Michael's web analytics blog has moved to a new location.
Please update your bookmarks and feeds!
About
I run Monitus Tools and use this blog to share ideas about web analytics for online retailers. You can also follow me on Twitter.Subscribe
Yahoo! Store webinar today
I will do a demo today of the cart recovery service as part of the Store Optimization Series organized by the Yahoo! Store small business team. I will also briefly touch upon the upcoming release of PersonaQuest, our web personalization and targeting platform. Plus there will of course be something about web analytics - you know me!
Please consider joining me and other partners who will be showing off their cool tools:
Yahoo! Merchant Solutions Add-ons & Features Showcase — Register Now
Yahoo! Merchant Solutions Developer Partners showcase some of their most useful product features to help make your online business more successful.
1:00-3:00pm PT / 4:00-6:00pm ET
About site reviews
Live site reviews are popular sessions in conferences and webinars, but are they actually all that useful?
In a site review, an expert takes a quick look at a site and makes recommendations on the fly as to how he would improve the site. I am just doubtful that you can make any valid recommendations without knowing at least a little bit about the business behind the site. This is not a criticism of the site reviewer, but rather of the site review process itself.
Even though it may be an expert opinion, it's still only one opinion from one person. This person will be shaped by his unique world view and past experiences and will have all sorts of hidden biases. For this reason I would take the combined behavior of thousands of actual website visitors over the opinion of one person any time. Put differently, I would trust web analytics data more than anecdotal evidence. I am also doubtful that you can easily apply the recommendations of the reviewed site to your own site.
I think the best outcome for a site review session, outside of the entertainment value, is that it gets you to think about your own site. And the starting point should be to delve in web analytics.
Cross sell works
...at least in a Yahoo! Store with the optional cross-sell feature. Average order values are consistently higher for this retailer:
Note that the cross-sell traffic segment includes only visitors who have clicked on a cross-sell link. It's not an AB test, but I think it does show that Yahoo!'s cross-sell algorithm does a good job of displaying relevant cross-sell items that lead to larger order sizes. Also, as I mentioned earlier, this is only one example, but I don't see how enabling cross-sells could hurt . It's easy to implement and doesn't cost anything extra.
Event Tracking value
I am a big fan of Google Analytics' event tracking feature and I think all web analytics programs should have that capability. There is one feature that would make event tracking even more useful in my opinion.
In addition to Category, Action and Label, you can send an event value, but this is supposed to be a positive integer. It would be more powerful if event value could be used also for cost data, i.e. dollars and cents, or a number with two decimals.
One of the strengths of event tracking is the ability to integrate other data sources, so wouldn't it be great to see either the cost or value of such external data sources? We could also assign a dollar value to outbound links, downloads or referral traffic coming to our site, and compare to e-commerce revenue.
How to arrange items on your section pages using analytics data
From the excellent Neuromarketing blog comes a great post entitled "Order Effect Affects Orders" that shows that you should put your most important items first, because that's what people will click on and buy.
In other words, the item in position 0 should get more clicks than the one in position 6 on this sample section page. I thought I'd see if this is found to be true by looking directly at web analytics data.
What I did is not only track the pageview, but also the position via an onclick event using event tracking. (Side note for Yahoo! Stores: you can pretty easily determine the position of the ids in the Contents field by using the POSITION RTML operator).
What's nice is that I can now aggregate the positions across all section pages. After just a few hours of gathering data it looks like there is already a confirming pattern emerging:
Here is what I would do with the data:
- Go through your top section pages and present your best items first, instead of other listing types, such as random/date/price/alphabetical. You can now get analytics data that shows you how important order is on your site.
- If you have an e-commerce site I think you might even be able to get a sense of the dollar value of the positions. Having a dollar value helps to make analytics actionable!
- Segment this data against screen resolution to see if there is an "above or below the fold" effect. With a low screen resolution you might find that items that are not visible without scrolling are not clicked on at all.
Segment by page type
Love it when web analytics data paints a clear picture (even if it is not a pretty one).
I was happily segmenting data in the quest for actionable insights when I came across segmentation by page type (hat tip to Gabriel), specifically for e-commerce sites. Outside of the homepage you typically have two main types in an online store:
- section/category pages and
- item/product detail pages
Then pull up your top landing page report, segmented by page type and look at some conversion and revenue metrics. Where does most of your traffic land? On a section or item page? What is the conversion rate per page type?
