Experiment Design

Solution Track — Step 4 of 4. Designs minimum viable tests with clear success/iterate/kill rules.

Inputs Required

An initiative with hypothesis and target metric (from .agents/mkt/prioritized-initiatives.md + .agents/mkt/targets.md)
OR: User describes what they want to test

Output

.agents/mkt/experiment-[name].md

Quality Gate

Before delivering, verify:

Executable in ≤2 weeks
Costs <10% of full initiative budget (or $0 for internal-only changes)
Success threshold is a specific number (not "significant improvement" or "we'll know it when we see it")
Sample size is sufficient (checked against table below) — or test type adjusted if insufficient

Chain Position

Previous: mkt-funnel-planner | Next: Scale (if success) / mkt-initiative-generate (if kill)

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
`prioritized-initiatives.md`	mkt-initiative-prioritize	INTERVIEW. Ask what to test.
`targets.md`	mkt-funnel-planner	INTERVIEW. Ask for baseline metrics.

Optional Artifacts

Artifact	Source	Benefit
`product-context.md`	mkt-copywriting	Test design context

Read .agents/mkt/prioritized-initiatives.md and .agents/mkt/targets.md if they exist. If not, interview the user:

"What are you trying to test? What's your hypothesis?"
"What metric are you measuring? What's the current baseline?"
"What would success look like? What would make you stop?"

You need at minimum: a hypothesis, a target metric, and a baseline number.

Step 1: Choose Test Type

Type	When	Rigor
A/B Test	Can split traffic randomly; enough volume	High — controls for time
Before-After	Can't split (infrastructure change, whole-site change)	Medium — time effects confound
Cohort	Testing on a specific segment first	Medium — segment differences
Pilot	New capability; need qualitative signal first	Low — directional only

If unsure: check sample size (Step 3). Not enough volume for A/B in 2 weeks → use Before-After or Pilot.

Step 2: Define Decision Rules

Be specific. Vague thresholds → endless "let's give it one more week."

Outcome	Threshold	Action
Success	[metric] ≥ [number]	Scale: roll out fully
Iterate	[metric] between [X] and [Y]	Change ONE variable: [specify which]. Rerun for [duration]. Max 2 iterations.
Kill	[metric] < [number] after [full duration]	Archive learnings. Return to `mkt-initiative-generate`.

Step 3: Sample Size Check

For A/B tests — minimum visitors PER VARIANT:

Baseline Rate	20% Relative Lift	50% Lift
1%	~40,000	~6,400
5%	~8,000	~1,300
10%	~4,000	~640
25%	~1,600	~260

Quick math: [Your daily traffic] × [test days] ÷ 2 = visitors per variant. Compare against table.

If insufficient for A/B: use Before-After, or make a bigger change (aim for 50%+ lift instead of 20%).

See references/sample-size-guide.md for full tables.

Artifact Template

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

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

# Experiment: [Name]

**Initiative:** [from prioritized list]

## Design

| Field | Value |
|-------|-------|
| Hypothesis | If [action], then [metric] changes from [baseline] to [target], because [reason] |
| Test Type | A/B / Before-After / Cohort / Pilot |
| Duration | [≤2 weeks] |
| Budget | [$X or $0] |
| Primary Metric | [metric name] |
| Baseline | [current value] |

## Decision Rules

| Outcome | Threshold | Action |
|---------|-----------|--------|
| Success | [metric] ≥ [X] | [Scale how] |
| Iterate | [metric] between [X-Y] | Change: [one specific variable]. Rerun: [duration]. |
| Kill | [metric] < [X] after [duration] | Archive. Next: [what to try instead] |

## Sample Size Check

- Daily traffic: [N]
- Test duration: [days]
- Visitors per variant: [N × days ÷ 2]
- Required for [X%] lift at baseline [Y%]: [from table]
- **Sufficient?** Yes / No → [if no, adjusted approach]

## Next Step

[When to check results. What to do with each outcome.]

