If you can't quite spell Hannah Montana or Boyz II Men you're not alone. But Ticketmaster found it doesn't pay to be too strict about spelling and modified its search engine to accommodate creative alternatives. Managers of enterprise and website search services could do well to follow Ticketmaster's lead.
"Search drives sales for a company like us. It is also the primary point of customer interaction. It is highly visible and it needs to work. Otherwise, we'd lose millions of dollars," Ticketmaster's Geoffrey Young said during a presentation at the ApacheCon conference earlier this month in Atlanta.
Ticketmaster improved the success rate of its searches by 30 percent by examining those queries that did not produce any results for the user, Young said.
"We approached things as a miss-driven solution and asked, 'How are we going to take care of our users in their existing usage patterns?'" he said.
When the company revamped its search service, it incorporated some of its findings into the new search logic. With Ticketmaster's quirky data set, it was a difficult challenge.
Ticketmaster processed US$1.3 billion in ticket sales in 2009. In order to buy tickets for a music performance, sports game or some other event, a user visits the site, types in the name of an event and is presented the option to buy tickets.
While this may seem like a straightforward process, a lot can go wrong, Young explained.
His team examined 2,000 unsuccessful searches to determine how people were failing to get what they were looking for. The idea was to "look at the historical data and let it drive the product that we're building," he said.
The most prominent form of unsuccessful searches was for cities or states. People were looking to find events in a certain area but, at the time, Ticketmaster only indexed artists, venues and events. It soon started indexing locations, and dates as well.
The next largest cause of unsuccessful searches came in the form of misspellings. For instance, the phrase "Circus Olay" is routinely used to find Cirque du Soleil.
Artist names can be a huge source of misspellings The Disney fictional pop starlet Hannah Montana gets spelled in a variety of ways, most involving too many n's and not enough h's. Adding to the confusion, Miley Cyrus, who plays Montana in the show, now performs concerts under her own name, even if many of her fans still look for these events under Montana's.
Also not helping matters are the unique spellings of performers such as Boyz II Men, P!nk and Flight of the Conchords, Even their fans miss the subtleties of these quirky spellings. Users also type unofficial, though widely accepted, names for acts, such as "NIN" for Nine Inch Nails, or STP for Stone Temple Pilots, or Amy Lee, who is the lead singer for Evanescence.
Finally, 10 percent of missed searches came from search phrases with no spaces between them, such as JanetJackson for Janet Jackson.
Young's team began switching over their search infrastructure in 2008. Previously, it used a combination of the Apache Web server, Perl-based scripts and a MySQL back end. Keeping Apache and the Perl scripts, they migrated the system to one that used Ajax on the front end and the Apache Software Foundation's Solr open-source search engine.
Overall, Ticketmaster does not maintain a huge collection of search data -- about 250,000 items, which amount to less than a gigabyte.
What people see, however, is subjected to a wide array of ever-changing business rules. Madonna may add a third show in a city if the first two sell out within a few hours. Björk will play at some Lollapalooza festivals but not others.
Solr promised to provide a flexible platform for capturing all these exceptions, Young said. With the replacement, the team started to add in a number of new hooks that would catch some of the most frequently incorrect searches. They added misspellings into the document keywords. They also developed a synonym list for items that the spell-checker would not catch.
When the system went live in 2009, its integrity came under fire almost immediately. On the day tickets for a Bruce Springsteen tour went on sale, the Ticketmaster site showed no shows available, while at same time linking to a broker site that offered more expensive tickets.
To many observers, it looked like Ticketmaster was giving first preference to the ticket brokers. And Young got some heat for this.
"Bruce Springsteen called a [U.S.] senator, the senator called the head of Ticketmaster. The head of Ticketmaster called me and asked why search wasn't working," Young recalled. He spent three days going through the software and, in the end, concluded it wasn't the search software after all.
Ticketmaster never revealed the source of the glitch, though Young did mention that, of some 30 articles that speculated what went wrong, only one correctly identified the culprit.
"Search is a convenient scapegoat when things go wrong," Young said.
Join the CIO Australia group on LinkedIn. The group is open to CIOs, IT Directors, COOs, CTOs and senior IT managers.