Landing Page Testing & Tuning: SES San Jose
Time to talk landing pages. Sage Lewis (Search Engine Watch, SageRock.com) is going to offer us an intro and then it’s all Tim Nash (Site Turners).
He wants us to admit we have a problem – our baby is ugly. The good news is that through testing we can improve this. We’re going to start off with some case studies.
Case Study: RealAge
Conversion action: Completion of free test registration
Tuning Method: SiteTuners Tuning Engine
They had a 40 percent lift in conversions with subtle tweaks. Shortened sales copy, moved stuff around.
Okay, he’s jetting through these case studies faster than I can type. Just know the results were impressive, and often only by tweaking small things on the page. There lies the power of landing page optimization. Point taken, Tim. Now let’s get to the meat.
Landing Page Optimization Background
Online marketing activities:
- Acquisition: getting people to your site
- Conversion: getting them to do what you want them to do
- Retention: getting them to keep doing it.
Conversion rate: The percentage of visors who take desired actions.
Landing page optimization: Improving conversion rate by testing Web site changes.
Why should you care? You have neglected your landing pages. Your conversion rate has dropped and now it’s costing you money.
Tims says your CPA equals your CPC over your clickthrough rate. Ew, Math.
Question: What is a camel?
Answer: A horse designed by a committee
Who should design your Web site? Ad agency? Your boss? Webmaster? Marketing? IT? The answer is none of the above. Your visitors should design your Web site.
You can either keep doing what you’re doing and become irrelevant or embrace your barbarian visitors. They’re going to do your messaging for you. They’re going to set your price. They’re going to set your copywriting for you. Your audience defines the interaction they want to see.
How to Test
- Size of test: 2-10,000,000+ recipes
- Test configuration: Restricted versus freeform
- Data collection: Full versus fractional factorial
- Data analysis: Parametric versus non-parametric
Tim gives some different interaction examples matching headlines and images. The moral of his examples is that it’s not the headline and it’s not the picture. It’s the context.
Picking a Tuning Method
AB Split Tests: Test one variable at a time. Send equal traffic to all versions. Very simple to implement. You get 1-10 recipes.
Multivariate Testing: Test several variables at the same time. Ignore variable interactions and try to predict best setting for each variable. You get 1-100 different versions.
Non-parametric tuning: Proprietary math for Digital marketing. Designed for large-scale tests. Takes variable interacts into account. Get millions of versions of your
Avoiding The Pitfalls aka The 7 Deadly Sins
- Squandering Attention: Stop screaming at your visitors. Eliminate choices and unclutter what remains.
- Frustrating User Experience: Guide people safely to their goal. Don’t make them feel stupid and avoid painting them into corners.
- Invisible Risk Reducers: Make risk reducers prominent. Show them on all pages.
- Lack of Social Proof: Transfer trust from larger brands in the form of endorsements, affiliations, and media coverage. Transfer trust from your clients with case studies, client logos and testimonials.
- Surprising or Confusing Visitors: No one likes surprising online. Don’t interrupt people unnecessarily.
- Ignore Your Baseline: Always devote some bandwidth to your current version. Measure relative to the baseline, not absolute performance. You test a change that increases your CR from 4.6 percent to 5.03 percent, should you get a promotion? No, because things have actually gotten worse relative to the baseline.
- Not Collecting Enough Data: Do not make decisions based on too little data. Pick a confidence level and wait to see which version is better.
The rules of Web action are: Get out of my way, make it easy and don’t surprise me.
Turns out we covered this panel back at SES NY 2008. You can read our recap there and find some of the stuff Tim didn’t mention this time around.