Lots of stuff happening.

1) HTTPS default search for logged in Google users.

I personally don’t think it will change analysis all that much, but it might be too early to tell as I don’t know how far along the rollout is.

You may recall that this is not first referrer change for Google Organic.  With the introduction of Ajax-powered search results came a new referrer. I tracked the rollout over time and the last time I checked in September 2010 the rollout was around 50%. The interesting thing about the new referrer is that the rank of the keyword is passed along in the referrer via the CD parameter:

http://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&sqi=2&ved=…

What’s different since October 18, 2011 is the proportion of Google referrers that have the CD parameter. As I said earlier, it used to be around 50%, but now it’s closer to 80%.

(Click for larger. There was a blip on September 29 – maybe a test run?)

So, while you may lose some information because of (not provided) keywords appearing now, you actually get better ranking information on all your Google Organic keywords.

2) Flow Visualization

I don’t have access to flow visualization yet so I can’t comment on what looks to be a beautiful feature, but what I am excited about is the fact that new goals you create will work on your historical data. First thing I’d always do in a new profile is set up goals since they’d only give you data going forward. Being able to work with past data is great news.

3) Event tracking impact on bounce rate

I love event tracking because of its elegant flexibility – you can just send anything to GA and it will accept it. No need to worry about naming conventions or slots. However, events have an impact on bounce rate if you fire them automatically on your landing pages. With a recent update you can now mark events as non-interaction, meaning that they don’t affect bounce rate.

This will open up quite a few options for cases where you had to use custom variables.

Date comparisons are great and allow you to see trends over time. As web analytics we care about trends rather than absolute values, right?

However in Google Analytics, by default you are comparing your selected date range to a date range in the past, but there are cases where you’d want to compare the data to a range in the future. It’s basically a different perspective, similar to saying that 20 is 100% larger than 10 (basis 10), as opposed to 10 being 50%  smaller than 20 (basis 20).

Let’s say you want to know how well your converting keywords from March are doing in April. In other words, did keywords that converted in March also convert in April? You cannot get the answer by looking at the past, i.e. by comparing April to March because in this case Google Analytics would list the converting keywords for April and then seeing if any of those April keywords are also in March. April is not right basis, you want March.

Fortunately, don’t let the Compare to past distract you from the fact that you can compare date ranges going forward. Just click in the dates and you are golden:

And now I can see that none of the keywords that converted in March converted in April. Good to know…

I enjoyed going to the Conversion Conference recently in San Francisco and picked up a few great nuggets to try out on my sites. I especially like it when speakers approach conversion rate optimization from a different perspective than AB or Multivariate testing. Michael Summers‘ presentation on eye tracking was quite literally an eye-opener for me.

I also appreciate hearing directly from retailers and practitioners doing conversion optimization, rather than consultants doing CRO for retailers. Don’t get me wrong, I got a lot of great ideas from consultants, but it is important to get different perspectives, if only to mitigate biases, however subtle they may be.

The bias I see is that many presentations talked about the successes of website testing. Somehow most experiments ended up beating the original. The problem is that if you only report successful experiments you might get the impression that it is about AB testing per se. Get a tool and watch the improvements roll in.

This type of bias, called survivorship bias, has been described in the book Fooled by Randomness by Nassim Taleb. An example he uses is the book “The Millionaire next door“, which studied and then recommended following the habits of those people who became millionaires. But as Taleb points out, “That all millionaires were persistent, hardworking people does not make persistent hard workers millionaires” and “We are trained to take advantage of the information that is lying in front of our eyes, ignoring the information we do not see”.

So we need to know about the failures, even though they are less sexy than successes.

You shouldn’t care that site XYZ improved conversion rates by changing the Call to Action on the Add to Cart button. What would be more interesting to know is the number of other sites who tried exactly the same experiment, and then look at the percentage of successful experiments. Kinda like an experiment of the experiment

Wouldn’t it be much more informative to know that out of 100 different sites that changed the Add to Cart button, only 10 of those reported higher conversion rates? If I only hear about the 10 successful ones, I don’t get the full picture if the other 90 are left out. Contrast that to a (hypothetical) meta experiment where 30% of sites reported higher conversion rates.

