product-delivery（产品交付）

Product Delivery System

"The goal isn't shipping. The goal is learning whether your bet was right."

This skill covers the Delivery System — how we ship, measure, and learn. It runs discovery and delivery in parallel (dual-track), ships with staged rollouts, measures with a clear hierarchy, reflects through retrospectives, and executes GTM with precision.

Part of: Modern Product Operating Model — a collection of composable product skills.

Related skills: product-strategy, product-discovery, product-architecture, ai-native-product, product-leadership

When to Use This Skill

Use this skill when:

• Planning how to roll out a new feature or product
• Designing a metrics hierarchy for a bet or product
• Running bet retrospectives after shipping
• Executing GTM launches
• Setting up dual-track development rhythm
• Deciding when to scale, iterate, or kill a bet

Cadence: Continuous | Owner: Product Trio + GTM Team

The Problem This Solves

Most teams either:

1. Ship features and never measure impact
2. Measure vanity metrics that don't connect to outcomes
3. Do "big bang" launches that create risk
4. Never officially conclude bets—zombies live forever
5. Treat GTM as marketing's problem after PM ships

The Delivery System ensures shipping is the beginning of learning, not the end.

Philosophy

Core Beliefs

1. Discovery and delivery run in parallel — Don't pause discovery to deliver
2. Staged rollouts are the default — Ship to 10% before 100%
3. Metrics exist in hierarchy — Leading → Core → Lagging
4. Every bet gets a retrospective — Explicit scale/iterate/kill decision
5. GTM is a product responsibility — PM owns adoption, not just availability

What This Framework Rejects

• Ship and forget (no measurement)
• Big bang launches (maximum risk)
• Vanity metrics (activity without outcome)
• Zombie bets (never concluded, never killed)
• Throwing features over the wall to marketing

Framework Components

1. Dual-Track Development

The Core Idea: Discovery and delivery happen simultaneously. While one bet is being built, the next bet is being shaped.

Week 1    Week 2    Week 3    Week 4    Week 5    Week 6
─────────────────────────────────────────────────────────
[  Discover Bet B  ][  Shape Bet B  ][  Discover Bet C  ]
         [    Build Bet A    ][   Build Bet B   ]
                    [ Ship A ]        [ Ship B ]

How It Works:

Time Allocation (Example):

Coordination Points:

• Weekly sync: What's in flight on each track
• Handoff moment: When a bet moves from "shaped" to "building"
• Learning moment: When shipped bet results inform discovery

0→1 Mode: Tracks may blur. Everyone does everything. Speed > separation.

Scaling Mode: Clear separation. Dedicated discovery time. Research ops support.

2. Staged Rollout

The Core Idea: Never ship to everyone at once. Start small, learn, expand.

Default Rollout Stages:

Progression Criteria:

Feature Flags:

• Every significant feature ships behind a flag
• Flags enable instant rollback
• Flags enable % rollout control
• Flags are cleaned up after GA (don't accumulate debt)

Rollback Triggers:

• Guardrail metric breached
• Error rate > threshold
• Customer-reported critical issue
• Leading metrics trending wrong

0→1 Mode: Stages can be compressed. Alpha might be 3 customers for 2 days.

Scaling Mode: Formal stage gates. Release management. Beta programs.

3. Metrics Hierarchy

The Three-Tier Model:

┌─────────────────────────────────────────────────────┐
│                 LAGGING METRICS                     │
│         (Revenue, Retention, NPS)                   │
│         Move slowly, hard to attribute              │
├─────────────────────────────────────────────────────┤
│                  CORE METRICS                       │
│       (Activation, Engagement, Conversion)          │
│       The outcomes your bets target                 │
├─────────────────────────────────────────────────────┤
│                LEADING METRICS                      │
│         (Feature adoption, Task completion)         │
│         Move fast, early signal                     │
└─────────────────────────────────────────────────────┘

Metric Types:

Hierarchy Example (Activation Bet):

Lagging:    Revenue growth (quarterly)
               ↑
Core:       Activation rate (weekly)
               ↑
Leading:    Onboarding completion (daily)
            First value action (daily)
               ↑
Guardrail:  Support tickets (daily)
            Error rate (real-time)

Metric Selection Criteria:

Dashboard Design:

• Leading metrics: Real-time or daily
• Core metrics: Weekly view with trend
• Lagging metrics: Monthly/quarterly view
• Guardrails: Alerting, not just reporting

4. Bet Retrospectives

The Core Idea: Every bet concludes with an explicit decision: Scale, Iterate, or Kill.

