Root Cause Analysis

Problem Track — Step 3 of 3. Reviews evidence against hypotheses and identifies the root cause.

Inputs Required

• Hypotheses from .agents/mkt/hypotheses.md
• User-provided data or data files for each hypothesis

Output

• .agents/mkt/root-cause.md

Quality Gate

Before delivering, verify:

• Every hypothesis has a verdict: Confirmed, Rejected, or Inconclusive
• Every verdict cites a specific number or data point (not "seems like" or "probably")
• Root cause statement includes gap percentage (e.g., "explains ~60% of gap")
• Every Inconclusive item has a concrete next step (what data, where, who)

Chain Position

Previous: mkt-hypothesis | Next: mkt-initiative-generate

---

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
hypotheses.md	mkt-hypothesis	STOP. "Run mkt-hypothesis first to develop testable hypotheses."

Optional Artifacts

Artifact	Source	Benefit
product-context.md	mkt-copywriting	Industry context for data interpretation

Read .agents/mkt/hypotheses.md. Quote the ranked hypotheses.

Then ask: "What did you find for each hypothesis? Share data, screenshots, reports, CSVs, or describe what you saw."

If user provides data files, read and analyze them directly — don't ask the user to summarize data you can read yourself.

If user hasn't gathered data yet:

• Use WebSearch for publicly available evidence: "[competitor] launch [date range]", "[industry] [metric] trend [year]", "[tool name] outage [date]"
• Point them to the specific sources listed in the hypotheses file

---

Step 1: Verdict Per Hypothesis

For each hypothesis:

Verdict	Criteria	What to Write
Confirmed	Evidence matches the "then" clause and supports the "because"	"[Data point] confirms: [what it shows]. Mechanism supported."
Rejected	Evidence contradicts the "then" clause	"[Data point] shows [opposite of prediction]. Cause eliminated."
Inconclusive	Evidence is ambiguous or insufficient	"[What we found]. Need: [specific additional data], from [source], owned by [who]."

Be direct. "The data doesn't support this" is useful. "It's hard to tell" is not.

---

Step 2: Root Cause Statement

Synthesize confirmed hypotheses:

Root Cause 1 (~X% of gap): [Cause], evidenced by [specific data]. [Mechanism]. Root Cause 2 (~Y% of gap): [Cause], evidenced by [specific data]. [Mechanism]. Unexplained (~Z%): [What's still unknown]. Next step: [what to investigate].

Gap percentages should roughly sum to 100%. They don't need to be precise — "~60%" is fine. The point is to show which cause matters most.

---

Step 3: Next Steps

Situation	Action
Clear root cause(s) identified	"Run mkt-initiative-generate to brainstorm solutions targeting: [root cause]"
Inconclusive items remain	"Before generating solutions, gather: [specific data]. Then re-run this skill."
Multiple root causes, different sizes	"Address Root Cause 1 first (~X% of gap). It's the highest-leverage fix."

---

Artifact Template

On re-run: rename existing artifact to root-cause.v[N].md and create new with incremented version.

---
skill: mkt-root-cause
version: 1
date: {{today}}
status: draft
---

# Root Cause Analysis

**Problem:** [problem statement]

## Verdict Table

| # | Hypothesis | Verdict | Evidence | Gap |
|---|-----------|---------|----------|-----|
| 1 | [Short name] | Confirmed / Rejected / Inconclusive | [Specific data cited] | ~X% |
| 2 | ... | ... | ... | ... |

## Root Cause Statement

**Root Cause 1 (~X% of gap):** [Cause], evidenced by [data].
[One sentence explaining the mechanism.]

**Root Cause 2 (~Y% of gap):** [Cause], evidenced by [data].
[One sentence explaining the mechanism.]

**Unexplained (~Z%):** [Description]. Next data needed: [what], from [where].

## Next Step

Run `mkt-initiative-generate` targeting:
1. [Root cause 1 — specific aspect to solve]
2. [Root cause 2 — specific aspect to solve]

---

Worked Example

User provides data for the signup decline hypotheses:

• "Checked DB vs GA4 — both show the same drop. Tracking is fine."
• "Paid traffic conversion dropped from 3.5% to 1.2%. Organic is still at 4.1%."
• "Homepage bounce rate increased from 35% to 52% for all traffic."

# Root Cause Analysis

**Date:** 2026-03-13
**Skill:** mkt-root-cause
**Problem:** Weekly signups 200 instead of 350 (43% gap)

## Verdict Table

| # | Hypothesis | Verdict | Evidence | Gap |
|---|-----------|---------|----------|-----|
| 1 | Tracking broken | **Rejected** | GA4 and DB both show same drop (200/week). Tracking is working correctly. | 0% |
| 2 | Ad targeting reduced quality | **Confirmed** | Paid conversion dropped from 3.5% to 1.2% (66% decline). Organic stable at 4.1%. Paid traffic is 60% of total. | ~55% |
| 3 | Homepage redesign reduced conversion | **Confirmed** | Bounce rate up from 35% to 52% across ALL sources. Even organic visitors are bouncing more, suggesting the page itself is the problem. | ~35% |

## Root Cause Statement

**Root Cause 1 (~55% of gap):** Ad targeting change brought low-intent visitors, evidenced by paid conversion dropping 66% while organic held steady. The broader targeting reaches people outside the ICP.

**Root Cause 2 (~35% of gap):** Homepage redesign reduced trust/clarity, evidenced by bounce rate rising 17 points across all traffic sources. The new page likely lacks the social proof or clear value prop of the old design.

**Unexplained (~10%):** Minor variance, possibly seasonal. No action needed.

## Next Step

Run `mkt-initiative-generate` targeting:
1. Paid targeting — restore or improve audience quality
2. Homepage — restore trust signals and clarify value proposition

---

References

None needed — this skill's framework is simple. The hypothesis-guide.md in mkt-hypothesis covers the upstream methodology.