A/B Testing Uptime Per-User Calculator Examples
Worked examples showing how uptime levels, traffic splits, test duration, and user volume affect A/B test availability exposure.
These examples show how the calculator translates uptime differences into average user downtime and aggregate lost user-hours. Each scenario uses a different traffic allocation, test length, or availability gap.
50/50 test with a modest uptime gap
A product team runs a 30-day experiment with 10,000 users and routes half of users to each variant.
Input Summary
Variant A uptime
99.90%
Variant B uptime
99.50%
Variant A traffic
50%
Test duration
30 days
Total test users
10,000
Calculation Breakdown
- 1Convert duration30 * 24720 hours
- 2Calculate A downtime720 * (1 - 0.999)0.72 hours per A user
- 3Calculate B downtime720 * (1 - 0.995)3.60 hours per B user
- 4Weight average downtime0.72 * 0.50 + 3.60 * 0.502.16 hours per user
- 5Aggregate exposure10,000 * 2.1621,600 lost user-hours
Result Summary
Aggregate exposure
21,600 lost user-hours
A/B Testing Uptime Per-User Calculator
Average uptime is 99.70%, with 2.16 expected downtime hours per user and 21,600 lost user-hours.
Conservative 10% rollout of a lower-uptime variant
A team exposes 10% of 50,000 users to Variant A at 99.00% uptime while the established Variant B operates at 99.95% uptime for 14 days.
Input Summary
Variant A uptime
99.00%
Variant B uptime
99.95%
Variant A traffic
10%
Test duration
14 days
Total test users
50,000
Calculation Breakdown
- 1Convert duration14 * 24336 hours
- 2Calculate A downtime336 * (1 - 0.9900)3.36 hours per A user
- 3Calculate B downtime336 * (1 - 0.9995)0.168 hours per B user
- 4Weight average uptime99.00 * 0.10 + 99.95 * 0.9099.855%
- 5Calculate total exposure(3.36 * 0.10 + 0.168 * 0.90) * 50,00024,360 lost user-hours
Result Summary
Calculate total exposure
24,360 lost user-hours
A/B Testing Uptime Per-User Calculator
The expected average downtime is 0.487 hours per user, or about 29.2 minutes, across the 14-day test.
Long test with nearly equal availability
A platform runs a 90-day, 70/30 test for 200,000 users. Variant A has 99.98% uptime and Variant B has 99.95% uptime.
Input Summary
Variant A uptime
99.98%
Variant B uptime
99.95%
Variant A traffic
70%
Test duration
90 days
Total test users
200,000
Calculation Breakdown
- 1Convert duration90 * 242,160 hours
- 2Calculate A downtime2,160 * (1 - 0.9998)0.432 hours per A user
- 3Calculate B downtime2,160 * (1 - 0.9995)1.08 hours per B user
- 4Weight average downtime0.432 * 0.70 + 1.08 * 0.300.6264 hours per user
- 5Calculate total exposure200,000 * 0.6264125,280 lost user-hours
Result Summary
Calculate total exposure
125,280 lost user-hours
A/B Testing Uptime Per-User Calculator
Average uptime is 99.971%, while expected downtime is about 37.6 minutes per user over 90 days.
Short 80/20 launch validation
A team tests Variant A at 99.80% uptime against Variant B at 99.60% uptime for 7 days, assigning 80% of 5,000 users to A.
Input Summary
Variant A uptime
99.80%
Variant B uptime
99.60%
Variant A traffic
80%
Test duration
7 days
Total test users
5,000
Calculation Breakdown
- 1Convert duration7 * 24168 hours
- 2Calculate A downtime168 * (1 - 0.998)0.336 hours per A user
- 3Calculate B downtime168 * (1 - 0.996)0.672 hours per B user
- 4Weight average uptime99.80 * 0.80 + 99.60 * 0.2099.76%
- 5Calculate total exposure(0.336 * 0.80 + 0.672 * 0.20) * 5,0002,016 lost user-hours
Result Summary
Calculate total exposure
2,016 lost user-hours
A/B Testing Uptime Per-User Calculator
The test estimates 0.4032 downtime hours per user, or about 24.2 minutes, with 2,016 lost user-hours overall.
How to Read Your Results
Average per-user uptime is the traffic-weighted availability across everyone in the test, not a separate measured uptime for every individual.
Expected downtime per user is an average across both variant assignments during the full test period.
A positive B versus A difference means Variant B has higher uptime; a negative value means Variant A has higher uptime.
Total lost user-hours aggregate time across users and should not be read as the number of affected users.
Compare results using the same duration and traffic assumptions when evaluating alternative test plans.
Assumptions & Important Notes
- Each scenario treats uptime as stable throughout the stated test period.
- User assignment follows the selected traffic split.
- All users are assumed to have equal potential exposure to downtime.
- Results estimate time availability only and do not measure failure severity or business outcomes.
Related Examples
Frequently Asked Questions
Can a 10% rollout still produce many lost user-hours?
Yes. A limited rollout can still create substantial aggregate exposure if the test is long, the user population is large, or the variant has much lower uptime.
Why does test duration affect downtime but not weighted uptime?
Weighted uptime is a percentage based on variant uptime and traffic shares. Duration converts the unavailable percentage into hours of expected downtime.
Does a 50/50 split always give the most reliable comparison?
A 50/50 split gives both variants equal weight in the average result, but the suitable allocation depends on the experiment's objectives and risk tolerance.
How should I compare a short test with a long test?
Review both uptime percentages and expected downtime hours. The longer test usually produces more downtime hours at the same uptime level.
Ready to calculate your own result?
Use the live calculator with your own inputs, timing, and preferences.