Almost two years ago (!), I started to think about a new framework to link Time on Site and Bounces, since they are so closely related. This framework also develops a new way of calculating Time Spent on Site, using the same technique that demographers use to measure Average Life Expectancy. I wrote an introductory article here, check it out:
http://www.tomsanalytics.com/2007/09/a-new-approach-to-the-time-spent-on-site-metric/

1) If you wanted to change time on site would you want to increase it or decrease it ?

2) How would you do that (increase or decrease) ?

You are on the money talking of segmenting – no one metric, average or absolute, can have relevance out of context nor be meaningful outside of meaningful context.

To take context further, segmenting can also be done by metric (eg page views, conversion rate, etc) in addition to by dimension (campaign, keyword, landing page – by data values)

Think of visits of 2 minutes with 2 pv’s vs. those with 5 pv’s.

@Jeff
Its not the metric that is a failure – very few metrics are (something like “hits” may be).
Taking anything out of context is en route to failure.

Brian Katz – Analytics – VKI

http://actionable-analytics.com/2009/06/measuring-engagement-the-dangers-of-dwell-time/

Segmenting this measure can prove useful though – checking things like average time online for PPC visitors (and into specific campaigns) can help to detect potential clickfraud and poor quality visitors. Likewise, running this value against various organic keyword referrals, or specific landing page visitors can help to identify if some of these are performing poorly with very little time online.

Generalizations like “Longer or shorter time on site does not indicate site effectiveness” really need to be qualified, imo.

Surely the answer is to do better analysis, rather than just throw the metric out. That means segment, segment, segment.

For instance, to find out if it *does* indicate effectiveness, surely you would just segment out your quality customers from the riff-raff and see if time on site is different.

Time on site could be the same for successful users who browse lots of catalog or product pages, and unsuccessful users who browse a lot of support pages, but again, segment them out and find out what’s going on.

If it turns out that unsuccessful users do have a lower (or higher) time on site, then that gives you a clue that something is hindering their experience – maybe they’re sending time looking for something they can’t find again, or hitting a road block (poor navigation, for eg) & leaving. Segment them out and see if you can uncover what it is

In the end “time on site” is a backwards metric in my view. It’s like the page view. It creates more questions and, thus far, in terms of standards development it’s been a FAILURE. Systems/software designers and users alike cannot come to agreement on what constitutes a “video view” let alone PROVE the ULTIMATE goal: creating meaningful customer behavior.

Creating meaningful customer behavior is, in this new economy, should be (IMHO) defined as:

An organized series of prompts that ultimately leads to a sales lead or transaction.

If the time-on-site/”engagement” doesn’t actually CAPTURE customer intent (ie. is the customer’s need immediate? latent?) and then plug it into an ORGANIZED SYSTEM (call it CRM if you’d like) then the “engagement” or “time on site” is just more waste — “branded entertainment”.

Then again, DrsFosterSmith http://budurl.com/ljpl likely disagrees with my hard core view of direct response marketing… and they’re direct marketers too! Very successful ones in the pet industry who feel that “time spent on site” translates to “increased trust” in a trust-based retail niche. Trust is “enough” for them and time spent on site is a measure of it.

I chart monthly sales plus now I can compare sales numbers year to year. So I can compare month against month, because each month has a different expection of sales.

So if your sales are going up each year compared to each month, you are doing something right.