• The javascript integration was never that good. Comment counts were almost never accurate and flaky to boot.


• People had to log in to typepad or they could comment anonymously. What about logging in via Facebook or Google?

• Force me to be explicit in my experimental design goals.
• Act as a guide for a pilot course I am teaching on optimizing non-profit landing pages.

We're planning on turning the pilot course into a full offering in the Fall. Current plans for the course include approximately the following agenda:

• First five weeks: Become familiar with non-profit business model; learn non-profit search engine marketing campaigns and goals; relate campaigns to measures provided by Google Analytics; perform initial LIFT analysis culminating in presentation to non-profit.
• Second five weeks: Conceptualize, design, and execute landing page variations for search engine marketing campaigns; start experiment using Google Website Optimizer; present completed design to non-profit.
• Final four weeks: Track results of Google Website Optimizer experiment along with other metrics and prepare final report for non-profit.

The thing I'm excited about here is that we have a chance to make this a sort of open laboratory on the web. That will give the students plenty of opportunity for exposure and feedback.

As people who have been followin my Twitter stream know, I recently embarked on a project to use web site optimizer to improve conversions on a landing page I'm using for a Google AdWords campaign. The experiment has currently been running for about a day and suggests that I might be able to raise my conversion rate, the rate at which people undertake one of several desired actions on this page, by 25%. My goal in writing this post is to lay out the rationale of the experiment and how I designed it. In a later post, I'll talk about the actual results. There appears to be a lift but not 25%.

Michigan Innovators: An Educational Non-profit

Michigan Innovators is an educational non-profit that connects the innovation community through video case studies and other content on its web site. The site's audience includes anyone interested in innovation, in particular those who run innovative companies and those who would like to work in innovative companies.

The Jobs Campaign

In November, 2008, Michigan Innovators received a Google AdWords grant allowing us up to $10,000 /month in in-kind advertising on Google's search enginge results pages. After experimenting with several campaigns, we hit upon a highly successful jobs campaign. The main goal of the jobs campaign is to connect people looking for jobs with those offering jobs in Michigan's troubled economy. In spite of what one may read in the papers, there are jobs on offer in Michigan. These jobs tend to require more education and mastery of technical subject matter.

The screen shot opening this post shows the landing page we developed for the campaign. The landing page uses a job search engine provided by Indeed. Michigan Innovators' content is then arrayed to the right of the search results.  In February, the jobs campaign pulled in just more than 300 daily visits to the landing page. Visitors spent an average of over 10 minutes on the site and looked at an average of over 10 pages of job search results.

Michigan Innovators monetizes the page as follows. Every time someone uses the search engine embedded on the page, we are paid a small amount. Every time someone clicks through on a job link, we are paid more. At the end of February 2009, about half the visitors clicked through on a jobs link at an average rate of four times per visit.

We decided to design an experiment using Google Website Optimizer to see if we could improve the rate of click through.

In designing this experiment, I used WiderFunnel's LIFT methodology. WiderFunnel is a Conversion Optimization agency located in Vancouver, BC. WiderFunnel's LIFT methodolgy provides a framework for analyzing factors that impact conversion on landing pages. This analysis can then be used to design alternative landing pages with a good likelihood of raising conversion.

Analyzing the Landing Page using LIFT principles

The first element in LIFT is to analyze the original landing page. To facilitate the anlysis, here is an annotated version of Michigan Innovators' landing page:

The first thing to note on the page are the blue ovals on the first job listing. They indicate the location in each job listed of the links that lead to better monetization for Michigan Innovators. If a visitor clicks on either of those links for any of the jobs listed, they go to the Indeed page for that listing, and Michigan Innovators subsequently receives revenue based on how far the visitor penetrates into Indeed's site.

Turning to an analysis of how we might improve conversion on the page, LIFT identifies five factors that influence the likelihood of landing page conversion. We'll examine these factors in light of how they might influence visitors' likelihood of clicking on the links of interest.

