Twitter still has 320 million monthly active users, but its monthly active user totals in the United States went down.
How Easy Ask is delivering better search results.
Coldwater Creek’s customers like the Coldwater Creek web site. In fact, last year, Jupiter/Media Metrix rated it the stickiest e-retailing site. But customers coming to the site couldn’t search for what they wanted. Rather, they had to drill down through the various product categories before they zeroed in on their purchase.
Coldwater Creek was well aware of the drawbacks of a site with no search ability. “With that system, the customer can’t define what she wants to see,” a spokesman says. But Coldwater Creek was hearing clearly from its customers that they wanted to zero in quickly on what they were seeking. “This is a customer service issue,” he says.
Many retailers are familiar with the search phenomenon of goofy results. A customer who types in “no-iron slacks” is as likely to get a listing of irons as he is of slacks. And in many cases, if the product is not described as slacks in the database, but rather as trousers, the customer is likely to not find at all what he is looking for.
Reluctant to create such an experience, Coldwater Creek did without a search function. But in March, Coldwater Creek customers were able to start searching for what they wanted. No longer did they need to map their way through the site to find the goods. The difference: Coldwater Creek had been sold on natural language search and implemented a system from Easy Ask Inc.
Most retail web sites today experience a 2% browse-to-buy ratio, meaning that 98% of the customers who visit a site leave without buying. “Search is one of the primary areas that people complain about,” says Larry R. Harris, chairman and founder of EasyAsk.
EasyAsk and rival Mercado Software Inc. are out to change that. Easy Ask today is rolling out its search capability. Mercado has been installing natural language searches on e-retailing sites since 1999.
The most important thing
Unlike most searches, a natural language search does not rely only on the descriptions of products that reside in databases. Rather, it performs a synonym search on all the information in the database. Thus the customer who types in “pants” will see listings for trousers if the retailer uses the word “trousers” but not “pants” in the database. It also incorporates automated functions to match unfamiliar words with familiar words, corrects spelling and makes decisions about proper names.
The importance of quick, accurate search results cannot be underestimated, say market observers. “Search is one of the most important things consumers do online,” says David Schatsky, research director for Jupiter Research. “It’s vital to shopping and one thing that retailers should be paying close attention to.”
And an efficient search is particularly important. “The online shopper’s patience is very, very finite,” says the spokesman for Coldwater Creek. Coldwater Creek spent several months testing the EasyAsk system before rolling it out. “This will expedite the whole process and we’re expecting it will close a lot more sales,” he says.
Harris launched EasyAsk in 1999 to develop a natural language search engine. The company grew out of Harris’s work with computational language as far back as the early 1980s. Harris, who holds a Ph.D. in computational linguistics from Cornell University, was focusing on the use of natural language technology to retrieve information from computer systems. In 1981, he developed a system to access data on IBM mainframe computers. Eventually more than 600 corporations used the system, called Intellect, for a variety of applications.
That success led to the development of an Internet-oriented search application. “This product was really the result of frustration from doing searches on the web and seeing how bad they really are,” Harris says.
Most search engines today are fairly rudimentary operations. Most web users are accustomed to having to word their searches a number of different ways before the information they need shows up. That is because searches are looking for the exact phrase the user types into the search box. And while searches can find combinations of words and have developed parameters so that a user can specify that results show all the words, any of the words or the exact phrase, the success of a search still relies on the searcher describing the sought data in the exact same terms the database being searched uses.
By contrast, EasyAsk, as well as Mercado, applies thesauruses to the search function and instructs the software that someone who types “slacks” is looking for trousers and pants. Similarly, someone who types in “cardigan” is looking for a sweater and if she types in “rose” she might be looking for pink.
Furthermore, EasyAsk also can ask for a clarification. For instance, if a consumer types “Simmons” into the search box, EasyAsk can inquire if the consumer wants Simmons mattresses or Richard Simmons exercise tapes. Harris says the company feared the follow-up question would make the system look less than adequate. “But we found many retailers viewed asking for clarification as an enhancement in terms of customer service,” Harris says. “Instead of looking stupid, you look quite smart.”
To start developing the search function, EasyAsk will run its software through a retailer’s catalog looking for words for which the system already knows synonyms. It also will perform stemming functions, in which the system looks for the stems or roots of words in an effort to match an unfamiliar word with a familiar word. When that process is finished, the system can answer about 80% of questions correctly on the first try. From there, EasyAsk monitors the user logs and adds phrases that customers have tried that have come up empty. “Within a week or two, we should be up to 95% plus,” says David C. Harris, vice president of marketing and no relation to Larry Harris.