Comment by hansvm | Hacker News Reader

Comment by hansvm | Hacker News Reader

Parent Full thread

Dwell time is challenging to pull out from server logs without constant interaction. E.g., if a landing page has a low conversion rate, your solutions are going to be pretty different if you're trying to solve users immediately bouncing, reading the whole thing and bouncing, or scrolling to some particular piece of content and bouncing.

There's also an implicit assumption in your question that every notable user event has an associated server-side request. If page changes and whatnot are being handled completely client-side after the first request you might need a little extra monitoring just to get the same level of detail you would otherwise have from traditional server logs.

I could be wrong here, but I think the big feature in client-side analytics isn't that they offer some sort of special data or insight; the selling point is that anyone can add them to basically any web hosting solution (including CDNs you don't control or random WordPress sites) with a single copy/paste. It optimizes ease of adoption _at the cost of_ probably better analytics.

Elaborating on that "better analytics" point -- the web is varied enough that even if you want to track your users' activities it's really hard to interpret client-side data correctly. E.g., in the landing page example above, if your heartbeats stop hitting the server did the user navigate away or did they go to another tab for a bit? In that same example, are your heartbeats going through because the user is reading and pondering your content or because they've gone to make a sandwich (you can disambiguate this a bit if you also keep track of something like the last time the user did something, but it's not perfect)? On top of that, you don't get metrics from people with JS disabled, with certain DNS blocklists, people with slow connections, etc.

Is dwell time a meaningful/valuable metric to track? I consistently open links in tabs I keep open for a long time and I suspect other people might do the same. Like you said, it's really hard to interpret this data, so why bother trying?

I think you're right about the ease of adoption aspect. My take is that we do it because we've always done it - primarily because the cost/benefit ratio has justified it. We've normalized the idea that it's not only acceptable but also prudent to study and analyze the detailed behavior of end-users. The consensus seems to be that not adding end-user tracking is similar to "leaving money on the table".

On reflection, I'm starting to feel that it's perhaps not an acceptable set of norms to accept without a justified need. Some metrics seem almost voyeuristic and most imply an approach to content marketing that only snake-oil salesmen should consider. The use of analytics itself implies a certain potential for inauthentic or intentionally manipulative content. Optimizing content to have as much mass-appeal as possible is perhaps more problematic than we initially thought.

Maybe it's time we rethink the norm in the first place. Perhaps it's better to not know everything.

> Is dwell time a meaningful/valuable metric to track?

Like most other questions of this nature, the answer is "maybe." Throwing out one possible scenario: We know some people keep tabs open for a long time. Does our dwell time data reflect that? If not then that's a red flag that we have work to do (maybe a memory leak from some framework lifecycle issue makes our page prone to crashes or being manually closed, maybe the content is such hot garbage that nobody thinks it's worth even keeping a background tab open with our site, ...).

> Like you said, it's really hard to interpret this data, so why bother trying?

Something can have a lot of value without that value being easy to extract. Like...I'm sure Amazon has detailed metrics about every mouse and keyboard event associated with any pixel with the gall to glance sideways at the shopping cart. It's probably hard to use correctly, but when a 1% increase can pay for hundreds or thousands of dedicated engineers and scientists the ease of interpretation starts to not matter as much.

> I'm starting to feel that it's perhaps not an acceptable set of norms to accept without a justified need

I'll need to think on it further, but on the surface that seems like a reasonable baseline. If you don't at a bare minimum know enough about your business and the data in question to be able to anticipate some actionable way you might actually need the data you want to collect then you'll probably do more harm than good for yourself and your users by collecting it.

It looks like Umami does some of that. But I found that Google Analytics does none of it, and in fact year-over-year seems to cut back on the amount of useful information. The main advantage now is they show referring search terms, which servers logs would show, but they broke that on purpose with their search result self-referring links.

Google analytics provides large amounts of useful information to Google to help them sell ads.

The information available to the sites is not as easy to interpret.

Reply on news.ycombinator.com