ICE Scorer Skill

Automatically score growth experiments using the ICE (Impact, Confidence, Ease) prioritization framework.

When to Activate

This skill should activate when:

• User creates a new experiment without providing ICE scores
• User mentions "score", "prioritize", or "ICE"
• User asks "which experiment should I run first?"
• User wants to evaluate experiment backlog
• User compares multiple experiments

ICE Framework Scoring Guidelines

Impact (1-10): How much will this move the key metric?

Score 8-10: High Impact

• Affects North Star metric directly
• Expected change ≥15%
• Targets large user segment
• Critical business metric

Score 4-7: Medium Impact

• Affects important but secondary metrics
• Expected change 5-15%
• Targets meaningful user segment
• Supports key business goals

Score 1-3: Low Impact

• Affects minor or vanity metrics
• Expected change <5%
• Targets small user segment
• Nice-to-have improvement

Confidence (1-10): How certain are we this will work?

Score 8-10: High Confidence

• Strong quantitative data supporting hypothesis
• User research validates the problem
• Similar experiments succeeded elsewhere
• Multiple sources of evidence
• Detailed rationale (>100 characters)

Score 4-7: Medium Confidence

• Some supporting data or research
• Analogous experiments showed promise
• Logical reasoning with limited evidence
• Moderate rationale (50-100 characters)

Score 1-3: Low Confidence

• Speculative or gut feeling
• No supporting data
• Untested assumption
• Minimal rationale (<50 characters)

Ease (1-10): How easy is this to implement?

Score 8-10: High Ease

• < 1 day of work
• No engineering required, or minimal changes
• No external dependencies
• Can be done with existing tools

Score 4-7: Medium Ease

• 1-2 days of work
• Some engineering work required
• May need design support
• Uses existing infrastructure

Score 1-3: Low Ease

• 2 days of work

• Significant engineering effort
• Requires design and multiple teams
• Needs external resources or new tools

Scoring Process

When scoring an experiment:

1. Read the experiment file from the experiments folder

2. Analyze the hypothesis components:

   • Proposed change
   • Target audience
   • Expected outcome (look for specific percentages)
   • Rationale (check length and evidence quality)

3. Evaluate Impact:

   • Is this a North Star metric or secondary metric?
   • What's the expected percentage change?
   • How many users will this affect?
   • Consider the experiment category (acquisition, activation, etc.)

4. Evaluate Confidence:

   • How much evidence supports the hypothesis?
   • Is there user research or data mentioned?
   • How detailed is the rationale?
   • Are there comparable experiments?

5. Evaluate Ease:

   • Estimate implementation time
   • Does it need engineering? Design? External resources?
   • How complex is the proposed change?
   • Look for keywords: "redesign" (low ease), "copy change" (high ease)

6. Calculate total ICE score: Impact × Confidence × Ease

7. Interpret the score:

   • 700+: Critical Priority - implement immediately
   • 500-699: High Priority - strong candidate
   • 300-499: Medium Priority - good experiment
   • 150-299: Low Priority
   • <150: Very Low Priority - deprioritize

8. Update the experiment JSON with ICE scores

9. Move to pipeline if score ≥ 300

Scoring Examples

Example 1: Onboarding Progress Indicators

Experiment: Add progress indicators to 5-step onboarding flow

Analysis:

• Impact: 7 - Activation is important, expected 15% increase
• Confidence: 6 - User research supports it, but not tested yet
• Ease: 9 - Simple UI element, <1 day of work
• Total: 378 - Medium-High Priority

Reasoning:

• Impact: Activation is a key metric but not the only North Star
• Confidence: User research provides evidence but no previous tests
• Ease: Adding progress bar is straightforward UI work

Example 2: Social Proof on Pricing Page

Experiment: Add customer logos and testimonials to pricing page

Analysis:

• Impact: 7 - Affects acquisition and conversion
• Confidence: 8 - Strong industry evidence for B2B social proof
• Ease: 9 - Design change only, no engineering
• Total: 504 - High Priority

Reasoning:

• Impact: Pricing page is high-traffic, affects key conversion
• Confidence: Multiple case studies show 10-15% improvement
• Ease: Simple asset placement, quick implementation

Example 3: Complete Platform Redesign

Experiment: Redesign entire user interface

Analysis:

• Impact: 9 - Could affect all metrics significantly
• Confidence: 4 - No data supporting specific improvements
• Ease: 2 - Months of work, multiple teams
• Total: 72 - Very Low Priority

Reasoning:

• Impact: Broad changes could have major impact
• Confidence: Too vague, no specific hypothesis about what will improve
• Ease: Massive undertaking, not a growth "experiment"

Keywords to Watch

Low Ease indicators:

• redesign, rebuild, refactor, overhaul, migration, infrastructure

High Ease indicators:

• copy change, button, color, image, text, email, simple

High Confidence indicators:

• "data shows", "research indicates", "we tested", "similar experiment"

High Impact indicators:

• North Star, conversion, activation, retention, revenue
• Specific percentages (e.g., "15% increase")
• Large user segments

Output Format

When providing ICE scores, explain your reasoning:

ICE Score Analysis for: [Experiment Title]

Impact: [score]/10
Reasoning: [Why this score based on metric importance, expected change, audience size]

Confidence: [score]/10
Reasoning: [Why this score based on evidence, data, research quality]

Ease: [score]/10
Reasoning: [Why this score based on time, resources, complexity]

Total ICE Score: [Impact × Confidence × Ease] = [total]

Priority: [Critical/High/Medium/Low/Very Low]
Recommendation: [What to do with this experiment]

[If score >= 300:]
✓ Moving to pipeline based on strong ICE score

Integration with Commands

This skill works automatically when:

• /experiment-create completes - offer to score immediately
• /hypothesis-generate creates ideas - suggest preliminary scores
• User asks about prioritization

Continuous Learning

After experiments complete:

• Compare predicted Impact vs actual results
• Adjust scoring calibration based on outcomes
• Learn patterns for better Confidence scoring
• Refine Ease estimates based on actual time taken