Everyone working in Digital Analytics knows the default setting/logic for Sessions – that they end after 30 min of inactivity. Which makes the entire concept of Sessions very artificial. I am fine with it, I find them very useful as consistent and reliable, more so than Users. But I have been asked why 30 minutes (noting that Amplitude at least uses 5 min as their default for app)?
My explanation over the years has been that, based on my observations of human behaviour, I can imagine the creator of an early digital analytics tool identifying the challenge of defining sessions, deciding to use a period of inactivity as the end point and then temporarily inserting 30 min as a placeholder – something to return to with a more correct number in the near future. And then never getting around to that with everyone else just falling into line until it became the standard because everyone uses it.
That has always amused me as feels so likely to be true.
Then yesterday I asked Claude to look into it, the sort of research I expect I would find too difficult to do myself. Much to my surprise, it actually found an answer and a source for the original 30 minutes. It wasn’t a placeholder after all, although it is a fairly dodgy piece of maths.
It is from a 1995 paper “Characterizing browsing strategies in the World-Wide web” by Catledge & Pitkow. Which I can’t access but if anyone has a Science Direct subscription, I will put a link in the comments. This paper is referenced by another paper “User Session Identification Based on Strong Regularities in Inter-activity Time” by Halfaker et al. from the Wikimedia Foundation and University of Minnesota, with the critical piece of information being:
“In their work, they reported that the mean time between user observed user events in their data was 9.3 minutes. They choose to add 1.5 standard deviations to that mean to achieve a 25.5 minutes inactivity threshold. Over time this proposed inactivity threshold has gradually been smoothed out to 30 minutes.”
See what I mean about dodgy maths? And this second paper (which I really should read in its entirety – will share a link to it as well) does go into that and the lack of evidence for 30 min being the standard, apparently it has been widely debated.
I am not suggesting we should change anything, although it would be nice if someone created a solution that quickly identifies the optimal gap of inactivity to define a session for any organisation, based on their data. But it is nice to have the official origin story (I admit to being surprised when Claude found something and then that it was able to back up findings with actual sources).
What do you think? Did everyone else already know this? Does this change anything for you?
