stindardlogic/product-management-sft-100k

Product Management SFT (100K)

100,000 ShareGPT conversations demonstrating expert-level product management across PRD writing, feature prioritization, OKR setting, roadmap planning, user research, competitive analysis, and stakeholder communication.

Motivation

AI assistants for product management commonly fail by:

• —Generic frameworks without application: Explaining RICE scoring without actually scoring the user's features; describing OKRs without writing them
• —Ignoring constraints: Writing PRDs that assume unlimited resources or ignore technical feasibility
• —Missing the business case: Prioritization without quantifying business impact (ARR at stake, retention improvement, cost reduction)
• —Vague user research guidance: "Talk to customers" without explaining recruiting, question design, or synthesis
• —Template delivery without thinking: Providing a PRD template instead of a filled-out PRD
• —No stakeholder awareness: Advice that ignores how to get organizational alignment and manage competing priorities

This dataset trains models to do the actual work of product management — not explain what PM is, but produce the artifacts and decisions that PMs produce.

Dataset Description

100,000 conversations across 8 product management categories:

Category Distribution

Format

{
  "conversations": [
    {
      "from": "human",
      "value": "Write a PRD for a bulk user import feature in our B2B SaaS project management tool..."
    },
    {
      "from": "gpt",
      "value": "## Product Requirements Document: Bulk User Import\n\n**Feature:** Bulk User Import via CSV\n..."
    }
  ],
  "metadata": {
    "category": "prd_writing",
    "context": "B2B SaaS"
  },
  "id": "abc123"
}

Key Properties of Responses

1. Artifact delivery, not advice: Responses produce the actual PRD, the scored RICE matrix, the OKR set, the roadmap — not descriptions of what those things should look like.

2. Business case embedded: Every feature decision includes quantitative business impact: ARR at risk, retention improvement target, cost reduction, or pipeline value.

3. Constraint-aware: PRDs address non-goals, open questions, and risks — not just what will be built but what won't and why.

4. Stakeholder thinking: Roadmap and communication responses acknowledge how different stakeholders (sales, engineering, executives) interpret the same information differently and how to frame accordingly.

5. Frameworks applied, not explained: RICE doesn't get defined abstractly — it gets applied to the user's actual features with specific scores and explicit reasoning for each estimate.

6. Anti-patterns included: Responses name common mistakes in the domain — bad sprint planning patterns, survey design errors, prioritization fallacies — giving readers contrast that builds judgment.

7. Metrics specified: Every major decision includes success metrics with baselines and targets, not vague "we'll know it's working when people use it" qualifications.

Use Cases

• —SFT fine-tuning for AI product management tools (Productboard AI, Linear AI, Notion AI)
• —Training AI copilots for product managers and founders
• —Building AI advisors for startup incubators and accelerators
• —Improving model performance on structured business reasoning and planning tasks
• —Training models for executive coaching, business strategy, and product leadership applications
• —Fine-tuning AI assistants for MBA programs and product management bootcamps

License

Apache 2.0