Kevin Hillstrom: MineThatData

Exploring How Customers Interact With Advertising, Products, Brands, and Channels, using Multichannel Forensics.

July 26, 2009

Conversion Rates Across Time

Customers. They mess up all of our conversion rate metrics, don't they?

Two weeks ago, I visited Zappos on a Monday night. I copied thumbnails of twelve pair of shoes that I liked, pasted the thumbnails into an e-mail, and sent the e-mail to my wife.

The next morning, my wife reviewed the thumbnails, and gave her thumbs up to one pair of shoes.

On Wednesday, I purchased the pair of shoes that my wife liked.

From a customer standpoint (i.e. from my standpoint), this process was a complete success.

From a metrics standpoint, this could be considered a flawed process. We have two visits with just one conversion. Your software tool might consider this to be a unique visitor with 100% conversion. Your software tool might consider this to be two visits yielding a 50% conversion rate.

From a Multichannel Forensics standpoint, we try to take some of the mystery out of conversion rate. Instead of looking specifically at the mystery of conversion rate, we look at how a customer with specific attributes behaves in the future.

Let's look at a very simple example. We take all visitors from the month of May, and categorize them into one of several segments, based on the depth of website activity during May:
  • Homepage or Landing Page Visit Only.
  • Multiple Pages Visited.
  • Shopping Cart Abandoned.
  • Purchaser.
Clearly, this segmentation strategy can be expanded upon (source = PPC or e-mail ... or looking at visitors from April or March, or looking at new buyers vs. multi-buyers, you get the picture).

The next step is to create a "grid". For all customers who visited the site in the month of May, we categorize them based on behavior during the month of June:
  • No Subsequent Visit.
  • Homepage or Landing Page Visit Only.
  • Multiple Pages Visited.
  • Shopping Cart Abandoned.
  • Purchaser.
Our job is to populate the grid, categorizing customers based on May behavior and June behavior. The grid looks something like this:



Home or
Shop

NoLandingMulti-CartPur-

VisitPagePagesAbandonchaser!!!
Homepage/Landing Page78.2%10.3%5.2%4.3%2.0%
Multiple Pages49.6%12.5%22.4%7.3%8.2%
Shopping Cart Abandoned32.3%15.7%24.4%13.7%13.9%
Purchaser25.8%20.4%29.4%9.3%15.1%

This table is pretty simplistic, but it gives us a lot of "business intelligence", if you will.

For instance, take a look at the "No Visit" column. The converse of this column is the "Re-Visit Rate". The table suggests that 21.8% of homepage/landing page visitors re-visited.

Now look at the Shopping Cart Abandoned customer. The "Re-Visit Rate" for these customers is 67.7%. That's important. If you know that customers who abandon a shopping cart are unlikely to come back to your website, then shopping cart abandonment is a REALLY bad thing. If you know that two-thirds of the visitors will come back over the course of the next thirty days, well, you think about abandonment with a little bit less fear.

In my Multichannel Forensics projects with online marketers, the important word is always "context". In the table above, the customer who drills down into the site and then leaves is not necessarily considered a failure. More than half of the visitors in that segment come back to the website next month, with 8% converting. The data suggest an element of "engagement" that may not be easily conveyed by a 3% conversion rate or a 47% shopping cart abandonment rate.

Context is derived by knowing the state of a customer in pre/post timeframes. When we step outside of measuring campaigns, focusing instead on measuring customer behavior across time, we obtain a different level of customer understanding.

8 Comments:

At 7:01 AM , Blogger David said...

This problem could have been simplified early in the research process when the user had to copy thumbnails of 12 different pairs of shoes. Why not provide the consumer with a simple technology solution that provides a project-based shopping experience. Persistent shopping can be accomplished through our application that allows multiple users to share product information without having to leave the active research/shopping process. We capture the service information that has previously been unavailable for the reason you had to copy/paste the thumbnails!

 
At 3:40 PM , Blogger Jeff_Molander said...

Hey, David... I was hoping you'd show up and comment on the experience here.

Kevin... David's solution is the real deal. I hope you may have a chance to check it out at some point soon. Elegant, simple and very much needed. I think you just proved the point!

 
At 3:47 PM , Blogger Kevin said...

Good for David!

Now let's see some folks create the analysis outlined in this post --- I'm asked to do this kind of stuff, very easy for those of us with SAS/SPSS background, very hard for the average web analytics analyst.

 
At 7:04 AM , Anonymous Michael Feiner said...

Hi Kevin,


I love the framework you propose above. It provides actionable insight and creates that essential context you talk about. Thank you.

Reading through the post I did wonder how I would get a web analytics tool to provide me the correct data.

I can get the data using tools such as Omniture SiteCatalyst. But not sure about the likes of Google Analytics.

You mention SPSS and SAS in your comment. Are you importing all the web analytics data into a BI tool and running the analysis there?

Would be intrigued to hear more about the practical steps of this process.

Thank you again.

Michael Feiner

 
At 7:52 AM , Blogger Kevin said...

Personally, I import web analytics data into SPSS and do all of my work there. Some tools, like Google Analytics, make that process harder, though not impossible.

I use SPSS because I can then analyze customer behavior any way I want, without limitation, including analysis via statistical algorithms.

 
At 8:22 AM , Anonymous Michael Feiner said...

Yeah, that certainly would make things easier.
Without a tool like SPSS it could be extremely difficult.

Even with Omniture it would require implementing an additional time stamp or using one of their customised data processing plugins (VISTA Rule).

Not easy, unfortunately.

But as I said before, it is a good idea worth pursuing.

 
At 11:00 AM , Anonymous Matthew Tod said...

This is a nice analysis, I love anything that nails the abandoned baskets myth!

But the analysis is made less good by the need to make the time period a month. This is an arbitrary requirement of most analytics tools - you have to specify a time period rather than the next session. The scenario painted at the start of this post took place over three days / two sessions - how much better it would be if the analysis could be based upon sessions. No criticism of the technique, just the way the vendors kindly process the data and make it available to us.

Matthew Tod
Logan Tod & Co.

 
At 11:12 AM , Blogger Kevin said...

Since I export the data to my own platform and write my own code, I get to pick my own time intervals. The time intervals can be defined however I want them to be defined ... in this case, I used 30 days, for other clients, the right timeframe is 1-2 days, for some, 1-2 months, for others, 1-2 years. Each business is different!

 

Post a Comment

Links to this post:

Create a Link

<< Home