After reading the latest news from InternetRetailer, about how Tower Records increased conversion rates by changing site search, from Endecca to Mercado Software. I had to see it for myself.
I had to see if the relevence was good, the interface was clean, to see if it could find what I was looking for.
Now I may not be a big music hound, but I did remember April Lavigne.
Now it occured to me, that most recent search software should have spelling correction, right? Wrong!
I typed in Lavine, thinking, okay this new search would be able to recognize my miss-spelling and find April Lavigne, well I was wrong.
This led me to think, not so much to blame a particulare site search software, but to identify what kind of features does a site search really need?
1. It needs to know how to interpret search query requests to bring the relevent results.
2. Then it needs to provide an interface to search/browse/sort thru those results
Sounds simple right?
Well let's for example show you what it took Celebros, a very nice site search company from Israel, to create spelling corrections.
Basically they had to go thru every piece of data, every title, category, to hopefully provide clue that if a customer typed x they really meant y.
That took days and weeks to come up with spelling corrections, related searches, categorization etc...
But it still failed thru no fault of their own.
It's a failure of an assumption, that the customers really know the product data, or that they can easily figure out what to say to get what results they want....
Before you can create an efficient and profitable site search, you have to understand the customer, not the client, but the people who are going to query your site search....
Most customers tend to divide into three or more searching categories:
1. Looking for something specific
2. Just browsing, surfing
3. Researching to Buy
Each has different needs, and you have to create a solution that meets all their needs.
Then you need to realize you can create your own cheaper solution using whatever database you have. Whether it be SQL Server 2000's Full-Text Searching, Verity, Oracle SQL Statements.
The key is how to create good experiences when you fail to deliver relevent results, I mean, they may be relevent to the people who work at your company, but not to the customer.
So then there is two approachs, do you spend lots of time or money trying to create a data set, that will fix any search queries?
Or do you try your best, then watch the Conversion Rates, and hope things grow....
At first any new search solution for your site, will show improvements, especially if it's an experience change...
I mean, people can adapt to the worst site search, and even being offered two choices will choose the old one, just because they are more used to it.
So my theory is that you need an adaptive/learning site search, that immediately gives a good experience to the customer, as much as possible, such as offerng a discount/promotion for failing to get good search results or even a live chat link to help not lose this customer, and make sure they can find what they are looking for....
Then you need to get feedback from customer and from your site search programming to identify what types of search failures are occuring, and then to inform your site search/merchandising team to tweak the site search data, to provide the correct results, for the search query the customer used...
Then over time, you'll have a more liquid and more human site search...
It's either that or calling the boy scout to go visit the customer and bring them across the street to your store?