7.7Using Historical Data to Improve Estimation

The Problem with QA Estimation

QA engineers consistently underestimate testing effort because they estimate based on the happy path and forget about:

• Environment setup and troubleshooting
• Bug investigation and re-testing
• Flaky test investigation
• Blocked testing due to dependencies
• Unplanned exploratory testing triggered by suspicious behavior

Historical Calibration

Use past data to calibrate future estimates:

Historical Data (Last 10 Stories):
  Estimated test effort: 2 days average
  Actual test effort: 3.2 days average
  Calibration factor: 3.2 / 2 = 1.6x

Next story estimate: 2 days
Calibrated estimate: 2 x 1.6 = 3.2 days

Estimation by Analogy

For each new feature, find the most similar past feature and use its actual effort as the baseline:

New Feature	Most Similar Past Feature	Past Actual Effort	Adjustment	Estimate
"Add coupon system"	"Add gift card system" (Sprint 40)	5 days	+1 day (more edge cases)	6 days
"API rate limiting"	"API authentication" (Sprint 35)	3 days	-0.5 days (simpler)	2.5 days
"Mobile push notifications"	None (new territory)	N/A	Use calibration factor on raw estimate	4 x 1.6 = 6.4 days