The headline might infer that the article outlines Thursday's major power outage that cut electricity to more than 1.4 million homes and businesses in California's South Orange County and San Diego County, as well as parts of Arizona. Commuters were stuck on empty at gas stations, ATM machines couldn't spit out cash, and traffic snarled to a near standstill because signals didn't flash green or red.
On the contrary, today's post calls attention to the inability of search engines to serve up relevant results in one query -- no matter how much personal information a searcher provides.
On Thursday, a friend calls to ask me a favor. He's driving around with his son looking for a store selling Halloween costumes, and only has one hour before returning him back to school. His phone call requests my help in doing a search online for local stores selling costumes and beards. I'll admit -- when in the car he relies on me and my iPhone to find local stores, restaurants and other locations.
While he holds patiently on the other end of the phone, I launch a Chrome browser and Google search engine on my computer, and type in "Halloween costumes Huntington Beach." The engine serves up six phone numbers and store locations from Google Places at the top of the query, but all six have been disconnected, connected to vacant physical locations, or are located miles from Huntington Beach. These listings in Places were not marked as "closed."
Several searches later, I find a local store selling beards, but must try several different keywords to arrive at the results. While my friend did find the costume before taking his son back to school, my frustration arose from thinking the process should take less time -- less queries. The first query should have returned more relevant, targeted results, but instead I got outdated information.
In fact, I not only got outdated information, but had to query several keywords I would not have normally used to find the beard, which is probably a result of the stores optimizing content with non-friendly consumer keywords.
Why all the problems for older businesses that might be seasonal or go out of business? Mike Blumenthal -- partner at Blumenthal.com, who specializes in local optimization -- explains that Localeze, InfoGrid and other primary list generators sell business listings to business list sites that repurpose and resell to Yelp, Facebook and others. Google takes the lists from feeds and other sources, triangulating the data, and compares it with what its bots find on the Internet. While the system works well to bring in new numbers and listings, it doesn't do as well when companies close or disconnect their phone number.
"Engines are not good at getting rid of old numbers," Blumenthal says. "Sites like InfoGrid could take 18 months to two years before double-checking a listing and making the change."
Incorrect business listings are a reflection of the noise available on the Web, according to Gib Olander, vice president of market development, Localeze. Today more than ever, there are competing sources with varying degrees of accuracy online, he said. Halloween stores or Fourth of July firework stands are notoriously difficult, especially when relying on a wiki approach -- as these stores do not provide audience-based or Web-based signals, positive or negative, of their status for nine months of the year.
Olander said it's a classic illustration of why a wiki approach to data quality only works for urban areas or high-traffic business categories, not for Halloween or July 4th.
So, Google, Bing, and Yahoo -- can you continue to improve search accuracy, please? It shouldn't take me more than a couple of tries to find a Halloween store selling costumes.