prioritization

Quantitative Priority Scoring

Applicable Scenarios

• Too many items in the backlog and need to rank them
• The team disagrees on priorities
• Quarterly planning needs to select 10 from 50 candidates
• Ranking validation experiments (which one to do first)

Inapplicable Scenarios

• Still in the opportunity exploration stage (should first use opportunity-tree）
• MVP slicing (should directly use mvp-scoping 's MoSCoW)
• Strategic-level direction selection (quantitative frameworks cannot replace strategic judgment)

Two Core Frameworks

RICE Scoring (preferred when user data is available)

Reference source:Intercom RICE

公式：RICE = (Reach × Impact × Confidence) / Effort

Four factors:

Factor	Meaning	Scoring Method
Reach	How many users are affected	Specific number (e.g., 2,000 users per month)
Impact	Degree of impact on each user	3=Massive / 2=High / 1=Medium / 0.5=Low / 0.25=Minimal
Confidence	Estimated confidence	100%=Has data / 80%=Has evidence / 50%=Guess
Effort	Effort	AI workflow measured in minutes

Example:

功能 A：商品搜索功能
  Reach: 5000（每月活跃用户都会用）
  Impact: 2（高，直接影响转化）
  Confidence: 80%（竞品数据支持）
  Effort: 90 min（AI 实现）
  RICE = (5000 × 2 × 0.8) / 90 = 88.9

功能 B：商品评分系统
  Reach: 1000（只有买家会用）
  Impact: 1（中，影响决策）
  Confidence: 50%（猜测）
  Effort: 60 min
  RICE = (1000 × 1 × 0.5) / 60 = 8.3

→ 功能 A 优先（RICE 更高）

ICE Scoring (preferred for quick decisions)

Reference source: Sean Ellis (Father of Growth Hacking)

公式：ICE = Impact × Confidence × Ease

Three factors (each scored 1–10):

Factor	Meaning
Impact	Impact on the goal
Confidence	Estimated confidence
Ease	Implementation difficulty (10 = very easy, 1 = extremely difficult)

Applicable:

• Experiment ranking (A/B test candidates)
• Quick ranking of growth strategies
• When precise user data is not needed

Decision Tree to Use

有真实用户数据吗？
  ├─ 有 → RICE
  └─ 没有 → ICE

需要排序的项目数量？
  ├─ < 10 个 → ICE 即可
  ├─ 10~30 个 → RICE 或 ICE 都可
  └─ > 30 个 → 先 ICE 粗筛，再 RICE 精排前 10

时间预算？
  ├─ 1 小时 → ICE
  └─ 半天 → RICE

Workflow

1. 列出所有候选项
   ↓
2. 选择评分框架（RICE / ICE）
   ↓
3. 逐项打分
   ↓
4. 计算总分并排序
   ↓
5. 检查 Top 5 是否符合直觉
   - 符合 → 输出结果
   - 不符合 → 重新审视评分（哪里高估/低估了）
   ↓
6. 输出排序表 + 推荐执行顺序
   ↓
7. 标注前 N 项进入下一阶段

Output Format

## 优先级评分结果（RICE）

| # | 功能 | Reach | Impact | Confidence | Effort | RICE | 排序 |
|---|------|-------|--------|------------|--------|------|------|
| 1 | 商品搜索 | 5000 | 2 | 0.8 | 90 | 88.9 | P0 |
| 2 | 用户登录优化 | 8000 | 1 | 1.0 | 60 | 133.3 | P0 |
| 3 | 商品评分 | 1000 | 1 | 0.5 | 60 | 8.3 | P2 |
| ... |

### 推荐执行顺序

P0（本期必做）：用户登录优化、商品搜索
P1（本期争取）：[排名 3-5]
P2（下期考虑）：[排名 6+]

### 评分依据

- 用户登录优化的 Reach：来自最近一周的 PV 数据
- 商品评分的 Confidence 较低：缺乏类似产品对比数据

Quality Self-Check

□ 是否选对了框架（RICE 用于有数据，ICE 用于快速）
□ Reach 是否用了具体数字（不是"很多用户"）
□ Confidence 是否标注了依据（数据/证据/猜测）
□ Effort 是否包含了所有相关工作流的工作量
□ Top 5 排序是否符合直觉（不符合说明评分有问题）
□ 是否标注了 P0/P1/P2 切分线

Common Pitfalls

1. Using percentages instead of absolute numbers for Reach—"50% of users" → should be "X thousand users"
2. Scoring everything 3 for Impact—if all features are "massive," there is no differentiation
3. Confidence is all 100%—this hides uncertainty
4. Effort only counts development—forgetting design/QA/deployment/operations
5. Blindly trusting the formula—RICE is an aid and cannot replace strategic judgment
6. Frequently adjusting scores to fit the result—repeatedly changing scores to make a certain feature rank first
7. Not writing the rationale—others cannot understand or review it

Supporting Templates

• templates/rice-template.md — RICE scoring template
• templates/ice-template.md — ICE scoring template
• templates/prioritization-result-template.md — Ranking result output template

Collaboration with Other Skills

上游：
  opportunity-tree → 提供候选机会和方案
  mvp-scoping → 提供候选功能列表

平行：
  prd-writing → 排序结果作为 PRD 第 6 节优先级依据

下游：
  转交项目经理 → 按 P0/P1/P2 拆排期