Worked Example 1: A/B Test

# Experiment: Paid Targeting Lookalike Test

**Date:** 2026-03-13
**Skill:** mkt-experiment
**Initiative:** Restore + Refine Paid Targeting

## Design

| Field | Value |
|-------|-------|
| Hypothesis | If we use a 1% converters lookalike audience, then paid signup rate increases from 1.2% to ≥3%, because the lookalike matches our ICP better than current broad targeting |
| Test Type | A/B — split paid budget 50/50 between current and lookalike |
| Duration | 10 days |
| Budget | $500 (redirecting existing $50/day spend) |
| Primary Metric | Paid visitor-to-signup rate |
| Baseline | 1.2% |

## Decision Rules

| Outcome | Threshold | Action |
|---------|-----------|--------|
| Success | ≥2.5% signup rate | Roll out lookalike to 100% of paid budget |
| Iterate | 1.5-2.5% | Narrow lookalike from 1% to 0.5%, add interest-based exclusions. Rerun 10 days. |
| Kill | <1.5% after 10 days | Targeting isn't the primary issue. Test homepage messaging next. |

## Sample Size Check

- Daily paid traffic: ~300 visitors
- 10 days: 3,000 total → 1,500 per variant
- At 1.2% baseline, need ~1,300 per variant for 50% lift detection
- **Sufficient?** Yes — 1,500 > 1,300. Can detect a 50%+ relative lift (1.2% → 1.8%+).

Worked Example 2: Before-After

# Experiment: Homepage Social Proof Restoration

**Date:** 2026-03-13
**Skill:** mkt-experiment
**Initiative:** Restore Homepage Social Proof

## Design

| Field | Value |
|-------|-------|
| Hypothesis | If we restore customer logos + testimonial to homepage hero, then bounce rate drops from 52% to ≤42%, because the redesign removed trust signals that reduced visitor confidence |
| Test Type | Before-After — change affects all visitors |
| Duration | 7 days after deployment |
| Budget | $0 (copy/design work only) |
| Primary Metric | Homepage bounce rate |
| Baseline | 52% |

## Decision Rules

| Outcome | Threshold | Action |
|---------|-----------|--------|
| Success | Bounce ≤ 42% | Keep changes. Explore further trust signal additions. |
| Iterate | 42-48% bounce | Add more proof: case study snippet, user count, or G2 rating badge. Rerun 7 days. |
| Kill | >48% after 7 days | Social proof alone isn't enough. Run full Homepage A/B (old vs new) from parked list. |

## Sample Size Check

- Daily homepage visitors: ~800
- 7 days: 5,600 visitors
- Before-After with 52% baseline: 5,600 is ample to detect a 10-point bounce rate change
- **Sufficient?** Yes

Loop-Back Protocol

When an experiment concludes, follow the appropriate path:

Kill Decision

Append learnings to the experiment file under ## Post-Mortem
Check: Is the root cause (.agents/mkt/root-cause.md) still valid?
- Yes → Route to S1: Run mkt-initiative-generate with constraint: "[Failed approach] did not work because [reason]. Generate alternatives that avoid this failure mode."
- No → Route to P1: Problem may have shifted. Run mkt-diagnosis to redefine.
Update .agents/mkt/prioritized-initiatives.md — move killed initiative to "Kill" status with reason.

Iterate Decision

Change exactly ONE variable from the original test
Maximum 2 iterations allowed before triggering Kill protocol
Document iteration rationale in experiment file under ## Iterations

Scale Decision

Document final results in experiment file under ## Results
Move to full execution — no further testing needed for this initiative
Run mkt-attribution to map the scaled initiative to KPIs

After the Experiment

Result	Do This
Success	Scale. Set full targets via `mkt-funnel-planner`. Celebrate briefly, then test the next initiative.
Iterate	Change ONE variable. Document what and why. Max 2 iterations before escalating to kill/pivot.
Kill	Archive in the experiment file: what you tested, what happened, what you learned. Return to `mkt-initiative-generate` with the new learning.

References

references/experiment-templates.md — Templates for landing page, email, ad creative, feature flag tests
references/sample-size-guide.md — Full sample size tables and common mistakes