prd-saas-dev

Client Discovery Workflow

Purpose

Define the human-AI collaboration protocol for evaluating a new prospect: who finds the problem, who maps the assets, who decides direction. Prevents the AI from generating both problems and solutions (which leads to self-reinforcing confirmation bias).

When to Use

• A new client lead comes in
• The user shares a Perplexity / market research link
• The user says “help me look at this client”
• Before any /spectra-propose for a client engagement

---

Roles

Actor	Responsibility	Why
User (Fish)	Problem discovery via external sources (Perplexity, interviews)	External citations prevent self-justification
AI agent	Asset inventory and mapping	AI knows the internal codebase best
User (Fish)	Final go / no-go decision	Business judgment requires human owner

---

Three-Stage Pipeline

Stage 1: User runs external research (Perplexity / Gemini / interviews)
   |
   v
Stage 2: AI maps pain points to owned assets, produces options
   |
   v
Stage 3: User decides direction (A / B / C / modify / drop)
   |
   v
If go: trigger /spectra-propose

---

Stage 2: AI Execution Steps

When the user provides Perplexity output or research summary, execute in order:

Step 1: Confirm input quality

• Is there a named client?
• Are the pain points concrete (not abstract market needs)?
• Do pain points pass the “repetitive, time-consuming, hard-to-measure” filter?

If any check fails: ask the user to refine research before proceeding.

Step 2: Inventory owned assets

Do not rely on memory. Re-scan every time:

ls products/
ls 8-外掛/
ls skills/

For each asset, record:

• Original problem it solved
• Whether it can be retooled for the client’s pain (low / mid / high modification)

Step 3: Build pain-to-asset mapping table

Client Pain	Owned Asset	Modification Effort
A	X	low / mid / high

Apply the methodology in client-opportunity-mapping.md.

Step 4: Generate three-tier proposal

Tier	Model	Client Commitment	Our Risk
A	SaaS rental of existing tools	lowest	lowest
B	Co-development with client as case study	mid	mid
C	Full integration build + monthly	highest	highest (but most replicable)

Step 5: Recommend MVP

Selection criteria (both must hold):

1. Decision-maker (parent / boss / paying party) can feel the value within 7 days of launch
2. Operator (teacher / staff) has zero learning curve

If no candidate meets both: redo asset mapping or recommend dropping the client.

Step 6: Output decision package to user

Format:

1. Mapping table
2. Three tier options (A/B/C) with scope, timeline, pricing basis
3. MVP recommendation with rationale
4. Anti-pattern flags (when NOT to take this client)

---

Signal-to-Action Table

User signal	AI action
Drops a Perplexity link or research dump	Enter Stage 2
“Help me look at this client” without research	Suggest running Stage 1 first
“This proposal is wrong”	Return to mapping table, do not argue
“Go with A”	Trigger /spectra-propose

---

Anti-Patterns

• Do NOT generate pain points yourself (no external validation)
• Do NOT skip asset inventory and jump to “let’s build X”
• Do NOT propose only one option (always three tiers)
• Do NOT recommend a new project when an existing asset covers the pain (violates entropy reduction)

---

Cross-References

• Stage 2 detailed methodology: client-opportunity-mapping.md
• Pain validation criteria: persona-pain-scenario.md
• Post-decision workflow: sdd-workflow.md