ab-test

A/B Test Documentation Skill

Purpose

Create and maintain structured documentation for A/B tests, capturing the hypothesis, variants, measurement approach, timeline, results, and conclusions. This ensures tests are properly documented and learnings are preserved.

When to Use

Invoke this skill when any of these conditions occur:

1. User mentions running an A/B test - "I'm running an A/B test on...", "We're testing two versions..."
2. User is comparing variants - "Should we try A or B?", "I want to compare these two approaches..."
3. User wants to measure something - "How do we know which is better?", "I want to test if this works..."
4. Explicit user request - When user invokes /project-docs:ab-test command

How to Document an A/B Test

Arguments

• test-name (required): Name of the A/B test (used as document title and default filename)
• file-path (optional): Where to save the document. Defaults to ./{test-name}.md in current working directory

Workflow

1. Determine file path:

   • If file path provided, use it
   • Otherwise, create ./{test-name}.md in current working directory
   • Sanitize test name for filename (lowercase, replace spaces with hyphens)

2. Check if file exists:

   • If file exists, read current content and present to user
   • Walk through updating each section
   • If new file, start fresh with all sections

3. Walk through each section interactively:

   Ask questions one at a time, waiting for the user's response before moving to the next question.

   For new tests:

   • "When does/did this test start?" (Test Date)
   • "What's your hypothesis? What do you expect to happen and why?" (Hypothesis)
   • "What's the name of your control variant (the existing/baseline)?"
   • "Describe the control variant:"
   • "What's the name of your treatment variant (the new/experimental)?"
   • "Describe the treatment variant:"
   • "How are users assigned to test vs control groups?" (Group Assignment - audience segmentation, randomization method, percentage split, etc.)
   • "How will you measure success? What metrics matter?" (Measurement)
   • "When will you evaluate the results?" (Results Timeline)

   For existing tests (updating):

   • Show current content for each section
   • Ask "Do you want to update this section?" for each
   • Focus especially on Results and Conclusions sections:
     • "What were the actual results?"
     • "What conclusions did you draw? What's the decision?"

4. Write the document using the format below

Document Format

# A/B Test: {test-name}

## Test Date
{when the test started/starts}

## Hypothesis
{what we expect to happen and why}

## Variants

### Control: {control-name}
{control description}

### Treatment: {treatment-name}
{treatment description}

## Group Assignment
{how users are assigned to control vs treatment - randomization, segmentation, percentage split}

## Measurement
{how success will be measured, what metrics}

## Results Timeline
{when results will be evaluated}

## Results
{actual results - may be empty initially}

## Conclusions
{what we learned, decision made - may be empty initially}

Guidelines

Test Date

• Be specific: "2025-01-15" or "Week of January 15, 2025"
• Can include end date if known: "January 15-29, 2025"

Hypothesis

• State clearly what you expect to happen
• Include the "why" - the reasoning behind the expectation
• Example: "We expect the larger CTA button to increase click-through rate by 10-15% because it will be more visually prominent and easier to tap on mobile."

Variants

• Control: The existing behavior or baseline
• Treatment: The new approach being tested
• Describe each variant clearly enough that someone could recreate it
• Include relevant details: copy, design, behavior, etc.

Group Assignment

• How are users randomly assigned?
• What's the percentage split? (e.g., 50/50, 90/10)
• Any audience segmentation or targeting criteria?
• How is consistency maintained (same user always sees same variant)?
• Example: "50/50 random split using Optimizely. Users assigned by cookie on first visit. Only including users from US and Canada."

Measurement

• Specify the primary metric
• List any secondary metrics
• Define how you'll collect the data
• Note sample size requirements if relevant
• Example: "Primary: Click-through rate on CTA. Secondary: Time on page, bounce rate. Measured via Google Analytics events."

Results Timeline

• When will you have enough data to decide?
• What's the minimum sample size needed?
• Include planned review date
• Example: "Review after 2 weeks or 1,000 visitors per variant, whichever comes first. Planned review: January 29, 2025."

Results

• Record actual data observed
• Include confidence intervals if relevant
• Note any unexpected observations
• Can be updated multiple times as data comes in

Conclusions

• What did you learn?
• What decision was made?
• What will you do next?
• Any follow-up tests needed?

Example Document

# A/B Test: Homepage CTA Button Size

## Test Date
January 15, 2025

## Hypothesis
We expect the larger CTA button (treatment) to increase click-through rate by 10-15% because it will be more visually prominent and easier to tap on mobile devices, where 60% of our traffic comes from.

## Variants

### Control: Standard Button
- Size: 120x40px
- Text: "Get Started"
- Color: Blue (#0066CC)
- Current conversion rate: 3.2%

### Treatment: Large Button
- Size: 160x56px
- Text: "Get Started Free"
- Color: Blue (#0066CC) with subtle gradient
- Added micro-copy below: "No credit card required"

## Group Assignment
- 50/50 random split
- Assignment via Optimizely, cookie-based on first homepage visit
- All traffic included (no geographic or device filtering)
- Users see consistent variant across sessions

## Measurement
Primary metric: Click-through rate on homepage CTA
Secondary metrics:
- Signup completion rate
- Time to first click
- Mobile vs desktop CTR difference

Data collection: Google Analytics events + Mixpanel funnel

## Results Timeline
Review after 2 weeks or 5,000 visitors per variant.
Planned review date: January 29, 2025
Statistical significance threshold: 95%

## Results
After 2 weeks (5,200 visitors per variant):
- Control CTR: 3.2%
- Treatment CTR: 4.1%
- Lift: 28% (p < 0.01)
- Mobile lift was higher (35%) than desktop (18%)

## Conclusions
The larger button significantly outperformed the control. The "No credit card required" micro-copy likely contributed to the lift beyond just size.

Decision: Roll out the larger button to 100% of traffic.

Follow-up test: Test the micro-copy separately to isolate its effect.

Important Notes

• Ask questions one at a time - Don't overwhelm the user with all questions at once
• For updates, show current content - Let users see what's there before deciding to change
• Results and Conclusions can be empty initially - Tests are often documented before they complete
• Preserve the structure - Keep all sections even if some are empty (use "TBD" or "Pending")
• Be specific - Vague tests lead to vague conclusions