a product team could address in several ways. Each statement needs evidence from the fictional interviews.

Review interviews I1 through I6 under Interview evidence.
Compare each excerpt with your table under Contradictions and new insights.
Draft three opportunity statements in the Claude message box without submitting them.
Phrase each statement as a user need, pain point, or desired outcome.
Attach every supporting interview ID to its statement.
Test each statement by asking whether several different solutions could satisfy it.
Reframe any statement that names a specific feature.

Opportunity or solution?

An opportunity names an unmet need, pain point, or desired outcome. A solution names a product or feature that could address it.

A statement that supports several possible approaches leaves room for discovery. This keeps the evidence in charge of the next decision.

Your draft now shows three solution-free opportunities. Each one has a visible trail back to the fictional interviews.

Score and rank the opportunities

This scoring rubric is an instructional heuristic for the simulation. Its value comes from making your reasoning visible.

Add columns for Opportunity, Supporting interview IDs, Evidence breadth, User consequence, Outcome fit, Total, and Rank.
Assign each opportunity an evidence-breadth score. Use 1 for one interview ID. Use 2 for two IDs. Use 3 for three or more IDs.
Assign each opportunity a user-consequence score. Use 1 for inconvenience. Use 2 for repeated delay or rework. Use 3 for an incorrect outcome or trust risk.
Assign each opportunity an outcome-fit score. Use 1 for indirect support. Use 2 for plausible support. Use 3 for direct support of a correct first response.
Add the three scores for a total from 3 to 9.
Rank the opportunities from the highest total to the lowest total.
Break any tie using the strongest consequence first.
Record one sentence of tie-break reasoning using the clearest evidence.

You've made the trade-off visible. Your top rank now rests on evidence breadth, user consequence, and outcome fit.

Write the revised primary user need

A user-need statement turns the top opportunity into a decision anchor without prescribing a feature. The final constraint captures what a future approach must respect.

Use the same primary target user from your provisional hypothesis.
Write the revised need with this structure:

As a [[PRIMARY_USER="primary user"]], when [[TRIGGER="trigger or situation"]], I need [[SOLUTION_FREE_NEED="solution-free need"]], so that [[DESIRED_OUTCOME="desired outcome"]], because [[IMPORTANT_CONSTRAINT="important constraint"]].

What does this statement capture?

The primary user keeps the statement focused on the support agent.
The trigger names the moment when the need becomes important.
The solution-free need describes what the user must accomplish.
The desired outcome connects the need to a correct first response.
The important constraint records a condition that the evidence says matters.

Fill each placeholder with your own wording.
Attach at least three supporting interview IDs to the completed statement.
Audit the statement for feature names.

Before you submit this, do you think the recorder will preserve your ranking without adding a feature?

Send your completed opportunity table and revised statement to Claude with this prompt:

Place my table under Opportunity ranking and my statement under Primary user need. Preserve my scores, evidence IDs, and wording. Do not propose solutions.

[[YOUR_WORK="your opportunity table and revised statement"]]

What does this prompt protect?

The prompt limits Claude to placement and formatting. Your wording, scores, evidence links, and product judgment remain unchanged.

Review the Opportunity ranking section.
Confirm it shows three opportunities with supporting IDs, three scores, totals, and rank order.
Review the Primary user need section.
Confirm the revised statement cites at least three interview IDs.
Compare Claude's formatted text with your original wording.

Strong work. Your brief now shows why one solution-free opportunity ranks above the others.

Did Claude change your reasoning?

Claude can drift beyond the recorder role when it sees an incomplete table or ambiguous wording. Your manual draft remains the source of truth.

Restate the formatting-only boundary from earlier.
Replace any rewritten wording with your original text.
Remove any feature proposal that Claude introduced.

Help me restore my discovery brief without changing my reasoning.

Your top opportunity now has a traceable case behind it. Next up, you'll turn that decision into a concise discovery brief and stakeholder pitch.

Defend Your Discovery Brief

Your product discovery work now links six fictional interviews to a ranked opportunity set. This step turns that reasoning into a brief a stakeholder can challenge.

Research changes a product decision when stakeholders can trace a recommendation to evidence. Its uncertainty must point to a concrete next research action.

Claude will format the final Markdown brief. You remain responsible for every decision inside it.

In this step, get ready to:

Complete the brief's remaining decision fields.
Write a 60-second stakeholder pitch with traceable evidence.
Produce an audited final brief in Claude.

Complete the decision fields

A recommendation names the opportunity worth investigating next. The surrounding fields show where the fictional evidence stops.

Write one recommendation that names the single opportunity the team should investigate next.
Write one tempting conclusion that the fictional evidence cannot support.

Your draft now separates a research direction from an unsupported leap.

Label that conclusion What the evidence does not support.
Record two research gaps under Assumptions and unknowns.

These gaps set the boundary of your current confidence.

Write one research question under Next research action.
Name the target participant group for that question.

Your remaining uncertainty now has a specific audience.

State which product decision the answer would change.

That is the decision core complete. It gives stakeholders a direction without hiding the gaps.

Write the stakeholder pitch

A stakeholder pitch compresses the brief without cutting traceability. Each part moves the listener from the original request toward a justified next action.

Open your pitch with the original request for AI-generated replies.
Describe what the evidence changed with interview IDs beside each claim.

The opening now establishes why the original feature request needs deeper investigation.

Name the top opportunity.
Explain why it ranks first using the opportunity scores.

