klen/hypothesis-tree

Hypothesis Tree - Structured Problem Decomposition

A Hypothesis Tree is a structured approach to breaking down complex questions into testable hypotheses. Originally from management consulting (McKinsey), it ensures MECE (Mutually Exclusive, Collectively Exhaustive) coverage of a problem space.

When to Use This Skill

• Validating new product or feature ideas
• Investigating why metrics are underperforming
• Planning user research or experiments
• Breaking down ambiguous strategic questions
• Prioritizing what to test first
• Communicating analysis structure to stakeholders

Core Concepts

Structure of a Hypothesis Tree

                    Main Question
                    "Why is X happening?"
                          |
          +---------------+---------------+
          |               |               |
     Hypothesis A    Hypothesis B    Hypothesis C
          |               |               |
       +--+--+         +--+--+         +--+--+
       |     |         |     |         |     |
     Sub-   Sub-     Sub-   Sub-     Sub-   Sub-
     hyp    hyp      hyp    hyp      hyp    hyp

MECE Principle

Mutually Exclusive: No overlap between branches Collectively Exhaustive: All possibilities covered

Good MECE:                    Bad (not MECE):
+----------------+            +----------------+
| New users      |            | Mobile users   | <- Overlap
|----------------|            |----------------|
| Returning      |            | New users      | <- Overlap
| users          |            |----------------|
+----------------+            | Some users     | <- Vague
                              +----------------+

Hypothesis Format

Strong hypotheses are:

| Element | Description | Example | | --------------- | ----------------------- | ---------------------------------------- | | Specific | Clear, measurable | "Checkout abandonment is >70% on mobile" | | Testable | Can be proven/disproven | Not "users don't like it" | | Falsifiable | Could be wrong | Has clear failure criteria | | Actionable | Leads to decision | If true → do X, if false → do Y |

Analysis Framework

Step 1: Frame the Question

Convert vague concerns into structured questions:

| Vague | Structured | | ----------------------- | --------------------------------- | | "Growth is slow" | "Why is our MoM user growth <5%?" | | "Users aren't engaged" | "Why is D7 retention below 20%?" | | "Feature isn't working" | "Why is feature X adoption <10%?" |

Step 2: Generate First-Level Hypotheses

Brainstorm potential explanations, then organize MECE:

Question: "Why is signup conversion <30%?"

Level 1 Hypotheses:
├── Awareness: Users don't understand the value proposition
├── Ability: The signup process is too difficult
├── Motivation: The perceived benefit isn't worth the effort
└── Technical: Bugs/errors prevent completion

Step 3: Decompose to Testable Level

Keep breaking down until hypotheses are directly testable:

Ability: The signup process is too difficult
├── Too many fields required
├── Password requirements unclear
├── Form validation confusing
└── Mobile experience broken

Step 4: Prioritize and Test

| Hypothesis | Evidence Available | Test Effort | Impact if True | | ---------- | ------------------ | ----------- | -------------- | | [Hyp 1] | [None/Some/Strong] | [L/M/H] | [L/M/H] | | [Hyp 2] | [None/Some/Strong] | [L/M/H] | [L/M/H] |

Priority = High Impact + Low Effort + Little Existing Evidence

Output Template

## Hypothesis Tree Analysis

**Central Question:** [Clear, specific question] **Date:** [Date] **Owner:**
[Name]

### Hypothesis Tree Structure

[Main Question] ├── H1: [First major hypothesis] │ ├── H1.1: [Sub-hypothesis] │
└── H1.2: [Sub-hypothesis] ├── H2: [Second major hypothesis] │ ├── H2.1:
[Sub-hypothesis] │ └── H2.2: [Sub-hypothesis] └── H3: [Third major hypothesis]
└── H3.1: [Sub-hypothesis]

### Prioritized Testing Plan

| Priority | Hypothesis | Test Method | Timeline | Owner |
| -------- | ---------- | ----------- | -------- | ----- |
| 1        | [H1.2]     | [Method]    | [Time]   | [Who] |
| 2        | [H2.1]     | [Method]    | [Time]   | [Who] |

### Current Evidence Summary

| Hypothesis | Status                       | Evidence  |
| ---------- | ---------------------------- | --------- |
| [H1]       | [Confirmed/Rejected/Testing] | [Summary] |

Real-World Examples

Example 1: Low Feature Adoption

Question: "Why is our new reporting feature only used by 8% of users?"

Low Feature Adoption
├── Awareness
│   ├── Users don't know it exists
│   └── Announcement wasn't clear
├── Value
│   ├── Feature doesn't solve their problem
│   └── Existing workarounds are "good enough"
├── Ability
│   ├── Feature is hard to find
│   └── Feature is hard to use
└── Timing
    └── Users don't need reports frequently

Example 2: Churn Investigation

Question: "Why did monthly churn increase from 5% to 8%?"

Increased Churn
├── Product Changes
│   ├── Recent feature change caused issues
│   └── Performance degradation
├── Market Changes
│   ├── Competitor launched better alternative
│   └── Economic conditions changed
├── Customer Mix
│   ├── Acquired lower-quality leads
│   └── Channel mix shifted
└── Service Issues
    └── Support quality declined

Best Practices

Do

• Start with clear, specific question
• Check MECE at each level
• Get to testable hypotheses quickly (3 levels usually enough)
• Update tree as evidence comes in
• Share tree with stakeholders for alignment

Avoid

• Overlapping hypotheses (not mutually exclusive)
• Hypotheses that can't be tested
• Going too deep without testing
• Confirmation bias (seeking to prove favorite hypothesis)

Integration with Other Methods

| Method | Combined Use | | ----------------------- | --------------------------------- | | Five Whys | Go deep on confirmed hypotheses | | Jobs-to-be-Done | Frame hypotheses around user jobs | | Fogg Behavior Model | Structure behavioral hypotheses |

Resources

• The Pyramid Principle - Barbara Minto
• Bulletproof Problem Solving - Charles Conn