Site search technology’s new goal: Making e-retail web sites think
By Mary Wagner
The term “dress up” means something quite specific to customers searching LillianVernon.com for one of the retailer’s most popular products, a trunk that holds scarves, hats and other costume pieces for children. But not long ago, Lillian Vernon’s site search engine wasn’t nearly so picky. A search on the term also delivered a barrage of other listings including one describing brass charger plates that “dress up” the table for the holidays.
On the flip side, a search on “jewelry box” delivered only a handful of results. Paul Goodman, vice president of Lillian Vernon Online, says the assortment could include 20 or more items that could be used as jewelry boxes. “But the language ‘great place to store jewelry’ wouldn’t be found under a ‘jewelry box’ search,” he says.
Too many, too few
Too many or too few results, relevant results mixed in with the irrelevant, no results at all: too often, this characterized the output of earlier generations of site search technology. More than frustrated shoppers and lost sales, ineffective site search also represented an untapped well of opportunity. For even unsuccessful searches might shed light on what shoppers want to find—if only the data could be captured and patterns analyzed.
CEO Steve Kusmer of search technology vendor Atomz Corp. started to realize that site search could be more than a set-it-and-forget it function in a previous job as head of electronic marketing at Macromedia. “People were searching our site and telling us in their own words what they were looking for, what their intentions were,” he says. “I thought, ‘What can we do with that?’”
As it turns out, plenty. Atomz is one of several search providers that now leverage the technology not only to improve search results, but also to do something more—improve on-site merchandising.
Newer functionalities underlying search products, such as natural language processing and guided navigation, are doing a better job of mapping computer logic to human thought. The new glamour technologies driving search results have high visibility among users, but site search has also benefited from smaller innovations behind the scenes, says Glenn Barnett of web technology consultants Molecular Inc. Take the fact that most search applications have increased their update frequency, for example.
More site search vendors have emerged to give e-retailers an alternative to the basic site search tools that arrive bundled into e-commerce platforms. Improved site search also has depended on a corresponding trend: the rise of web analytics.
Site search vendors are now integrating services with web analytics providers. Agreements between Atomz and WebSideStory Inc. and between Endeca Technologies Inc. and Coremetrics Inc., for example, have been announced this year. “Since users can now interact with search in more ways, it’s become harder to track what they are doing. So it’s critical for marketers to connect with something that is capable of crunching the numbers and giving a relevant business view of what their users are doing,” says Barnett.
New merchandising initiatives
Analytics not only capture data on site search behavior to point the way to new merchandising initiatives; they’re also delivering feedback on the performance of those initiatives in near real-time. In the future, say providers, analytics packages will not only collect data on site searches; but they will also feed back into and adjust search algorithms to populate search results as well.
“Say someone is searching for a product and they get a certain set of results. The third result is the one that gets the highest conversion rate. Why doesn’t that become the first result?” says Atomz’s Kusmer. “The future is having analytics automatically drive search relevancy.”
The next frontier
Several site search products are heading in that direction, already integrating limited analytic data into the presentation of search results. Like a number of other products, Qwiser, the search and navigation product of Celebros Inc., for example, automatically ranks listings within a set of search results based on products’ current best-seller status. Though compatible with outside analytics packages, Qwiser also has its own built-in analytics tool. Future releases will incorporate data such as personal or demographic information, into how it ranks search results on a site. “The next frontier for site search is combining search and navigation with analytics and merchandising rules,” says Michael Crandell, CEO. “That will create a self-tuning, automated system that constantly improves merchandising through the search results it delivers.”
For now, sites like LillianVernon.com are simply glad to fix their immediate search problems. Lillian Vernon last September licensed site search from vendor EasyAsk Inc. as a replacement for the site search function bundled into its IBM WebSphere e-commerce platform and saw a triple-digit increase in conversions off site search within weeks. Goodman won’t say what it cost Lillian Vernon, but he does say it paid for itself in improved sales before Christmas.
Natural language processing is at the core of a number of newer site search products. Its goal is to gain a deeper understanding of query intent. “We also use that information to be very directed about how we present the content we retrieve and how we leverage that content interaction to create opportunities to influence,” says Tony Frazier, senior vice president of marketing at iPhrase Technologies Inc., which provides natural language-based search technology both directly to its own customers as well as through OEM relationships with CRM technology vendors.
300% conversion boost
Here’s an example of how natural language processing works at iPhrase client eLuxury.com. Under earlier-generation search technologies, a search for “earrings under $200” would likely return results containing any of the three terms, reflecting a broad range of relevancy—including none. Natural language processing, however, parses the query—that is, breaks it into component parts to establish the meaning of each term individually and then in combination. That determines, for example, that “under $200” is a price constraint rather than two separate terms.
iPhrase’s natural language processing technology compares the individual and combined elements of the parsed query to what’s in the site’s searchable database, which includes category information, product information, brand names, longer merchandising descriptions, and more. “It takes what the user asks for and compares it against the different elements of back-end content,” says Frazier. That means results listings deliver more precise answers that get users to where they want to be faster, he says.
