Understanding Report Metrics in Convert
Mastering Convert Reports: Read Numbers with Clarity & Confidence
THIS ARTICLE WILL HELP YOU
-
Identify the core performance metrics in every Convert report
-
Pick the statistical engine that fits your traffic and risk profile
Performance Metrics
-
Visitors – Unique people bucketed into each variation.
Example: 5 000 visitors on Variation A means 5 000 different individuals saw that version. -
Conversions – Total goal completions.
Example: 100 “Buy Now” clicks from 1 000 visitors = 100 conversions. -
Conversion Rate – Conversions ÷ Visitors.
Example: 100 / 1 000 = 10 % conversion rate. -
Total Conversions – If multiple conversions per visitor are allowed, this is the aggregate count.
Example: 50 visitors × 2 purchases each = 100 total conversions. -
Revenue (when enabled) – Sum of all transaction amounts tracked for the variation.
-
Revenue per Visitor (RPV) – Revenue ÷ Visitors.
Example: $5 000 / 1 000 visitors = $5 RPV. -
Improvement – Percentage lift or loss v. baseline.
Example: Control 10 % → Var A 12 % ⇒ +20 % improvement (not +2 %).
Statistical Confidence Indicators
Frequentist
-
Confidence – Certainty that the observed difference is real. Wait for 95 %+ before deciding.
-
Statistical Significance – Flag showing whether confidence ≥ your preset threshold.
-
P-value – Probability of seeing the data if the variants were identical. < 0.05 is conventionally “significant”.
-
Confidence Interval – Range containing the true lift.
Example: +15 % [+10 %, +20 %].
Bayesian
-
Chance to Win – Probability a variation is best. Most teams ship at 95 %+.
-
Expected Loss – Average % conversion you might forfeit if you pick this variant and it is not actually the best.
-
Credible Interval – Bayesian version of the confidence interval (interpretation is the same).
Test Progress Indicators
-
Sample Size – Visitors collected vs. required.
Example: 5 000 / 10 000 visitors (50 %). -
Statistical Power – Probability the test will detect a true effect. Aim for ≥ 80 %.
-
Minimum Detectable Effect (MDE) – Smallest lift your current traffic can reliably spot.
-
Test Duration – Elapsed runtime; keep every test live for at least one full business cycle (7-14 days for most sites).
Warning Icons
-
⚠️ Low Sample Size – Fewer than ~5 000 visitors per variation. Let the test run.
-
⚠️ Not Yet Significant – Results still within the margin of error. Collect more data.
-
✅ Test Complete – All sample, power, and confidence criteria satisfied; safe to implement the winner.
Choosing Your Statistical Method
Method | Best For | Key Metrics | Notes |
---|---|---|---|
Fixed-Horizon Frequentist | Classic A/B with preset n | Confidence, p-value, power | Don’t peek early. |
Sequential Testing | Very high-traffic sites needing fast calls | Always-valid confidence, sequential bounds | You can look anytime without α-inflation. |
Bayesian | Most users | Chance to Win, Expected Loss, Credible Interval | Intuitive; resistant to early noise. |
Quick-Reference Checklist
-
Sample Size – ≥ 5 000 visitors per variation
-
Confidence / Chance to Win – ≥ 95 %
-
Test Duration – ≥ 7–14 days
-
Statistical Significance – Yes (Frequentist only)
Red flags
-
Wild swings or flip-flops in the first days
-
Lift claims > 50 % on tiny traffic
-
“Significant” results with < 1 000 visitors
-
Tests shorter than one complete business cycle
Patience pays. Resist the urge to act on early excitement; let the data mature before you call the winner.