Re:A 14kb page can load much faster than a 15kb page
by sokoloff on Thursday August 25, 2022 @10:41AM
(#32594291)
We ran some A/B tests where we intentionally slowed our site down (at the webserver page generation level) by set percentages (0%, 10%, 25%, and 33% IIRC). Those would be against a baseline page generation time of around 2000ms, so the slowdowns were not small.
All tests showed a drop in traffic, conversion, and average session value indistinguishable from zero. Our hypothesis is that our site (where a converting visitor is on average designing and ordering a custom product) is one where the difference between 30 minutes and 31 minutes is not meaningful and a lot of the time is spent in the web-based document editor. It would not generalize to a "give me a quick search result", "sell me an item from a warehouse shelf", or "let me argue with someone on a forum".
This was of course disappointing, because we ran this test as a sanity check after having done an extensive project to improve the page speed across the main conversion funnel. The project was technically successful, but a business zero (probably for the same hypothesis as above), so we decided to test in the other direction, because that's easier to implement than to force faster pages.
All tests showed a drop in traffic, conversion, and average session value indistinguishable from zero. Our hypothesis is that our site (where a converting visitor is on average designing and ordering a custom product) is one where the difference between 30 minutes and 31 minutes is not meaningful and a lot of the time is spent in the web-based document editor. It would not generalize to a "give me a quick search result", "sell me an item from a warehouse shelf", or "let me argue with someone on a forum".
This was of course disappointing, because we ran this test as a sanity check after having done an extensive project to improve the page speed across the main conversion funnel. The project was technically successful, but a business zero (probably for the same hypothesis as above), so we decided to test in the other direction, because that's easier to implement than to force faster pages.