微信扫一扫Hypothesis Development & Data Planning
Problem Track — Step 2 of 3. Turns logic tree leaves into ranked, testable hypotheses with data requirements.
Inputs Required
- Logic tree from
.agents/mkt/diagnosis.md
Output
.agents/mkt/hypotheses.md
Quality Gate
Before delivering, verify:
- [ ] Every hypothesis follows If/Then/Because format
- [ ] Every "then" clause names a specific metric or observable data point
- [ ] Every hypothesis has a named data source (not "check analytics" — which tool, which report)
- [ ] Hypotheses are ranked by testability (easiest to test first)
Chain Position
Previous: mkt-diagnosis | Next: mkt-root-cause
Before Starting
Step 0: Product Context
Check for .agents/mkt/product-context.md. If missing: INTERVIEW. Ask the user 8 product questions (what, who, problem, differentiator, proof points, pricing, objections, voice) and save to .agents/mkt/product-context.md. Or recommend running mkt-copywriting to bootstrap it.
Required Artifacts
| Artifact | Source | If Missing |
|----------|--------|------------|
| diagnosis.md | mkt-diagnosis | STOP. "Run mkt-diagnosis first to define the problem with a logic tree." |
Optional Artifacts
| Artifact | Source | Benefit |
|----------|--------|---------|
| product-context.md | mkt-copywriting | Better hypothesis framing |
Read .agents/mkt/diagnosis.md. Quote the problem statement and list the leaf nodes.
Step 1: Form Hypotheses
For each leaf of the logic tree:
If [this cause is true], then we'd see [specific observable evidence], because [mechanism that explains why].
What Makes a Strong Hypothesis
| Element | Weak | Strong | |---------|------|--------| | If | "onboarding is bad" | "onboarding emails are being spam-filtered since the domain migration on Jan 15" | | Then | "conversion drops" | "day-1 email open rate dropped below 20% (was 45%)" | | Because | "emails matter" | "the new sending domain has no reputation, triggering ESP spam filters" |
The "then" must be something you can look up in a specific tool. The "because" must explain the mechanism — it's what you learn from if the hypothesis is wrong.
Step 2: Map Data Requirements
For each hypothesis:
| Hypothesis | Deciding Data Point | Confirming Evidence | Rejecting Evidence | Source (Tool → Report) | Owner | |------------|-------------------|--------------------|--------------------|----------------------|-------| | [If/Then/Because] | [Single data point that decides] | [What you'd see if true] | [What you'd see if false] | [e.g., "GA4 → Acquisition → Traffic by source"] | [Person] |
WebSearch directive: Before finalizing, search "[topic] common causes" OR "[metric] drop reasons [industry]" to check if you're missing any high-probability hypotheses. Also search "[competitor name] [recent change]" if external factors might be at play.
Step 3: Rank by Testability
Rank hypotheses by: (a) how quickly you can get the data, AND (b) how much of the gap it would explain if confirmed.
| Priority | Hypothesis (short name) | Data Available Now? | Potential Gap Explained | Test Time | |----------|------------------------|--------------------|-----------------------|-----------| | 1 | [Easiest + highest impact] | Yes / Partial / No | ~X% | [Hours/Days/Weeks] | | 2 | ... | ... | ... | ... |
Logic: Test what you can test TODAY first. Eliminate hypotheses fast. Don't spend weeks on #4 when #1 takes an hour to check.
Artifact Template
On re-run: rename existing artifact to hypotheses.v[N].md and create new with incremented version.
---
skill: mkt-hypothesis
version: 1
date: {{today}}
status: draft
---
# Hypotheses
**Problem:** [problem statement from diagnosis]
## Hypotheses (Ranked by Testability)
### 1. [Short name] — Priority: HIGH
**If** [cause], **then** [observable evidence], **because** [mechanism].
- **Deciding data:** [specific data point]
- **Source:** [Tool → Report → Metric]
- **Owner:** [person/team]
- **Confirming:** [what you'd see if true]
- **Rejecting:** [what you'd see if false]
- **Potential gap explained:** ~X%
### 2. [Short name] — Priority: HIGH
[Same format]
### 3. [Short name] — Priority: MEDIUM
[Same format]
## Next Step
Gather data for hypotheses #1 and #2. Then run `mkt-root-cause` to analyze findings.
Worked Example
From diagnosis: Weekly signups are 200 instead of 350 (43% gap). Inflection: homepage redesign + ad targeting change 8 weeks ago.
# Hypotheses
**Date:** 2026-03-13
**Skill:** mkt-hypothesis
**Problem:** Weekly signups 200 instead of 350 (43% gap, 8 weeks)
## Hypotheses (Ranked by Testability)
### 1. Tracking broken on new homepage — Priority: HIGH
**If** the homepage redesign broke the signup tracking pixel, **then** GA4 shows a sudden drop in recorded signups coinciding with the deploy date while server-side signup records remain stable, **because** the new page template may not include the tracking script.
- **Deciding data:** Compare GA4 signup events vs. database signups for the last 8 weeks
- **Source:** GA4 → Events → sign_up AND production database signup count
- **Owner:** Engineering
- **Confirming:** GA4 shows drop but DB signups are flat
- **Rejecting:** GA4 and DB both show the same drop
- **Potential gap explained:** ~100% (if tracking, not real decline)
### 2. Ad targeting change reduced quality — Priority: HIGH
**If** the ad targeting change brought lower-intent visitors, **then** paid traffic volume stayed flat but paid visitor signup rate dropped, **because** broader targeting reaches people less likely to convert.
- **Deciding data:** Signup rate by traffic source (paid vs organic) before/after change
- **Source:** GA4 → Acquisition → Traffic source → Conversion rate
- **Owner:** Marketing (paid team)
- **Confirming:** Paid conversion rate dropped; organic stable
- **Rejecting:** Both paid and organic dropped equally
- **Potential gap explained:** ~50%
### 3. Homepage redesign reduced conversion — Priority: MEDIUM
**If** the new homepage is less clear or trustworthy, **then** homepage-to-signup conversion rate dropped for ALL traffic sources equally, **because** the new design may lack social proof or clear value proposition.
- **Deciding data:** Homepage → signup conversion rate, by traffic source, before/after
- **Source:** GA4 → Pages → Homepage → Next page flow
- **Owner:** Product/Design
- **Confirming:** All-source conversion dropped on homepage specifically
- **Rejecting:** Conversion dropped only for paid traffic (→ targeting issue, not homepage)
- **Potential gap explained:** ~40%
## Next Step
Gather data for #1 (takes 30 minutes — compare GA4 vs DB). Then #2 (segment paid vs organic in GA4). Run `mkt-root-cause` with findings.
References
- references/hypothesis-guide.md — If/Then/Because templates by problem type