AI Hypothesis Refiner
Transform rough experiment ideas into clear, testable, and measurable hypotheses using AI refinement.
Why Hypothesis Quality Matters
A poorly written hypothesis leads to ambiguous experiments, unclear success criteria, and unactionable learnings. Well-structured hypotheses ensure every test generates valuable insights - regardless of whether the experiment succeeds or fails.
The If/Then/Because Format
All hypotheses use a structured format:
"If we [action] for [audience segment], then [metric change], because [reasoning]."
This ensures every hypothesis includes four essential components:
- Action - What you're changing
- Segment - Who you're targeting
- Metric change - What measurable outcome you expect
- Reasoning - Why you believe this will work
What AI Refinement Does
Our AI refinement engine analyzes your hypothesis and improves it across four scored dimensions:
| Dimension | What It Measures | Score Range |
|---|---|---|
| Specificity | Is the action described precisely enough? | 1–10 |
| Testability | Can it be objectively tested within a timeframe? | 1–10 |
| Measurability | Is the expected metric change quantifiable? | 1–10 |
| Time-bound | Are there clear temporal boundaries? | 1–10 |
Combined Clarity Score
The four dimension scores are averaged into an overall Clarity Score (displayed as a percentage). Aim for 80% or higher before launching an experiment.
How to Use the Hypothesis Refiner
- Write your initial hypothesis in the If/Then/Because text area - even rough ideas work
- Click "Refine with AI"
- Review the refined hypothesis, clarity scores, and structured breakdown
- The refined version auto-populates the hypothesis field
- Review any warnings (e.g., missing success criteria, vague audience)
- Check the sample size guidance if provided
💡 Tip: The refiner works best when you also provide context - target metric, funnel stage, and experiment goal. This enables more relevant refinement.
Understanding the Analysis Results
After refinement, you'll see:
- Clarity Score - Overall quality percentage
- Dimension Breakdown - Visual scores for specificity, testability, measurability, and time-bound
- Structured Components - Action, Segment, Metric, and Reasoning parsed separately
- Feedback - AI commentary on the hypothesis quality
- Warnings - Issues that need attention (shown in amber)
- Sample Size Guidance - Estimated sample needed for statistical significance
Before and After Examples
Before: "Make the signup page better" After: "If we simplify the signup form by reducing fields from 5 to 3 (email, password, name) for first-time visitors from organic search, then signup completion rate will increase by 15%, because shorter forms reduce friction and abandonment at the critical first-value moment."
Before: "Test social proof" After: "If we add a rotating testimonial carousel with 5 customer reviews above the fold on the pricing page for B2B visitors, then demo request rate will increase by 12%, because social proof reduces purchase anxiety for high-consideration enterprise purchases."