(Click for larger versions)
Answering these questions is inherently interesting, but the main value is to further segment by paid source, e.g. Adwords or Yahoo! Search Marketing.
In one particular case, I found that most Adwords traffic was being sent to section pages (not sure if this was intentional), but the data now shows me that sending folks to item pages could yield far better results:
Note that no matter what tool you use, you should be able to get this sort of data pretty easily as long as you have proper campaign tagging enabled.
Even if your data is not as clear-cut as in this example, you can now review your PPC strategy and make changes if necessary as you are in charge of specifying the landing page URLs.
Surprising checkout error analysis
Time to look at some data after gathering checkout error data for a while.
Surprise #1
43% of all transactions had at least one error message during checkout. I expected there to be fewer although I don't know why.
Surprise #2
Seeing an error message does not necessarily mean that those people don't convert. In fact, 91.7% of visitors who saw an error message still completed the purchase. Forgot to put in the state or the email address? Just hit the back button and try again. This is what most people (thankfully for retailers) seemed to have done.
Surprise #3
Not all error messages are created equal. In particular, if the error message has to do with an incorrect CVV value, the conversion rate drops from 91.7% to just 61.9%. Or to put it another way: of all visitors who see an error message during checkout, those who don't see a CVV-related error message are almost 50% more likely to convert than those who do. Using the sample data above, it looks like over 210 transactions are lost due to CVV issues.
Perhaps this is not a huge surprise though. Here are some plausible reasons I can think of why CVV errors might lead to fewer conversions:
- If you supply the wrong CVV value or leave it blank the transaction may not go through at all (as opposed to a typo in your shipping address).
- Could it be a symptom of fraud? You need the card present to see the CVV value.
- Do visitors have trouble understanding or locating the CVV value?
- Or perhaps the error message itself (e.g. "You seem to have supplied an invalid Credit Card Verification value.") is not as helpful as others in actually helping visitors to correct the error.
One caveat
This sample data is from one source. I have seen checkout errors on other sites that are completely different, so use caution - as you always should - when using anecdotal data or drawing general conclusions.
Finally, I am wondering if your aim should be to reduce error messages in general, either by making (design) changes to your checkout pages or by limiting the amount of data you try to collect?
Demographic data in Analytics
I know I know, comparing data from different sources will give me different results, but what if the values of my report are binary: 0 or 1, yes or no, new or return visitor, female or male? In this case I don' care as much about the absolute occurrences of Zeroes and Ones, but rather the proportion between the two values. I will then optimize for the value that gets > 50%. So if I somehow knew the gender mix on my site I could change the color scheme? Sounds like a decent optimization idea, right?
Well, I looked at two recent/updated sources of demographic data: Google Ad Planner (via Jeff) and Yahoo! Web Analytics (via Dennis) and compared data from the same site, an established multimillion dollar online retailer.
From Ad Planner:
From YWA:
What I don't understand is why one tool says there are mainly men visiting the site whereas the other says just the opposite. Granted, the way the data is computed is undoubtedly different, but I am still left with the decision: do I optimize for one or the other?
Or are they both right? Perhaps I should optimize my site based on the source of the traffic: traffic from Google gets stylesheet A and traffic from Yahoo! properties gets stylesheet B.
Has anyone had luck in doing onsite personalization based on demographics, or are demographics best used in ad campaigns?
SEO ranking and Event Tracking
Love the flexibility of event tracking. Very quick to set up and I can send the data without having to specify ahead of time what the data has to look like. Event tracking reports are also conveniently separate from the standard reports, so no need to bend the system by creating "fake pageviews" to track events. In the normal web analytics paradigm I know when a page was viewed and generated a pageview, but with event tracking I can ask very specific questions about how visitors interact with a page, such as visualizing checkout error messages. This is really actionable stuff as I have data to help me design a better form.
André Scholten came up with a cool couple of filters to track SEO rankings - i.e. the search results page the referring keyword was on. Maybe there will be more user-defined variables available at some point in the future, but we can just use event tracking to do the same thing. Just use standard or custom segments.
(Click for larger)
As for the actionable part? You'll probably find that most traffic and revenue comes from being on the first page. Not much of an insight there. Try and get keywords from page 2 to page 1 by creating better targeted content or links...