Having this type of prior success probability would help you prioritize what to test in a more objective manner.

Not saying it’s easy to get this type of data, but you should still be aware that you may not always get the whole picture. Conferences naturally attract like-minded people, so it’s very important to have a diverse group of speakers and I hope that conferences like the Conversion Conference will continue to do that. I will certainly go again in the future.

One of the many benefits of having a Yahoo! Store as your e-commerce platform is that you get Yahoo! Web Analytics. Not only that, but the tagging is done automatically for you and aspects that typically require some sort of setup, such as e-commerce, site search and action tracking, are done for you out of the box.

The flipside is that the YWA tracker can’t be modified if you use auto-tagging. So, things like custom fields require that you install the tracking code manually. Maybe not the end of the world, but then you’d have to modify templates, and auto-tagging is so much easier.

I came across one case where the lack of auto-tagging customization prevented me from tracking 404 error pages unless I went to manual tagging. Or so I thought! Sometimes a little hack can go a long way.

Below is an excerpt of the presentation I gave at the recent Yahoo! Merchant Summit.

Guess which button I’d like to have click data for?

I know, I know. Along comes a great new feature from the Google Analytics team, which is still in beta, but if you give a mouse a cookie, he’ll want more

The add to cart button is typically a form post, and In-Page Analytics works with links, so I shouldn’t expect this data to be there. Would be great though to know how many people saw the product and then clicked that oh-so important button, wouldn’t it? You can of course get the data with event tracking for example, but it would be nice to get the data visually.

I am wondering if this can only be done with tools that track mouse movements and produce a heatmap?

I have made it a habit to check the roll-out of the new Ajax-powered search results page on google.com in Google Analytics.

December 2009: 10% (proportion of traffic from google.com to my site with the new Ajax referrer)

February 2010: 20%

September 2010: Over 50%

It looks like the most recent significant bump came on August 25, 2010.

And no, this is not the death to search term tracking. In fact, you get more data, such as exact keyword ranking.

Weighted Sort has got to be one of my favorite new features in Google Analytics. There is a nice demo for using weighted sort with bounce rate, but I thought I’d suggest another metric on which to use this new ranking algorithm.

Let’s start by asking a simple question:

What are my most valuable pages?

As a proxy for “valuable”, we’ll use the $ Index metric in the Google Analytics Top Content report. $ Index measures the e-commerce revenue divided by the number of pageviews. The problem with sorting by $ Index is that you’ll see very large $ Index values because the pages only have one or very few pageviews.

In other words, the top page only had one measly pageview out of 3.5 million and it happened to have been during a visit that resulted in one sale. Not exactly optimization material, so let’s try and find pages that are doing a bit more work by using weighted sort. Just click and look at the results:

Don’t you just love it when the data you look at tells a great story? The first few results are all checkout pages, and this of course makes perfect sense as every visitor has to go through the checkout to place an order.

But now we can go down the list and look at other pages and get a sense of their true importance. Note the relationship between pageviews and $ Index: low pageviews and high $ Index being equivalent to high pageviews and low $ Index.

What weighted sort does beautifully is put data into context. You see, I don’t care about $ Index or Bounce Rate or e-commerce conversion rate per se, I am more interested in the impact on business and prioritizing my optimization efforts.

So a logical follow-up question to the first one might be:

Which page should I optimize?

The one with a bounce rate of 60% or the one with 50%? The answer of course is: it depends. If both landing pages have the same amount of traffic I’d pick  the one with 60%. However, if landing page A has a bounce rate of 60% and 1,000 views and landing page B has a bounce rate of 50% and 10,000 views, the impact of reducing the bounce rate of B will be higher even though its bounce rate is already better than landing page A.

I really appreciate the fact that weighted sort bubbles up this valuable data with a simple click, unlike some other approaches. And once you know your most valuables pages, guess which pages make great candidates for AB or Multivariate experiments?