Frazier notes that retailers such as Neiman Marcus have seen as much as a 300% improvement in conversions directly off site search after implementing iPhrase’s search and navigation technology. Sites such as iPhrase client Sephora.com take it a step further, harnessing the technology for active merchandising duty. At Sephora, a query on “dry hair,” for example, may not only deliver relevant hair products Sephora wishes to promote, but also related content, such as a magazine article that offers guidance on the subject.
“They are using the same business rule facility to directly promote product and to promote non-product content that educates a user on how to solve the problem—and along the way, sell more product to facilitate that,” Frazier says “It’s one of the ways our product has evolved. We have moved from using the natural language processing capability to not just deliver more accurate answers, but also to understand how to influence a customer.”
Guided navigation lets shoppers narrow search results into a progressively smaller set, based on a series of choices presented to them as results are returned. The choices are tied to product attributes such as color, size and price. The challenge on outdoor gear retailer
Orvis.com was 4,000 products and 35,000 SKUs: in other words, many, many subcategories of a relatively small number of products. “We were looking to bring some intelligent navigation to our search results,” says director of e-commerce John Rogers. That led Orvis to vendor Endeca, where the guided navigation approach is a product centerpiece.
“Someone looks for a sweater on the site. They know what they have in mind, but we don’t know what we should show them—men’s? women’s? wool? cashmere?,” says Rogers. “And you don’t have much time, because if something doesn’t jump out at them right away when they key in sweater, they’re off to the next site.”
The merchandiser override
Endeca streamlined the existing search function, guiding shoppers more quickly to what they were looking for. And it did something more: it boosted site merchandising by automatically feeding related product listings into page real estate set aside by Orvis for that purpose. Orvis positioned a box that runs at the top of every search results listing under a major product category, “boots,” for example, that displays current best sellers or whatever Orvis wants to display, depending on what the customer is searching for. Endeca’s technology selects category best-sellers to populate the box automatically; Orvis’s merchandisers can override that to feature other products of their choice.
Within a month of implementing Endeca’s search and navigation technology, Orvis saw sales off site search gain by 50%. Sales from the merchandising feature, which Orvis calls the merchandising zone, now account for 24% of all sales from site search. Rogers says the new site search capability also pays for itself in better reporting. “The search shows you where you are missing and where your action is,” he says. “It is allowing us to be smarter about how we build our rules set.”
As with any kind of technology implementation, prices vary. Celebros, for example, which aims for small to mid-sized sites, offers a hosted service or will install its software on the retailer’s server. Licenses for its product start at $2,000 monthly, with no other fees. Atomz’s search product is offered exclusively on a hosted basis, with the typical annual pricing at about $60,000, higher for service that must support a very large number of SKUs; smaller for a lower number of SKUs, Kusmer says. At iPhrase, which targets large, complex sites, the standard pricing model is $150,000 for a perpetual license with an annual maintenance fee on top of that. Pricing may scale upward or downward depending on the amount of content and volume of interaction the software must support. IPhrase’s Frazier notes that some retailers buy the license on an annual subscription basis which significantly reduces their initial acquisition cost.
Tower Records uses search and navigation technology from Mercado Software Inc. to build a “Spotlight” feature similar to Orvis’s merchandising zone. The spotlight is a section that runs at the top of search results, into which Mercado technology feeds data on different artists and products. “We can set the merchandising rules on what appears in the spotlight based on when the shopper searches or browses on certain terms or refinements. We can also set third-column ad squares based on merchandising rules,” says Kevin Ertell, senior vice president of direct to consumer operations.
Fine-tuned
Tower maintains a ranking of its 5,000 best-sellers. Under one of its merchandising rules, if any results of a searched query fall within the 5,000, they’re highlighted as best-sellers in the search results. “That one’s completely automated, and it’s probably our most successful,” says Ertell. “The promotions we are merchandising contextually within search are getting strong click-through rates; in some cases, as high as 20%. Our promotional pages are our highest-converting pages, with anywhere from 8% to 30% conversion rates, so the contextual promotions are clearly benefiting sales.”
Nevertheless, Ertell already has his eye on future rounds of site search improvements. “In our type of product, there are a lot of weird band names and crazy spellings, so with 800,000 SKUs, there are things in there that we don’t catch and so people don’t find them. We do our best to look for them but it would be nice if it were all automated somehow,” he says. “We have really fine-tuned our search engine, but whatever it is in a human being that instantly processes and interprets correctly, computers aren’t there yet.”
mary@verticalwebmedia.com
Click Here for the Internet Retailer Guide to Site Search Solutions
