Three minutes



As a boy, 3 minutes was how long I cooked my eggs. As an adult, it seems to be the accepted industry benchmark for website dwell time in Google Analytics. Anything less is met with horror; anything longer commands respect and reverence. There is a strong belief that duration correlates directly to a quality experience, yet when you look deeper into the stats you see the curve which should look like this:





Actually looks like this:


The stats that result from these visits are served as a fairly meaningless average. Another way to describe dwell time is basically the average of people who spent a short amount of time and those who spent longer. Did anyone enjoy their visit? Who knows!


If you think about your own behavior online it is immediately clear that the idea of longer duration being a measure of success is only sometimes true. For a news website or anything with content, a longer visit is a good indicator that the website is doing something right. But for a utility site, arguably, faster is better. The trouble is that most sites are serving both needs at the same time. The Trainline website that I want to tell me train times in seconds also wants me to buy tickets - a considerably longer process. The Retrobike forum, better than it sounds and where browsing vintage mountain bikes can take longer than you think, also needs to show me my new instant message, instantly.


So how can we look beyond dwell time and is it ever a useful indicator of a good visit?


Here are three approaches:

      1. Use Google Analytics (most of our clients do), but go beyond the headline stats (most people don’t).


In some cases, filtering by visitor activity may provide insight. For example:

a) segment your pages into those where you believe a long visit is a good one and those where the reverse is true. What do the stats look like now? The same logic applies to bounce rate - home pages should have a low bounce rate because their job is to be the shop window, but 'contact us' pages are fine to bounce all day long.

b) explore the visits by mobile devices and look for differences to desktop visits. Can you make a case for a good mobile visit being a fast one? Could you consider targeting dwell time based on device - longer for desktop; shorter for mobile?


      2. Ask the people who use your website (face to face).

The challenge of a representative sample size notwithstanding, I have yet to be in a focus group where the tenth group said much different to the first - so this doesn’t need to be an exhaustive, expensive exercise, and is useful at three key points:

a) when re-imagining a website, card-sorting exercises are effective at seeing what tools and content customers want and how they would arrange them. Ask your customers to create a sitemap for you; it will almost always be different to how you imagined it.

b) during the design phase a clickable prototype will allow your customers to try various user journeys before your site is built. The key thing here is to think in terms of journeys rather than a series of website templates.

c) For live sites, ask people to perform some important tasks while you observe them. And consider tasks that are not in what you might call in the positive direction. How do I cancel something? How do I change my address? How do I phone you up?


      3. Create a new approach using multiple data sets.

What other factors should be considered when looking for insight? What was the weather doing? When was pay-day? That surge in dwell time might have had more to do with a rainy day and pay cheques than anything else. But therein lies the opportunity. Every email I subscribe to lands on the same day every week, to the same rhythm every month. I can’t think of a single one that arrives on pay-day when the temptation to shop is greatest. And one of the best campaigns I can remember used the weather in London in Celsius each day as the Summer sale percentage discount.


Whichever approach you take, the key is to persuade your colleagues that stats like dwell time aren't that meaningful by themselves. That conversation might take more than 3 minutes.


If Draw can help, please get in touch.


Related News