Author Archive

We've put an interesting press release out today - in which we use the site search behavior on DeepDiscountDVD to predict the outcome of the Oscars. This is a bit of fun and came from the wisdom of crowds theory. This suggests that groups are often smarter than any of the people in them. To me this is counter-intuitive - but I'm sure if I asked the rest of the people in the office then, as a group, we wouldn't think so :-). There is a book on this subject.

The idea was to compare the site search traffic on DeepDiscountDVD for Oscar nominees in a few of the main categories. We weren't sure that this would work - most of the recent movies have not been released on DVD. However we found that doesn't stop people searching for them and DeepDiscountDVD does sell posters of Brokeback mountain.

We found that rather than looking at the total number of searches for each nominee, it was necessary to look at how the search traffic had increased. This stopped George Clooney from automatically winning the best director - there are more searches for him than any of the other directors. This approach had it's own flaws - but I don't think it really matters. I'm looking forward to seeing how our predictions pan out.

It has been mentioned in the past that search has a long tail, that is, a large proportion of the queries are unique or don't get searched for very often. This is useful for search marketers to know about. e.g. the term digital cameras gets searched for often and many people trying are optimizing pages and buying keyword based advertising. However the term "7 megapixel canon camera with 2 day delivery" hardly gets search for at all, but would convert to a sale better, be easier to optimize for and cheaper to buy.

I wanted to see if site search has the same sort of tail as web search and explore what the implications are for site owners. First I reviewed the data from one of our web search customers and plotted the popularity of the top 1000 search terms - shown as a percentage of the overall traffic. The most popular search term ("google") accounted for only 0.0044% of all the traffic. The 1000 most popular search terms only accounted for 0.3% of the overall traffic. This definitely looks like a tail.

We then selected four of our site search customers and plotted the popularity of each of their 1000 most popular terms. As you can see from the chart below - although each site has completely different content and visitors - the tails are all the same shapes. The main difference is in the size of the tail. Not surprisingly the more traffic a site has, the longer the tail.

We then looked at what portion of the search traffic was covered by the top 1000 search terms and plotted it against the total search traffic.

So what implications does this have for site search owners - knowing that they own a long tail? One of the main implications is how site search owners go about searchandizing. For those unfamiliar with the term (I'm not sure who coined it) searchandizing is using site search for merchandising, for example promoting products within your own site search for certain terms. Most advanced site search solutions provide these facilities. If you have a highly trafficked site then promoting items for a small number of individual terms is not going to have a huge impact for the majority of your users - you need to merchandise the tail.

It was nice to see a a story about site search today. The crux of the story was that Forrester had released a report showing that 56% of web sites didn't have search engines that supported users' goals. This is very definitely supported by what we see and I think it happens because it is very easy to create a bad site search and users' expectations of search are continually increasing.

It is fairly easy to create a site search - a form where you type in some words and get back some results. Almost any web developer can use existing or free software to create this experience. Most content management solutions will come with some sort of basic search. But to create a good site search requires expertise and experience regardless of whether you've using free software or an expensive package you've bought for the job. Their is an art to;

• indexing all the content while not getting duplicates
  
• indexing the right text - for example it may not make sense to index the navigational text on each page
  
• presenting the search results in a format that most helpful to your users
  
• implementing and displaying refinement and sorting options
  
• and most importantly ensuring the results are as relevant as possible
  

This is one of the reasons why we chose to deliver our site search as a service rather than software. When we build site search for a customer the engineer assigned to the task has the benefit of years of experience across many different types of sites. The result is always a better experience than someone who is building site search for the first time.

The other reason why so many sites have search that does not meet their users' expectations is that their expectations are increasing. The article referred to the Google effect - where people expect all search to be as good as Google. This is reinforced by those sites that do get site search right - with relevant, well laid out results that are easy to drill into, that quickly and easily satisfy peoples desire to find the information that they seek.

This week we announced our new customer Chiasso. I love our customers - and these guys are no exception. They sell all this cool furniture on-line and through their catalog and they're about to open their first store. Like almost all of our customers they're growing. If you're looking for a different sort of gift - maybe a funky toaster or a stylish lamp - then you should take a look at what they've got to offer.

