If responses times are as low as 50 ms, it usually isn't a problem when it become 150 ms: people don't notice Similarly, if response times are 2 seconds, it doesn't matter that it becomes 2.1 seconds. However, when the response time is 400 ms, then it is a noticable difference when it becomes 500 ms. People notice that: the site feels more sluggish.

Now whenever I shop at Amazon, I browse around for a bit. It's not that a more sluggish site would cause me to abandon a purchase midway in the purchasing process, but it's rather that I might get annoyed at the sluggishness while browsing and postpone my purchase or take it elsewhere, where I can browse the available products at my leisure.

I think that is what they mean, but if the number is just based on assumptions about who intends to make a purchase, that also means the number is highly disputable. However, I think Amazon has enough data available to simply correlate response times with sales rates over a statistically signifcant period and statistically signifcant amount of sales. In that case, the number is pretty undisputable: if anyone has this kind of data, it's Amazon.

Basically, you just run the time to sale and how much cash you're processing and correlate that to time. True, it's not exactly causal, but Google found the same with I believe the quote is "1 second delay in search means $1M lost in revenue"

:) Thanks (all) for responding. I guess I never realised how much of amazons purchases must be spur-of-the-moment. (NOTE: I order stuff from amazon - but usually in groups as I'm shipping it to australia - so all of my stuff is planned - i'm going there to buy something)

But - if anyone has the metrics to prove a figure like that - it would be amazon.