These types of articles should quote error ranges.
"Optimizely's homepage redesign that yielded a 46% '+/- x%' increase in new accounts"
They never seem to do so.
If you have small traffic, forget about performing A/B tests. Instead learn from the A/B tests performed by companies with high traffic, and implement their findings.