Loading Form...
Jul 28, 2010 | 8 minute read
written by Linda Bustos
This is the fourth in a series of posts based on the material covered in our latest webinar: Taking Your Site Performance to The Next Level With Optimization Testing.
The ultimate goal of a Web optimization project is to improve revenue and profits driven through your site. Success is measured by tracking changes in KPIs (key performance indicators), most commonly AOV (average order value), conversion rate, days to purchase, bounce rate, cart and checkout abandonment. But don't forget the metrics that matter -- revenue, revenue per visit, and if you can track it, contribution margin. Remember, your conversion rate can skyrocket if you drop prices below cost and give free shipping, but you’ll quickly be out of business!
Your strategy may involve optimizing your website’s usability, or experimenting with product offers, promotions and messaging. Through site testing, you can quantify the impact of site changes and promotions against a “do nothing” control.
Though there are many approaches to site optimization, I am going to outline a process that uses Web analytics to discover and prioritize opportunities for improvement.
What are you looking for in your analytics?
Checkout Funnel Abandonment
The checkout is the best place to start your optimization. Visitors who make it all the way to your checkout are far more likely to convert than those who hit your home page, which means success is measured in completed sales rather than “reduced bounce rate” or “increased click through to a category page.” It’s also the process that can cause the most anxiety (giving personal and credit card information) and usability problems (forms, etc.)
It’s important that you configure your conversion goals in your Web analytics tool, including every step in the funnel process (from shopping cart review to thank you) so you can see visually where people abandon your checkout.
The page with the highest percentage drop off should be top priority. After you’ve tested your checkout to the point of diminishing returns, you can move on to other content pages, such as...
Top Entrance Pages
Your home page is not your only home page. When your visitor lands on your site through a search engine, online ad, referring website or email link, every “landing page” acts as a home page. These deep pages may receive more entrances than your home page, so they deserve more attention.
The terms entrance pages and landing pages are sometimes used interchangeably, but note that "landing pages" also refer to pages created specifically for a campaign or customer segment to land on, which are not part of your regular site structure. Your Entrance Page or Landing Page report may include both types of “landing page" - regular entrance pages and special landing pages.
Top Content
There’s also a difference between top entrance pages and top content. It's helpful to know which pages are viewed most frequently on your site, they are not necessarily the pages visitors start with. For example, you may have a product that is often searched, or that is featured on the home page and in cross-sell recommendations. Use reason when determining which pages are actually important. Segment by referral page and consider if the page contributes to your goal of improved revenue and profit. (E.g. many people view your Shipping Policy page, but that doesn’t mean it’s top priority for testing.)
Top Exit Pages
Where are customers leaking? Are these pages important, and can they be tweaked to keep folks on your site longer?
It’s dangerous to put faith in averages, as they conceal many insights. For example, using the pivot table tool in Google Analytics, I find that the site average is 27%. But some pages have unusually high bounce rates for returning visitors coupled with low bounce rates for new visitors. This warrants a closer look at these pages to figure out why.
Another example is segmenting by traffic source. You may be getting a lot of referrals to your “Hannah Montana Live in Concert” CD page from the search term “Hannah Montana concert,” but if searchers are looking for concert tickets, not merchandise, it will drag up your bounce rate and drag down your conversion rate.
Often organizations can’t move forward with testing because they aren’t using their Web analytics in any actionable way. If you don’t have an in-house Web analysis capability, you have the options to outsource analysis and testing to an agency or contract consultant, to add an in-house web analyst to your team or to train yourself on Web analytics principles. Two great books to get you up to speed are Avinash Kaushik’s Web Analytics an Hour a Day and Analytics 2.0.
Once you’ve weeded through your reports and found good candidate pages for improvement, it’s time to prioritize them. As I mentioned before, it makes sense to optimize your checkout process first – but sometimes this is not possible, especially if you are using a 3rd party payments solution that you can’t touch.
Aside from the checkout process, you’ll have to use your noggin to determine which pages require immediate attention. Priority should be given to pages that add the most to your bottom line. Pick “big fish” pages that have room to improve (especially high-margin products). You may be tempted to rescue pages that have the worst bounce rates or conversion rates, but avoid them if a substantial improvement to the page is only a drop in the bucket to your business.
You also want to consider what you can’t find from your analytics. Consult with your sales team. What products are they going to push in the next 6-12 months? These might be pages that historically haven’t received a lot of traffic but are expected to be big sellers next quarter or during the holidays, such as new product. (In this case, analytics data is irrelevant.) This is important if your product catalog has high turnover. It’s important to involve your merchandising team too, as they can use your support in testing promotions (banners, calls to action, home page layouts, cross-sell/upsells, etc.)
It’s a bit of a chicken-and-egg situation with testing elements and testing methods. If you read my last post about selecting A/B or multivariate testing, you’ll recall that your testing method depends in part on what things you want to test. But before you decide what elements to test, it helps to know which method you will use.
If you are indifferent between the two methods (you have enough traffic to support multivariate and no barriers to test radical redesigns), your decision comes down to gut feel and experience on what makes most sense to test. It helps to have a background in web design, usability and marketing, as your gut feel likely influenced by reasonable industry best-practices. Unfortunately, we can’t eradicate gut feel completely from conversion optimization, but at least we will be able to attach data to our educated guesses through testing.
Most pages have so many possible variables to test you can get overwhelmed, so focus on variables that are likely to influence each other. For example, this landing page has no lack of page sections to play with:
You can’t test every element at the same time and still run an efficient experiment. In this case, you may believe that the headline has a big impact on bounce rate along with the hero shot (images evoke strong emotions – positive or negative). The main call to action and the callout are also influence whether people click through to the next page. In this case, you may select a multivariate test with 4 variables.
You will also need to determine your “recipe budget” – how many recipes you can afford given your current traffic, conversion rates and expected improvement. This impacts how many variations of each variable you can test in reasonable time.
Estimating test length
Google's Website Optimizer Duration Calculator can help you figure out how long a test will take considering the number of recipes (test versions) you want considering daily page views and current conversion rate, and expected conversion improvement (break out your crystal ball for that one).
If your test duration is too long, you need to scale back on your number of recipes, or adjust your expected conversion improvement (the higher the estimate, the longer it takes to gather enough data to validate it).
Knowing a ballpark range of how long your test will run is important, since you may only have a short window before you need to roll out changes. You also don’t want to have your test overlap known traffic spikes or product launches that may affect visitor behavior and bias your results.
You may be tempted to stop your test and declare a winner before your test has reached statistical significance after a certain number of pre-determined observations. If you jump the gun, you may invalidate your test and actually implement a losing version on your site.
Sometimes you won't find a statistically significant winner. Don't worry, the test was not a loss. You learned that your test elements have no impact on site performance. That's worth something.
One last thing…
If your product assortment and merchandising is the problem, you can tweak your site all you want but still not persuade visitors to buy. Make sure you consider all aspects of your business that are impacting revenues and profits.
Introduction to A/B and MVT4 Reasons for Site TestingChoosing Between A/B or Multivariate Test Design