I Presented at Biddable World
I really enjoyed presenting at Kelvin Newmann's latest conference
Biddable World. My presentation is embedded from slideshare above and
the key ideas are bullet pointed below:
- It is possible to measure the true value of something by not doing
it and seeing what changes.
- But this is high risk (lots of missed sales) and it can sometimes
be difficult to compare before/after because of seasonality.
- If you can divide the audience up into homogeneous groups so that
similar groups can be divided into test and control then you can
run a proper test like this.
- Ebay divided their audience based on geographical regions and this
approach works well because it is possible to pause/activate PPC
ads based on this dimension
- The tricky part is then identifying similar areas.
- I did this firstly using clustering algorithms and then refined it
- (I have somehow broken my install of RGoogleAnalytics so I can't
verify the code is still working. When I've got that sorted out I
will post an example)