They've been great to work with and I'm pleased we've made a significant difference to their site by providing a better search experience. Somehow Chiasso manage to create good press for themselves. I was impressed when I read about them in Time magazine while they were trialing our service. Our announcement this week also received good coverage - DM News, Tech News World, and Chain Store Age all covered it - with more to come.

Click fraud has been an been an issue for as long as pay per click as been around. There has been renewed media interest in the topic since Google settled a lawsuit for $90 million. As it stands now click fraud is an annoyance - it's one of the costs associated with running a paid search campaign. A cost that can be difficult to determine and is often ignored until it starts having a noticeable impact on your campaign.

The search engines are trying to detect click fraud and not charge for the fraudulent clicks but this is a battle that is difficult to win. A skilled fraudster can probably outwit any algorithm they come up with.

One solution is to change the model from pay per click to pay per acquisition. There has already been a move towards this with the increasing popularity of pay per call. Google has also been testing this. Bill Gross, arguably the father of the PPC model, is promoting a pay per action model with one of his start-ups, snap.com. This model is a lot harder to defraud. If you want to spend your competitor's advertising dollars you can click on their ads and buy lots of their products - but I don't think they're going to get upset.

So how is this relevant to SLI? We have some experience in detecting this type of behavior. We have techniques to ensure that one person's activity can't unduly influence the learning on any of our customers' site searches. In the past when we have detected this behavior they have been crude attempts to influence the results. For example a keyword that normally gets 2 searches a day, suddenly gets 30,000 in one day from a single IP address. If you're going to do click fraud - don't use a sledge hammer - be subtle about it.

This is also relevant to us for our Ad Champion service. This automatically creates and manages paid search campaigns from the data we collect from site search activity. We are very wary of click fraud happening on these campaigns which we are running for our customers. We have an advantage though - because these campaigns are operating mainly in the search tail. In order for click fraud to be successfull our customers' competitors would have to spread the clicks over lots of keywords and it's difficult to work out what they are. If there's a sudden increase in clicks on a keyword that normally doesn't get searched for very often - then it stands out like a sore thumb.

For our Site Champion product, which automates search engine optimization for tail terms, we charge either on a cost per click or a cost per acquisition. The cost per click is susceptible to fraud (although we haven't seen it yet), but because it's operating in the search tail, it too will be difficult to defraud. The cost per acquisition model is risk free for our customers but there is an overhead to tracking the sales. I predict that the risk free appeal of the cost per acquisition model will mean we will see it more and more in the major search engines.

SLI is exhibiting at eComXpo next week, April 4-6, 2006. This is a virtual trade show for search, affiliate and interactive marketers. There are some fantastic speakers presenting and it is free to attend. So if you would like to find out more about our products and services then please come and visit. With our staff spread around the globe we are going to attempt to man the booth 24/7 so you should be able to drop in at any time.

This is our first virtual trade show - and I am very interested to see how it goes. I love the fact I don't have to travel. I'll be calling into the booth regularly and we're giving away one of our brass telescopes - call into our virtual booth to be in to win.

Free registrations for attendees.

One of SLI's tactics to provide a better search is very simple: measure quality of the search, try to improve it, then see if the metrics have improved. In a previous posting I discussed some of our other measures of search relevance. This week we're unveiling another way of measuring search quality: our site search survey. What's more we're making it available free for anyone to use.

This tool allows you to put a question on your site search page asking your visitors if they found the results useful or not. They simply click Yes or No and we record the responses and make them available to you via online reports. We also provide visitors the opportunity to add feedback.

Once you have this tool on your site you have a measure of the quality of your site search which we track for you. It's particularly useful to look at the keywords people are using when they complain and the comments they give. This allows you to understand the types of problems people are having and highlights areas that you should be improving. You can then try improving your site search and see if it has an impact on the number of complaints.