• Relevance: Does the landing page relate to what the person thought they were going to get? The landing page seems crystal clear in that regard. It's part of a campaign for people wanting to find a job in Michigan. The goal is clearly stated at the top of the page and the search box directly beneath confirms it.
• Clarity: Does the landing page clearly articulate what the user needs to do to receive benefits? In this case, the area enclosed by the blue box and labeled "3" indicates the job listings themselves may be of concern. First, the listings are in tabular format while most search result listings are presented as a list, possibly making them harder to read and digest. Second, the links are a non-standard color, possibly hiding the fact that they are links at all and making it less likely the person will click. Third, the items to click may lack prominence. The "learn more and apply!" link is way off to the right and trails a paragraph.
• Anxiety: What are potential misgivings the user could have about undertaking the conversion action? In this case, demands on the user are fairly minimal. All they have to do is click. A statement about Michigan Innovators' status as a non-profit and evidence of the kind of work it does is in the right hand column, allaying fears that clicking may lead the user somewhere harmful.
• Distractions: Are there items on the page potentially leading the user away the from the goal? The green rectangles labeled "1" and "2" represent two sources of concern in this regard. First, the link to Indeed is unnecessary. Contractually, a link at the bottom of the results is all that is desired. The "Indeed" link could be a source of clicks away from the page as could the link for Michigan Innovators for that matter. However, since Michigan Innovators is the main organization sponsoring the service, a link to that organization seems necessary to convey ownership and allay anxiety. However, the Michigan Innovators content, labeled "2" in the image may be an unnecessary distractor.
• Urgency: Is there an indication that the action needs to be taken now? The current design does contain a call to action, but it may be weak. Given the current job market, we felt that people arriving at our page might already possess a high sense of urgency, making this variable hard to manipulate.

Developing Alternative Hypotheses

Our analysis suggested that lack of clarity in our presentation of search results and distractions on the right side of the page could be hampering conversion. However, the analyis did not tell us exactly what to do about these factors, leaving an element of uncertainty to any hypotheses we might develop. Further, any alterations our analysis might suggest would likely involve changing groups of page elements and, possibly, significant aspects of page structure.

With this amount of uncertainty, a simple solution is to set up at least two landing page variations that act as alternative hypotheses about solutions to the issues raised. These alternative pages can then be tested to see if either outperforms the original.

Note that this approach does not lead to a factorial experiment design, eliminating our ability to determine the extent to which the different factors are responsible for the effects we see. However, when analyzing choice data, the data requirements for full factorial designs can often be prohibitive.

Alternative 1: Clarified Search Results and Substantially Reduced Sidebar

This alternative contains two major alterations to the page. First, the sidebar is reduced to just making a statement about Michigan Innovators' non-profit status and offering a link to additional content. We also removed the link to Indeed. As a result, distractions were substantially reduced.

Second, we completely changed the presentation of search results to more closely resemble those found on other search sites. The links are a standard color, and the first link now has a much larger typeface. Further, color is used to set off the company name, and the location is also highlighted.

Alternative 2: Clarified Search Results and Original Sidebar

This approach is similar to the first alternative, but we retain the original side bar. An alternative hypothesis is that this sidebar gives more weight to Michigan Innovators as an organization, thereby relieving potential anxiety about who controls the page.

Preliminary Conclusions

The experiment is not yet complete. Initial results suggested a rather dramatic improvement in conversion with Alternative 2. However, we have not yet collected enough data to fully test the significance of these results.

Twitter in some sense serves as a techy stream of consciousness. One way to get your blog noticed today (vs maybe later) is to inject it as an extension of yourself into that stream. How do you do that? Well, there's a myriad of tools and strategies. Here are two I recommend: 

• Twitter feed: Twitter feed let's you take your blog's syndication feed and republish it as a tweet from your twitter account. I find this method superior to others that are out there. Almost every blogging software produces feeds, so Twitter feed works with anything. Also, Twitter feed let's you choose which parts of each blog post get included in your tweet and also let's you avoid the indignity of having your blog tweets prepended with "New blog post:". That way, you can tailor twitter injections from your blog posts to be more tweet like. I recommend writing your summaries like a tweet and then telling Twitter feed to post those. Titles are another possibility. Just pick one (not both) and write them like tweets. In fact, I compose mine in the twitter input box to make sure they fit.
• Blog comments feeds: These can help keep the connection between twitter and your blog active. I'm currently using Typepad Connect for managing blog comments. It produces a feed of my personal comments. I make sure the first sentence of any comment using Typepad Connect could be meaningful in a twitter context as well as in the context of the blog comment. Typically, that means a provocative sentence that can stand alone. That way, when it's tweeted, people might take notice.

That's enough for now. It was this weekend's insight.

Too much work. 

The idea behind TypePad Connect is that they handle all comment management, including spam removal, and basically republish the comments on your blog. It's one of several competitors including Intense Debate and Disqus. These services offer several advantages for both blog commenters and blog owners that I'll be discussing in future posts.

