Wednesday, August 8, 2007

Omniture Discover 2.1: just like playing Lego

Back from vacation, I was eager to see if our Discover v2.1 license had been turned on. Sure thing, it was there, boldly saying "Launch Discover 2.1".

What's changed?

  • Licensing: At eMetrics I heard some people complaining they had to pay for the D2 upgrade even if they were already licensed for D1.5. In fact, D1.5 was based on a per-user/per-reportsuite model, while the new licensing is per-seat and for all your report-suites. So unless you only have one analyst and one report suite, the new licensing model is much more interesting.
  • Interface: The interface is totally changed and enhanced with new visualizations, including the traditional pies and bars, but also funnels and the amazing 3D site analysis. Hey! Omniture too can do fancy little gadgets like a time-slicer to easily pick date ranges just like Google Analytics :)
  • Reports/Segments/Metrics/Comparisons: the intuitive interface is straightforward:
    1. Pick a report: the same from SiteCatalyst, including all of your custom vars appropriately named
    2. Chose a segment: including pre-configured ones and those defined in SiteCatalyst, or define new ones using the Segment Builder
    3. Pick some additional metrics, including custom ones or define new ones trough the Metrics Builder
    4. Compare all of those with different segments or time frames
  • Slicing & dicing: playing with the data, drilling down (nested metrics), filtering or arranging the columns is a snap, plus the graphs gets updated dynamically and is easy to tweak (like pulling down a slice from a pie chart).
  • Other cool features: site analysis is the amazing 3D visualization of your site navigation along with the metrics of your choice, dynamic path flow lets you navigate the site and drill down in any way you want and the virtual focus group (a feature similar to ClickTale, TapeFailure and such playback tools) are just way cool!

First launch

After firring up Discover v2.1 trough Java Web Start and a couple of quick downloads, the new interface came up. A drastic change from v1.5! Here's some caveats to watch for if you have difficulties starting it:
  • Although the latest version of the required Java runtime is JRE 6 Update 2, make sure you are using JRE 5.11. If you don't, Discover will warn there might be some problems... and there are. See KB #1870.
  • If you have Google Web Accelerator installed, disable it while using Discover. Otherwise you will end up with the Java Web Start window looping continuously, trying to download the files from Omniture. Also refer to KB #1276.

Advanced analysis capabilities: real stories from the trenches

Like Craig Danuloff reported a couple of months ago when he first saw D2, SiteCatalyst "as a reporting plus a little analysis and the other (D2) as a hard-core analysis tool". Here's two examples (a bit simplified) of things I couldn't do with SiteCatalyst:
  • Complex funnel analysis: Imagine you have a car insurance quoter where you define the driver, the location, the car, and get a price (4 steps funnel). We have several transactions that share some of the steps (like "Add a driver" to an existing insurance policy), and some steps that have multiple alternative versions (like getting a quote for two cars or two drivers at once). SiteCatalyst allows you to build a funnel but each step must be a unique page, which made it very complex (even impossible) to address all our complex needs. With D2 page groups I was able to easily solve this problem, plus I can now dynamically change my segmenting (dynamic segmentation is a feature that WebSideStory HBX Analytics is very strong at).
  • Nested metrics: doing "as you go" analysis is often much more efficient then having to extract a bunch of reports and play with them in Excel pivot tables... I wanted to validate if a geotargeting feature would be efficient on our site. Using the nested metrics capabilities, I was able to easily drill down, switch segments and see if what the automated targeting would have chosen would effectively be what the real user ended up picking (comparing GeoSegmentation and a custom variable holding the real user's selected region). For example, the system think you are in Quebec, and you really are: great! The system think you are from Ontario, but you end up doing a quote saying you live in Quebec because Gatineau,QC and Ottawa, ON are just beside each other, but in two different provinces (check the map!). Are we comfortable with that? In the end, I was able to conclude the solution would be accurate 99% of the time in some regions, and would be very poor in others. Knowing that gives us actionable data and we can take the right business decision. (I lived a similar situation in a previous job, where a dealer locator feature would automatically recommend the nearest dealer - on the other side of a river that required a two hour drive to get across a bridge! What was the ratio of false-positive? What was the consequences on conversion rates? We never knew because we didn't have any web analytics tools).
  • For another real-life example, see A case about reporting vs. analysis

My take

In my view, once you get beyond reporting, once you've reached a certain level of analysis capabilities and brought the web analytics culture to your organization, the next logical step is to use powerful tools like Discover. But one thing remain: out of the critical activities of a web analyst, exploring new strategies is one of the most difficult and sadly, the one we have the less time to do. Discover is a really good product, but it won't do any magic: better make sure you have the right people, the right process and enough time to really use it as it should.