Web-only retailers, including Amazon, accounted for 42% of sales of all retailers ranked in the Read Now
(Page 2 of 3)
So far, TowerRecords.com is the only site using the search-and-browse technology, although the other vendors say deals are imminent. EasyAsk, for instance, rolled out Search Adviser toward the end of January and expects to announce an implementation soon. Mercado unveiled its new IntuiFind 5 in November and it, too, expects to announce implementations in the near future. Netrics has a cross-referencing feature on its product, but continues to stress search results over the browse capabilities.
That most retailers today rank the importance of search very highly and are looking for more effective search tools is not surprising. Failed search has been the cause of much lost revenue over the years. Jupiter Media Metrix reports that search is the first thing 16% of visitors do when they arrive at a site. In addition, 50% of visitors whose navigation attempts fail turn to search. But Jupiter says that as many as 85% of searches do not return what the customer is seeking. By that measure, only a slight improvement in search would yield great returns.
Thus if a site hosts 500,000 visitors a month, and 48% of visitors (the initial 16% who search plus half of the remaining 84%) search for a product and 85% of those searches fail, just over 200,000 potential buyers have left the site without being given a chance to even look at what they want to buy. “Most sites don’t know how to do search; they’re just bad at it,” Berk says.
But achieving accuracy has been difficult. And that has given rise to the search-and-browse phenomenon. “The whole text search metaphor is wanting,” says Alperin. “Search technologies are very brittle because you’re searching on words that people enter into the box. But you can’t search literally, because consumers don’t know the merchant’s terminology.”
Furthermore, because of that very inaccuracy, searches may generate more results than any consumer would be inclined to browse through. Jupiter’s research found that 35% of consumers abandon a search after the first page or two because it’s too time-consuming. Those results are much like not assisting customers in stores. “If a shopper goes into a Macy’s and asks for a shirt, the sales associate isn’t going to say, ‘Go to the fifth floor, it’s full of them,’” says Dycian. “The clerk will ask if the customer is looking for a dress shirt, or a T-shirt, men’s or women’s, and so on.”
Still other searches may fail to produce results because the customer is basing the inquiry on brands the merchant doesn’t carry. But the search still may represent products for which the merchant carries a competing brand that a customer might be interested in if the customer knew the merchant had the product. And so, say search-and-browse advocates, the search needs to be guided, just as a sales associate would guide a customer in a store. “If you go to a store and ask for a brand they don’t carry, the clerk will interpret what you want and lead you to that product,” Alperin says.
To be able to sort on attributes, those attributes must be in the database to begin with. “There’s got to be some hook that we can grab onto,” Alperin says. By necessity, most retailers’ databases contain several attributes for each product, so the search-and-browse technologies can grab those attributes. But retailers need to beef up the data in databases for some products. The amount of additional information needed will vary depending on the age of the database and the retailer’s desire to have a variety of information in it. “We have one customer who put 200 attributes into the database for no other reason than he thought it might be useful some day,” Alperin says. He notes, however, that the addition of attributes can be an incremental process. “You don’t have to do it all upfront,” he says.
Taking a test drive
Until now, the search battles have been waged on the accuracy of search, with each vendor claiming its accuracy is better than the others’. Compared to the early stages of search, the products of the past few years have been a major improvement. Most rely on spelling-correction technology-type in Led Zepelin, for instance, and the search engine will automatically give you results for Led Zeppelin-and matching of concepts-type in slacks at a site where the database contains pants and the customer will still get an answer. But all still have their quirks. Any retailer wishing to check out a vendor’s claims need go no further than to sites that feature that vendor’s technology and type in a few misspelled words, analysts say. “The caveat to that is that there’s not a search engine that’s so awful that a lot of mechanical work by the site owner won’t offset a lot of shortcomings,” says Forrester’s Manning. “So not finding problems is not indicative that the search engine is wonderful. You really need to talk to the site owner to find out how much mechanical work they had to do.”
There’s also a contingent that argues that search-and-browse will only be as good as the search. Thus accurate search is still key because only with accurate search can a shopper be guided toward navigation. “The quickest way to connect to customers is to let them tell you what they want and then provide accurate search,” says Damianakis of Netrics. “You can generate a list of other products that have affinity with the product they’re looking for, but you need to make sure you’ve provided accurate search results to begin with for the affinity list to be accurate.”
Improving search on a site typically costs about upwards of $200,000. The average installation in Forrester’s research ranged from about $100,000 to $400,000. Sites with as little in sales as $2 million a year would experience a quick payback by installing systems at the $100,000 range, says EasyAsk’s Alperin. Vendors all say their search tools can be implemented in a matter of weeks or months and offer immediate paybacks in terms of increased sales. Tower Records, for instance, licenses the Endeca technology with a monthly fee. “We pretty much get it back at the beginning of every month,” Bressler says.