I told her I had not seen her updates. She tends to do about one a day, usually at a time when I'm not watching twitter. I only follow about 200 people on twitter, but even at that level, updates are like a torrent, and it's hard to wade back through the past.

Twitter is an extreme case, but in many ways it typifies the modern web. Google has counted over 1 Trillion web pages. Four years ago, the web was around 20 Billion, and that seemed like a lot. You can't keep up with everything that's being produced on the web even as you attend to it. Only a very small amount is worth referring back to over time.

I've restarted this blog, after almost a three year absence, and I've kept the stuff from back then. It's almost all out of date but retains an emotional attachment for me. It's interesting to see the beginnings of some things that have become full blown and the witherings of others.

I have a lot to say about information communities, and my 140 character twitter account won't hold it all. This is a place to say more.

I'm looking forward to what is learned in this trial for Nature. I'm also looking forward for what it stimulates in the scientific community about the possibilities of openness. To move away from the publish or perish model, or to reimagine it to support cross boundary collaboration where the public good can be served, is a huge change. It would rattle down to the tiny bones of every organization. It would change donor and funding streams. It would change higher education.

I like the attitude of collaboration here. I'm just not sure it's practicable. Being inside academia, a clear issue is measurement. So much depends on getting certain kinds of counts.

Well, you might say, just count differently. The problem with that is that "differently" implies "different". Can I justify rewards that I gave based on set of counts using another set of counts? Can I give lifetime employment based on the new counts? There will be a lot of hesitancy around that because once you open debate on the basis of reward, you effectively open debate on the nature and extent of the reward itself. People who have already gained the rewards do not want that.

And it's not because people are reward grubbing or selfish necessarily. In the vast majority of cases, you're talking about people's life's bread. All of us are very conservative when it comes to life's bread.

I suspect that what will ultimately have to happen is that a new system will have to come up to replace the old. Disruptive forces like wikipedia might help with that. Can current academia be replaced by a creative commons?

Recently, we began to start releasing some of our podcast videos from MuscleVentures onto the popular service, youtube as well as google video. By far our biggest hits are coming from footage of a bodybuilding show. Why? Well, I suspect it is a least common denominator effect. People can look at it and appreciate it without really having to have any particular, arcane knowledge or exerting much effort to understand.

You might be tempted to recoil at that, but in almost all activities, only a few people have time to become experts. If you want to make money in your area of expertise, you have to find ways to immediately make people see the value of what you are doing without having to think a lot.

One thing I noticed in that exercise was that when I posted video clips, those posts would immediately gain more visibility for relevant search terms. Well, if that wasn't proof enough that video was hot, services like youtube and google video kept on popping up, with Om Malik reporting on constant venture money coming into the category

These services are like little walled communities unto themselves. That changes the dynamics of search a bit. If you just follow the basics of metadata and target areas that are less well covered, you win bigger right away with good search rankings on terms like bodybuilding that are maxed out by optimizers in the regular index. Further, with Google Video, these rankings seem to translate into better search placement in the overall index.

For instance, click on this search for my training partner Nancy Arnold. As of today, this google video and this second google video occupy positions 14 and 15 with 469 and 250 views each in addition to being linked from our video blog.

The key lesson here is that the space is wide open for anyone with even mediocre search optimization skills. Maybe people who produce media just don't like to write metadata. Metadata in these video services doesn't come for free like it does with regular blogging.

I won't go into what I think are the merits of that idea, but one of the key points that came up was identity. How do you consistently identify the commenters? Last November, Kaliya Hamlin and I had a conversation on that very topic, and she was convinced that establishing identity was key to community work.

It would be nice to loosen this up such that everyone could have an identity associated with a URL, as in the OpenID spec, and Mark Paschal's implementations for Movable Type. As a community builder right now, the problem you have is that you have to do a lot of infrastructure work to make that happen, not the least of which is educating users.

By the way, by community, I don't just mean site visitors but rather people who are aware of each other and care what each other is doing. So, I'm not just talking about pulling in readers with adwords and getting them to convert.

So, what we have done is put together a site called Muscle Ventures. It offers text content, exercise analysis videos, and podcasts. The site seems to be working for the local community. We'll see what we can do for the Internet community. I'll be writing more on this topic as the situation evolves.

Finally, there was a little break in the clouds, and I upgraded. Let's see how SPAM fighting works. At least there will only be one place to blat it out. Further, whatever I did to put myself on the SPAM blacklist has hopefully gone away making it possible to respond in commments again.