The only downside of weighted sort is that you don’t know it exists unless you sort in the first place. A visual cue might be nice to indicate whether a column is weight-sortable, but I definitely know where to look for it and I hope you do too.

There are a few reasons why clicks and visits from Adwords campaigns will likely never be same in your Google Analytics reports (or indeed any other web analytics tool), but you should of course make sure that you have a correct implementation as Adwords is a paid campaign…

But how can you tell if everything has been implemented correctly? If you have lots of Adwords campaigns driving traffic to lots of landing pages any errors might be hard to spot.

First of all, I would expect there to be fewer clicks than visits. If a visitor clicks multiple times on an Adwords ad during the same session you would see multiple clicks but only one visit in Google Analytics. Also, clicks are counted even before the visitor lands on your site, so if the tracking code does not execute in time on the landing page before the visitor navigates away you’d also lose visits.

In any case, even if there is a normal discrepancy you’d expect that discrepancy to be consistent. So if your normal percentage of visits to clicks is 90%, a percentage of 20% in one of your campaigns would not be normal.

You could of course look at the numbers directly in the reports table and look for unusually large differences between visits and clicks, but if you have lots of data to look at you might not spot any unusual patterns.

That’s why I like two visual approaches.

1) Using Compare Two Metrics

Pull up a new Adwords campaign report and click on the tab above the graph and select Compare Two Metrics. Choose Visits and Clicks. Data look OK to you and are clicks and visits in line?

You can then repeat this approach to drill down into Campaigns and Keywords. Unfortunately you’ll have to select those two metrics again every time you pull up a new report.

But fortunately, there is a sexier way…

2) Using Motion Charts

Pull up your Campaign or Keyword report and click on Visualize in the top half of the screen.

Motion charts allow you to see how several dimensions move over time in a graphical way. Visits should already be selected on the Y-axis. Select Clicks for the X-axis. Now you can see how visits and clicks behave over time. As I said earlier, you should expect clicks and visits to be correlated, but the motion chart should allow you to spot any outliers pretty easily.

As with all web analytics, you are looking back in time, but this is a pretty cool and powerful way to spot any errors.

What’s the only thing better than customers? Repeat Customers. According to the book Flip the Funnel the conventional wisdom is that it is far more expensive to acquire new customers than it is to retain existing ones. Many companies now base their strategy on increasing purchases from repeat customers, which in the case of Zappos is 75% of their business (again from the book Flip the Funnel).

You should be able to get repeat customer metrics directly in your e-commerce order system. If you have a Yahoo! Store, you can get a baseline metric by pulling up the Repeats report. This will show you the number of customers who have placed multiple orders. You can then divide this number by the number of total orders to get the percentage of repeat orders.

In this example, 159/3511 = 5% of orders came from repeat customers. This metric will probably not change that much over short time periods, but I would check it from time to time – perhaps on a monthly or quarterly basis. Are you able to do a better job of retaining your existing customers?

From a web analytics perspective, what I would like to examine is the behavior of repeat customers vs one-time customers vs non-customers. Are repeat customer behaving differently than other segments? Is there anything I can do to target one-time customers to get them to become loyal customers?

To address some of those questions in your web analytics you’ll have to be able to query your e-commerce order system and tag your visitors appropriately, eg as “new customer” or “repeat customer”, at the time of ordering. Implementation approaches will vary depending on your e-commerce platform, but web analytics tools such as Yahoo! Web Analytics or Google Analytics allow you to set custom variables so that you can track these customer segments.

Once you have set up your tracking, what does the data tell you? One of the first questions I would ask is how repeat customers come back to your site. Are they coming back because they know your brand and therefore type in your brand name in a search engine or visit your site directly?

My take so far is that unless you proactively target and try to get more repeat customers, don’t expect repeat customers to behave differently than one-time customers. You will have to work just as hard as getting new customers to get repeat customers in your search marketing campaigns. You will have to earn their business again every time by competing for non-branded terms:

In this example, the new and repeat customer segments show virtually no difference.