The idea for this tool was inspired by one of our customers (Veritas, I think). We were trying to put together a case study around why they chose our search. We were looking for some hard ROI numbers. But these can be hard to get and often the effort that goes into getting the numbers isn't justified. They said the main reason they chose to keep using us after their trial was because their complaints about search stopped once they put our search up.

Our customer, Eastwood, used the feedback tool on their old search and then on their new search and saw a dramatic reduction in the amount of feedback once they upgraded their search. Their complaints dropped by 96%. With this sort of improvement you just know that the search is better.

So why are we releasing this tool for free? It's not just because we're generous and want to make the world a better place. We are very much a business and this may raise our profile and help us find more customers. If you sign up and look like a good fit for our technology then we'll be in touch with you. There are lots of enhancements we could make to this tool - let us know if you have any ideas.

You can sign up to use the feedback tool here. Enjoy.

Google recently confirmed that they have made a change to their interface. For certain keywords they will show 3 related results inserted into the main results. For example, here's what you see when you search for tokenize.

There have been reports of this for quite some time, so Google have obviously done a reasonable amount of testing with this before they decided to roll it out to everyone. However, I don't like it. I find them intrusive and the times I have seen them appear I was frustrated - I wanted to see results for the term I typed in - not for what Google thought I wanted to see.

I think something like this can significantly enhance the user experience but it needs to be less intrusive. Most other search engines chose to show related searches but few of them have got it exactly right. I like Yahoo's interface where they show related searches below the search box.

This is nice and simple - but suffers from the problem that all the related search terms contain the word insulin, graphically demonstrated if you see all the related searches, e.g for insulin:

Ask have almost got it right - but chose to complicate their interface by grouping the related searches that expand or narrow your search or are related names. This is cool from a geeks perspective but I don't think people care whether they're expanding or narrowing their search. If a related search term that expands your search is more useful then it should be shown at the top of the list.

AOL used to use a similar concept - of clustering - from Vivisimo but when I looked just now I didn't see it. They did show some related terms - but it searched only AOL results. MSN doesn't offer anything like this.

There are some, enlightened smaller search engines who have got it right IMHO. For example search.com and mamma.com show related terms on the right hand side of the results. You just click them to fire off another search It's simple, non-intrusive and really useful.

Of course, I am biased because we provide the related searches for those enlightened search engines :-). But I am truthfully less than impressed with Google's effort.

We're busy at SLI. We've got 6 events on over the next few weeks. We currently have two shows on at the same time: Annual Catalog Conference in Chicago and Internet World in London. Next week we've got the Novator Client Summit in Toronto, Canada and the South Florida Interactive Marketing Association networking cruise. Shortly after that we have Search Engine Strategies in London, followed by Internet Retailer 2006 Conference in Chicago.

That's a busy month - but it means we're going to see lots of our customers and prospects. If you're going to be near any of these then come and say hi. We'd love to see you and show you what we can do for you.

Since my last post we've also announced a couple of new customers - cosmetics retailer, Ulta and Green Mountain Roasted Coffee. We are proud to have these companies as customers. Even though they're in completely different industries they can both see the benefits of having site search that works and of leveraging the site search activity to drive more traffic through automated search engine optimization.

Further to my previous post about Google's related results, I see that Google has an option to refine your results for some queries.

---

---

This is like the faceted search that we offer for our site search (for example search for gift sets on Ulta and we offer 7 different types of refinement). Clicking on one of the refine options will restrict the results that you see. In order to provide these refinement options Google needs to have some structured information about the pages its indexing. For example it needs to know whether a page is for patients or for health professionals. Google appears to have labelled a bunch of pages. You can see this when you click on one of the refinement options:

---

---

Presumably Google is somehow automatically categorizing these pages - they are against manually doing this type of thing. I've only seen the refine option for health related queries - e.g. hypertension. I presume this is part of the rumored Google Health.

I like this feature - it is fairly unobtrusive, and easy to use. I think related search terms would compliment this nicely. They could be shown for queries that don't return results with structured data.

As a side note - internet explorer crashed twice while I was viewing these pages.

Thanks to Paul Denhup from Gartner for pointing this out