Like most (all) companies with websites, improving the conversion rate of paid traffic is a priority for us. When this optimization was, we sent all traffic to our home page and we tested our homepage content and layout to identify the best composition to convert visitors to customers. We use A/B Split testing; Multi Variable Testing and Taguchi Testing on our landing page in the past, but now use Taguchi Testing most often for its unique ability to identify significant improvements quickly.

The Taguchi Method was developed 50 years ago and has been employed with great success to optimize automobile and product manufacturing. Genichi Taguchi devised a methodology for testing multiple variables in the manufacturing process using a subset of the number of tests required by traditional testing methods. In large-scale production scenarios, every test is expensive so the Taguchi Method proved extremely valuable. In addition to reducing the number of sequential test, the Taguchi Method discriminates between signal and noise and estimates the positive or negative value of each element (factor) in each alternative (level). The Taguchi Method works so well in manufacturing that it is now being applied to traditional advertising, Internet PPC advertising and Landing Pages.

We use a slight variation of the Taguchi Method in our Landing Page Optimization testing to figure out which elements influence visitors to sign up to test our software. The process is relatively simple and has 4 steps.

Step 1: Select elements on the landing page that we believe will influence sign-up

Step 2: Create alternatives for each element

Step 3: Create test "recipes" that combine these elements according to the Taguchi Methodology

Step 4: Set up and run a concurrent multi variable test using the Offermatica ASP testing framework

Step 1: Select elements on the page that we believe will influence sign-up
Taguchi testing proves its value when a large number of elements are tested simultaneously.

We chose to iteratively improve our existing home page. The first step was to select elements on the page from which we could create alternatives.

Our default page looked like this:

We selected 5 elements to vary. The elements included the:

• Headline
• Graphic at the top of the page
• Content of the Bulleted List next to the free trial button
• The Large Body Text
• Color of the Navigation Bar at the bottom of the page
• A Credibility Line of text above the bottom navigation bar that describes Offermatica's history and credentials

Step 2: Create an Alternative for Each Element
Alternatives should be levels where possible, existing vs. not existing if the element does not have levels and finally, strong alternative treatments.

The alternatives were:

Headline:
Existing: "Lower Your Customer Acquisition Cost"
New: "Scientific Marketing Methods to Boost Conversion"

Graphic at Top of Page:
Existing: Square, colored "bubbles"
New: No Graphic

Bulleted List Contents:
Existing:

• Start testing on your site now!
• 30-Day Free Trial
• No software to buy

New:

• Use Taguchi method to increase conversion by 2-3x
• Have statistical confidence in marketing decisions
• Improve average order value without new software

Large Body Text Area:
Existing:

New:

Navigation Bar Color:
Existing: Blue with white text
New: Light green with black text

Credibility Line:
Existing: None
New: "Since 1996, Offermatica as Fort Point Partners Inc., built custom ecommerce solutions for Best Buy, Nike, Intel, Barclays Global Investors and many others. Offermatica reflects the most successful efforts from our 50+ projects."

Step 3: Create "Recipes" that Combine these Elements According to Taguchi
Using Taguchi you can achieve the same test results with fewer tests and a smaller test population.

With 5 elements in 2 alternatives (default and new) we would have to create 32 unique pages to test every combination. We estimated that it would take more than six months to have enough conversions on each of the 32 "recipes" to obtain an accurate reading if we were to run all of the recipes.

The Taguchi Method uses linear algebra to determine a subset of these recipes that can be tested to estimate the impact of each element in each alternative. In our case we had eight recipes.

Step 4: Set up and Run a Concurrent Multi Variable Test Using Offermatica
Offermatica took about two hours to instrument the landing page and upload the new content.

In order to test the recipes, we added mboxes to the landing page where the content would change. This involved cutting and pasting a line of java script around the existing content and took less than an hour. In Offermatica, we uploaded the new content by pasting the HTML, text or images into form fields in the Offermatica administrative interface. We created the eight recipes and instructed the Offermatica engine to send 1/8 of all traffic to each of the recipes.

Each night we looked at three metrics:

• Which recipe was converting best
• What was the positive or negative effect of each element/alternative
• Were the results stabilizing so that we had confidence in the numbers

The graph that described recipe performance and stability looked like this:

The element effect graph looked like this:

After two weeks, we concluded that the test had achieved sufficient confidence to draw conclusions. Based on the data, we found that Main Text, Headline and the Bulleted List had a greater than 10% impact on likelihood to convert. These results suggest two things; the best page would be the existing page with the new body text and with the new call to action. The theoretical best page, which was not one of the recipes tested, should perform as much as 13% better than the existing page.

Interestingly, if we had simply A/B Split tested the existing page versus a new page with the new alternatives of each element; the two would have had the same conversion rate. The higher performing element/alternatives would have been obscured by other lower-performing elements.

Confirmation
The most exciting part of Taguchi testing is when you find hidden elements that provide improvement and adding them together to create large improvement.

We were surprised by the results of the first part of the test. We expected the new main text to have a stronger impact. We were surprised that the existence of the graphic did not impact conversion either way. And whereas we expected the "credibility text" would help conversion; actually it hurt, a lot.

In the confirmation test, we were able to run all of the traffic through either the existing page or the new "best" page so it took less than a week for the results to stabilize. Instead of the predicted up to 13% improvement we saw a 14.5% better conversion on the new page.