Could you be doing a better job of targeting your one-time customers either directly on your site or via email campaigns so that they become loyal repeat customers? Do you have a great process in place once customers place an order or when they call your customer support number?

Hopefully over time you will find that loyal customers will come back directly to your site thereby lowering your direct SEM costs. And finally I should mention that I have not even taken into account other advantages such as word of mouth and referrals from your loyal customers that should help to lower your overall marketing costs.

In terms of web analytics and statistics more data is always better. Ask 1000 people who they’ll vote for and you get a 3.1% margin of error. Ask a 100 people and the margin of error goes up to 9.8%. Since you are asking fewer people you are less confident about saying something about the underlying population.

Same thing for AB and MVT testing. You can see for yourself with the handy Google Website Optimizer calculator. Assuming a 2% baseline conversion rate, a 20% expected conversion rate improvement and the only change is the number of visitors who see your experiment page. If 1000 visitors see it a day, the expected experiment duration is 17 days, but if only 100 people see it per day it is 175 days! Ouch…

Add in the fact that web analysis is likely done on an ad-hoc basis without dedicated people (thereby violating Avinash Kaushik’s 10/90 rule), what’s a busy small retailer to do?

I think the answer lies in focusing on big segments, mini goals and the checkout funnel. Let me explain:

Big segments

Segmentation is the key to finding valuable nuggets in your web analytics. An average metric, such as your site conversion rate, hides the fact that your site traffic is made up of many different segments, some of which are above average and some below average. Once you see differences you can start looking at ways of doing more of the above average stuff and less of the below average stuff. However, if you don’t have much traffic to begin with, there is a risk that you will quickly look at segments that are too small to be significant. An example might be keywords that are converting very well, but only convert once or twice in a period of months. Hard to optimize that. Instead, try to work with big segments, such as:

• Visitors who added to cart. This is a much bigger segment than visitors who placed an order. What can we learn from people who add to cart vs those that don’t? Assuming that your add to cart rate is 6% (which is in line with a 2% site conversion rate), plug in this number into the Website Optimizer calculator. You now get a far less depressing duration of 56 days for 100 visitors if you were to try to optimize the add to cart rate.
• Visitors who use internal site search. I love internal site search anyway because it has lots of tactical stuff you can do, such as fixing zero results searches. But also look at how visitors are using site search, such as by source. Are you sending paid search traffic to a particular landing page, only to find that a large percentage immediately use your site search? Perhaps the landing page is not relevant enough.
• Branded keywords vs non-branded keywords. How are these segments behaving differently? Can you bucket groups of keywords to help you create more and better content that targets both the head and the long tail?

Mini goals

Getting more people to add to cart can also be viewed as a mini goal. Instead of only focusing on sales conversions try to get more visitors to add to cart first. Or you can try to get more visitors to look at product detail pages by making smart categorization choices.

Here is a very interesting metric to look at: Visits to Purchase. Pull it up in your favourite web analytics tool. You’ll probably find that most transactions happen in one visit. Big pat on the back? Not so fast. The way I see it is that you only have one shot to make a sale. After the initial visit, transactions go down rapidly. For this reason consider adding visitors to your mailing list first. This great mini goal helps you keep alive the conversation with your prospects. Obviously you should make sure you have something interesting to say to your prospects. Another example would be to provide your buyer’s guide via PDF in exchange for an email address.

Checkout Funnel

Last, but certainly not least is the checkout funnel. The checkout funnel is of course hugely important for an online store. Your visitors have to go through it before placing an order. If you have not done so, set up a checkout funnel in your web analytics tool, beginning with the cart page and ending with the transaction page. You will have one or more to steps. Make a note of the funnel conversion rate and go back to the Website Optimizer calculator. Remember the example of 175 days? Let’s plug in a typical 25% funnel conversion rate, i.e. a 75% cart abandonment rate, and keep everything else the same. You get an expected experiment duration of only 10 days for 100 visitors. If anything, you can see how important the cart page is and should feature prominently in any optimization plan.