Re: [webanalytics] KPI's for Social Media Websites
- View Source
> Depending on the makeupIt's great that people are thinking about Engagement in terms of activity
> of the site, you can assign various engagement levels to each of these
> success events, in addition to time spent per visit, number of logins
> made in a particular time period, etc.
over a given period of time or even better, activity in some Recent time
frame. I'd like to suggest people begin thinking about where this kind of
metric leads longer term and the possibility of modifying it slightly before
it becomes "embedded" in the culture and leads to some frustration on the
Frequency and Cycle Time are really two very different concepts from a
Database Marketing perspective. Mixed metrics like "number of events in a
time period" are fine to start with. But when you start testing programs
based on these mixed metrics you will eventually find mixing the Frequency
component with the Recency or Time component in the same metric is
problematic. You will lose detail you can use to make the information
derived much more actionable for Marketing, Design, Content, and
In other words, instead of getting consistent, repeatable test results,
you'll get an unreliable mix of test results that are all over the place and
don't seem to have any pattern, because the basis of the metric "X events
over Y time" is a mixed bag itself. It doesn't really speak to the
Marketing issues at hand.
Think about it this way:
Frequency - the Consumption or Current Value component : represents what the
visitor / customer is worth to the company now. Measure: Sum Total of
events, is a *weak* predictive variable for future activity.
Recency - the Engagement or Potential Value component : represents what the
visitor / customer is worth to the company in the Future. Measure: Time
since Last Event, is a *strong* predictive variable for future activity.
These are two very different ideas that make a lot of sense by themselves,
but when they are forced into a single metric, lose a lot of the power they
have separately. Of course, if you're not interested in predicting the
behavior of a segment, then it doesn't really matter. But most Marketers
are *very* interested in prediction, because you can drive much higher ROI
if you can predict behavior.
Using these 2 variables separately, you get 4 segments:
High Current Value, High Potential Value : best visitors
Low Current Value, High Potential Value : up-and-coming visitors
High Current Value, Low Potential Value : former best visitors
Low Current Value, Low Potential Value : dreck visitors
and each of these segments requires a different marketing approach to
optimize the value of the segment. Think about it - what are you going to
get for response / results if you make the same offer or treat these 4 very
different groups the same way? Right. An unreliable mixed bag.
Further, you can use this same 4 segment model above to analyze any action
segmentation - campaigns, search phrases, product categories, content
groups, blog posters, uploaders - any action.
If you're still with me and care about the marketing implications of
Engagement analysis - as in, will the marketers be able to *do anything*
with the analysis once you generate it - check out this series on my blog.
If you are familiar with the concept of Recency, you can skip the Intro and
If you're not clear on the predictive powers of Recency online and you want
to see the "demo" from my eMetrics Summit 2004 presentation, start here:
The idea of "X times over Y period" is a good start, but history has proven
"X times, last time Y ago" is where you will eventually end up if what you
are concerned with is maximizing response and profitability.
Unless, of course, you are prepared to build your own regression or other
advanced model, in which case, why are you looking at "X times over Y
period" as a metric at all?
See ya at the 'Summit...
Web Site: http://www.jimnovo.com