Reference source: Teresa Torres, Continuous Discovery Habits
Applicable Scenarios
- The business goal is clear, but it is unclear what feature to build
- There are too many candidate solutions and it is unclear which to choose
- The team disagrees on "what the user pain points are"
- Exploration of high-uncertainty requirements (before writing the PRD)
- Organizing the outputs of continuous user research
Inapplicable Scenarios
- The requirements are already clear and only the PRD needs to be written (directly use prd-writing)
- Single-feature small iteration (directly use user-story)
- Urgent bug fix (no exploration needed)
Core Idea
First diverge on opportunities, then converge on solutions. First validate assumptions, then invest in development.
OST is a tree structure that starts from the product goal and decomposes layer by layer into:
- User opportunities (pain points/needs/expectations)
- Candidate solutions
- Validation experiments
[产品目标 / Outcome]
↓
┌───┴───┐
│机会 A │ ← 用户的痛点/需求
└───┬───┘
↓
┌───┴───┐ ┌───────┐
│方案 A1│ │方案 A2│ ← 解决这个痛点的不同方式
└───┬───┘ └───┬───┘
↓ ↓
[实验 A1a] [实验 A2a] ← 验证方案可行的最小成本测试
Four-Layer Structure
1. Outcome (product goal)
A quantifiable business result, not "what to do" but "what to achieve".
✅ 好的:
- 7 日留存从 X% 提升到 Y%
- 月活用户从 1 万到 5 万
- 客单价从 50 元提升到 80 元
❌ 差的:
- 提升用户体验
- 让产品更好用
- 增加用户活跃度(没有数字)
2. Opportunities
Pain points, needs, and expectations from the user's perspective.Use user language, not solution language。
✅ 好的:
- "我搜索商品时找不到想要的"
- "我下单时不确定能不能按时收到"
- "我退款流程太复杂不想再买"
❌ 差的:
- "需要搜索算法优化"(解决方案)
- "改进物流系统"(解决方案)
- "退款一键操作"(解决方案)
Opportunities must have evidence: user interviews, customer service records, behavioral data, questionnaires.
3. Solutions
Specific ways to solve a certain opportunity.There should be multiple candidate solutions under one opportunity, not just one。
机会:"我搜索商品时找不到想要的"
方案候选:
- 搜索算法升级(语义搜索)
- 智能推荐补充("找不到?试试这些")
- 分类导航优化
- 多维度筛选器
- 视觉搜索(拍照搜图)
4. Experiments
Before investing in full development, validate whether the solution is feasible at the lowest cost.
See pol-probe skill。
Workflow
1. 明确 Outcome(必须可量化)
↓
2. 收集用户洞察(访谈/数据/客服记录)
↓
3. 识别用户机会(用户语言,至少 5~10 个)
↓
4. 按重要性筛选 Top 3 机会
↓
5. 每个机会下生成 3+ 个候选方案
↓
6. 每个方案设计 1 个最小验证实验
↓
7. 优先执行成本最低、信息量最大的实验
↓
8. 实验结果反馈到树上 → 决定继续 / 调整 / 放弃
Output Format (Mermaid)
flowchart TD
O[7日留存从 30% 提升到 45%]
O --> Op1[用户搜索时找不到想要的]
O --> Op2[用户对推荐内容不感兴趣]
O --> Op3[用户登录后不知道下一步做什么]
Op1 --> S1[语义搜索升级]
Op1 --> S2[智能推荐补充]
Op1 --> S3[多维度筛选]
S1 --> E1[A/B 测试: 5% 流量]
S2 --> E2[假门测试: 落地页]
S3 --> E3[原型用户测试]
Output Format (Markdown)
## OST: 提升 7 日留存
### Outcome(目标)
7 日留存从 30% 提升到 45%(季度目标)
### Opportunities(机会)
#### Op1: "我搜索商品时找不到想要的"
- 证据:客服记录显示每周 200+ 次"找不到"投诉
- 优先级:高
- 影响留存:中
#### Op2: "我对推荐内容不感兴趣"
- 证据:推荐位 CTR 仅 2%(行业平均 5%)
- 优先级:高
- 影响留存:高
#### Op3: "我登录后不知道下一步做什么"
- 证据:新用户 D1 跳出率 60%
- 优先级:中
- 影响留存:高
### Solutions(候选方案)
#### Op1 的方案
| # | 方案 | 成本 | 影响 | 推荐 |
|---|------|------|------|------|
| S1 | 语义搜索升级 | 高 | 高 | ⭐⭐⭐ |
| S2 | 智能推荐补充 | 中 | 中 | ⭐⭐ |
| S3 | 多维度筛选 | 低 | 中 | ⭐⭐⭐ |
#### Op2 的方案
...
### Experiments(验证实验)
| 实验 | 类型 | 验证什么 | 成本 | 时间 |
|------|------|---------|------|------|
| E1 | A/B 测试 | 语义搜索 vs 关键词搜索的 CTR | 中 | 30min |
| E2 | 假门测试 | 智能推荐的接受度 | 低 | 15min |
### 下一步
优先执行 E2(成本最低,信息量大)
Quality Self-Check
□ Outcome 是否可量化(有具体数字)
□ Opportunities 是否用了用户语言(不是解决方案语言)
□ 每个 Opportunity 是否有证据支持(不是凭空想象)
□ 每个 Opportunity 下是否有 3+ 个候选方案
□ 每个方案是否有验证实验(不是直接投入开发)
□ 是否优先执行成本最低、信息量最大的实验
Common Pitfalls
- Skipping Opportunity and going straight to Solution—"users need AI recommendations" → you should first ask "what are the user's pain points"
- Writing Opportunity as a solution—"needs a search upgrade" → it should be "users cannot find what they want"
- Only one solution per opportunity—loses the opportunity for side-by-side comparison
- No evidence supporting the opportunity—pain points made up by the team on the spot
- Outcome is not quantifiable—"improve experience" → cannot verify whether it has been achieved
- Starting development without validation—bypassing experiments and directly investing in the full feature
- The tree is one-off—OST should be continuously updated as experiment results come in
Supporting Templates
templates/opportunity-solution-tree-template.md— Complete OST templatetemplates/discovery-brief-template.md— Opportunity evidence collection template
Collaboration with Other Skills
上游:
用户研究 / 数据分析 / 客服反馈
平行:
pol-probe → 设计每个方案的验证实验
prioritization → 当方案太多时,用 RICE/ICE 排序
下游:
实验通过 → mvp-scoping 划分范围
范围确定 → prd-writing 输出 PRD