Retrospective Format:

BET RETROSPECTIVE: [Name]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Timebox: [Duration] | Shipped: [Date]

HYPOTHESIS REVIEW
Original: "We believed [X] would result in [Y]"
Result: [ ] Confirmed  [ ] Disproved  [ ] Inconclusive

METRICS REVIEW
| Metric    | Target | Actual | Verdict |
|-----------|--------|--------|---------|
| Primary   | [X]    | [Y]    | ✅ / ❌  |
| Secondary | [X]    | [Y]    | ✅ / ❌  |
| Guardrail | [X]    | [Y]    | ✅ / ❌  |

KEY LEARNINGS
• [Learning 1]
• [Learning 2]
• [Learning 3]

DECISION: [ ] SCALE  [ ] ITERATE  [ ] KILL

NEXT STEPS
• [Action 1]
• [Action 2]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Decision Framework:

Iteration Limits:

• Maximum 2-3 iteration cycles before forcing Scale or Kill
• Each iteration must have a new hypothesis
• Time-bound iterations (don't let them drag)

Retrospective Cadence:

• Run at timebox end, regardless of "completion"
• Include full trio + relevant stakeholders
• 30-60 minutes maximum
• Document in central repository

0→1 Mode: Informal but still explicit. Even a Slack message: "Bet X result: [Y]. Decision: [Z]."

Scaling Mode: Formal retrospective meetings. Learning database. Cross-team sharing.

5. GTM Execution

The Core Idea: PM owns adoption, not just availability. GTM is part of delivery, not an afterthought.

GTM Components:

Launch Tiers:

Launch Checklist (Tier 1/2):

Adoption Metrics (PM Responsibility):

0→1 Mode: PM does GTM. Scrappy launches. Focus on learning, not polish.

Scaling Mode: PMM partnership. Launch playbooks. Integrated marketing calendar.

Key Outputs

Templates

This skill includes templates in the templates/ directory:

• rollout-plan.md — Staged rollout checklist
• metrics-hierarchy.md — Metrics design template
• bet-retrospective.md — Retrospective format and decision framework
• launch-checklist.md — GTM execution checklist by tier

Using This Skill with Claude

Ask Claude to:

1. Design rollout plan: "Create a staged rollout plan for [feature]"
2. Build metrics hierarchy: "Design a metrics hierarchy for [bet/product]"
3. Set success criteria: "What metrics should determine success for [bet]?"
4. Plan retrospective: "Create a retrospective agenda for [bet] that shipped [X]"
5. Analyze results: "Given these metrics, should we scale, iterate, or kill?"
6. Design launch tier: "What launch tier should [feature] be? What GTM is needed?"
7. Create launch checklist: "Build a GTM checklist for [Tier X] launch"
8. Identify guardrails: "What guardrail metrics should we watch for [bet]?"
9. Plan dual-track: "Help me design a dual-track rhythm for [team size]"
10. Write changelog: "Write release notes for [feature] at [tier]"

Connection to Other Skills

Anti-Patterns to Avoid

Quick Reference: Delivery Quality Checklist

Before GA:

• Rollout staged — Not shipping to 100% first
• Feature flagged — Can roll back instantly
• Metrics instrumented — Can measure leading/core/lagging
• Guardrails defined — Know what triggers rollback
• GTM prepared — Docs, training, comms ready
• Retrospective scheduled — Date on calendar at timebox end
• Success criteria agreed — Team knows what Scale/Iterate/Kill means

Sources & Influences

• Marty Cagan — INSPIRED, EMPOWERED
• Eric Ries — The Lean Startup
• Teresa Torres — Continuous Discovery Habits
• April Dunford — Obviously Awesome (for GTM)

*Part of the Modern Product Operating Model by Yannick Maurice*