The middle of your pitch now connects the recommendation to visible reasoning.

State one unresolved question.
Close with the next research action.

The ending now makes uncertainty useful.

Read your completed pitch aloud once to test its 60-second length.

You should hear a clear path from the original request to the top opportunity. The final line should point to the next research action.

Your stakeholder story is now ready to defend the recommendation without overstating the evidence.

Format and audit the final brief

The final formatting pass should improve readability without changing your reasoning. Your audit catches unsupported certainty before a stakeholder sees it.

Prepare Claude to format the complete artifact by copying this prompt:

Format my completed material as one concise Markdown discovery brief using the existing headings. Preserve my decisions and wording. Do not add evidence, recommendations, or claims. Keep every interview ID attached to the claim it supports. After the brief, format my stakeholder pitch under 60-second pitch.

What does this prompt protect?

The instruction Preserve my decisions and wording. holds Claude to its recorder role.
The instruction Keep every interview ID attached to the claim it supports. protects traceability.
The 60-second pitch heading keeps the spoken summary inside the final artifact.

Paste your completed decision fields directly below the prompt in the same message.
Paste your completed stakeholder pitch beneath the decision fields.

Before you submit, ask yourself whether Claude will preserve every decision without adding a claim. Hold your prediction in mind.

Submit the completed message in the open Claude conversation.

Claude should return one concise Markdown brief under the existing headings. Your decision fields should appear in your original wording with interview IDs still attached.

The 60-second pitch should appear under its own heading.

A polished format can still hide a changed meaning.

Confirm every section from Discovery goal through Primary user need remains present.
Confirm Interview evidence still contains I1 through I6.
Check every factual claim for at least one interview ID.
Confirm every interview excerpt is clearly labeled fictional.
Confirm the original Unvalidated solution idea label remains attached to the early feature idea.
Remove any line that presents the fictional evidence as real market validation.
Rewrite any feature statement that sounds approved.

This pass protects traceability without inflating the evidence.

Confirm Recommendation names one opportunity to investigate next.
Confirm What the evidence does not support remains clearly labeled.
Confirm Assumptions and unknowns contains two research gaps.
Confirm Next research action includes one research question.
Confirm the target participant group is named.
Confirm the decision impact is stated.

Your brief now shows what the team should investigate next. It also shows what the current evidence cannot settle.

Confirm the pitch opens with the original AI-generated-replies request.
Confirm the pitch explains what the evidence changed.
Confirm the pitch names the top opportunity with its rationale.
Confirm the pitch states one unresolved question.
Confirm the final line gives the next research action.

Did Claude change your reasoning?

Compare the formatted brief with your original draft.
Delete any inference Claude added.
Resubmit the formatting prompt with a direct reminder to preserve your wording.

If a change is hard to spot, help me audit Claude's formatting changes.

Strong finish. Your discovery brief now supports its recommendation with traceable evidence. Its visible uncertainty points to the next research action.

Secret mission

Add the Customer Perspective

Your brief already centers the support agent as the primary user. Extend the evidence chain to show how the same problem affects Chen in I6, then add a customer-trust guardrail without changing the primary user need.

Clean Up Your Resources

This project adds no cost beyond your existing Claude subscription because it created no additional billable resources. Choose whether to keep, pause, or delete the Claude conversation that holds your discovery brief.

Resources you used:

One Claude conversation containing your completed discovery brief, fictional interview evidence, contradiction table, ranked opportunities, stakeholder pitch, secondary user impact, and customer-trust guardrail.

Keep everything running

Keep the conversation available so its evidence trail can become the starting artifact for the next roadmap project.

Find the conversation in Claude's left sidebar or Chats and tasks.
Click the ⋮ menu beside the conversation.
Select Rename.
Enter AI PM Roadmap 1 - Discovery Brief as the conversation title.

Your discovery work now has a clear name and a ready-made place in the next roadmap project.

Pause - I'll come back to this later

Nothing runs in the background for this project. Leave the conversation unchanged if you want to return to the brief later.

Keep the Claude conversation in your chat history.

Your completed brief remains exactly where you left it with no additional project cost.

Delete - I don't want to use this again

Deleting the conversation is permanent. Your discovery brief and stakeholder pitch disappear with it.

On the web:

Find the conversation in the left sidebar or Chats and tasks.
Click the ⋮ menu beside the conversation.
Select Delete.
Confirm by selecting Delete again.

On Claude for iOS:

Touch and hold the conversation.
Tap Delete.
Confirm the deletion.

On Claude for Android:

Return to the open conversation from earlier.
Tap the top-right ⋮ menu.
Tap Delete.
Confirm the deletion.

You should no longer see the discovery conversation in your chat history.

Nice Work!

You did it! You built an evidence-backed product discovery brief that traces every factual claim to six fictional interviews. Your 60-second pitch now turns that brief into a defensible stakeholder recommendation.

You've learned how to:

Built a traceable evidence trail from six fictional interviews. You kept Claude in a neutral recorder role throughout.
Used contradiction analysis to expose the limits of your first hypothesis. You ranked three solution-free opportunities with a transparent scoring rubric.
Wrote a solution-free user need grounded in interview IDs. You defended an evidence-backed recommendation through a 60-second pitch. The pitch names uncertainty plus the next research action.
Secret Mission: Added a secondary-user impact section grounded in I6. Added a customer-trust guardrail without changing the primary support-agent need.

Ready to quiz yourself?