The social network says acquiring Gnip will help companies better understand what consumers and other brands are saying across Twitter.
They’re not really customer service reps—but they play the part on the web.
What’s more up-to-the-minute, postmodern than the Internet? But the success of one of the latest retail applications on the Internet depends on human behavior formed in caveman days: the rules of social interaction. For instance, survival needs have hardwired into the human brain that both good and bad information have more impact when delivered from a face at close range to one’s own. And looking down into a face conveys a sense of control that’s missing when forced to look up.
Guess what: it works that way online, too. That key finding of scientists at Stanford University is being developed into a web application by Westminster, Colo.-based Finali Corp. Finali’s NetSage, a virtual agent that appears on web sites as a human figure, can be deployed to assist online shoppers through complex purchases like ordering an audio system or to field customer queries that might otherwise require live human service at $5 to $10 a pop. The humanized presence ups the effectiveness of what’s essentially online customer self-service and automated cross- and up-selling functions, researchers say, by eliciting responses from users that are so deeply rooted in the rules of social behavior they’re instinctive.
“We can rationally say that an image that’s just a bit map on a screen isn’t going to make anyone feel god or bad, but that’s not the way we’re built to respond, and that response is applicable to media,” says Byron Reeves, director of Stanford University’s Center for the Study of Language and Information and an adviser to Finali. “Many of the features of this technology are so similar to real human-to-human interaction, especially when a face is displayed, that the same rules come tumbling into play.”
Taking the cost out
Virtual agents have come a long way since Microsoft embedded its annoying paperclip character in software in an attempt to help users performing simple tasks, such as writing letters, on computers. The proof is in early results from marketers already using or testing NetSage technology on their own sites. The NetSage figure added to Buy.com’s web site last summer has had a positive reception among customers, as measured in its resolution of about 23% of the customer e-mails that would otherwise require human help, and in the resulting reduction of customer service costs from about 2% to less than 1% of Buy.com’s revenues. In a test at Dell.com, a program more oriented to cross-selling and up-selling, a NetSage character created especially for the computer manufacturer boosted order size by about $50 and doubled customers’ willingness to accept product recommendations presented online.
Critical to NetSage’s effectiveness is the casting, character development, stage direction and talent of the actors selected to play NetSages on different sites. Getting from concept to finished product online is a process that at times has seemed more like a Hollywood production than product development at a technology company. When a company orders up a NetSage for its web site, Finali starts by working with the marketers to develop a persona for the character that meshes with the company’s brand, customer profile and program goals. “Ian Stone,” for example, the NetSage figure that launched on Buy.com last July, reflects Buy’s typical customer: young, male, and techno-savvy. The character’s “bio” as posted on the site uses some Southern California jargon to give the figure’s personality an extra bit of edge.
Once the character, backstory, and sense of how the Net Sage should look and even dress are developed, Finali checks its bank of images. The company can choose from thousands of stills captured from actors it’s videotaped in a series of motion sequences. Under a producer’s direction, the actors perform tasks that a real-life customer service rep or sales agent would perform when interacting with a customer, such as looking up product specs in a catalog, encouraging the customer to complete a form, even delivering the bad news that an order is lost.
Extracting stills from a motion sequence rather than asking the actor to strike a pose invests the NetSage with an extra degree of realism that makes its exchange with an online customer more compelling, Reeves says. The stills capture actors in dozens of different behaviors and facial expressions and at different distances from the camera to evoke different reactions. The images are matched to text responses and prompts that the NetSage can give online, depending on the course of its interaction with the customer.
Finali also sends out casting calls when a company wants to look beyond the stock library to find a NetSage, as Dell did. “That leads to us sitting down with about 1,000 8-by-10 glossies to look for people who look like our character,” says Reeves.
A consumer test provides the feedback that puts the finishing touches on any NetSage before it launches online. Testers see different versions of the NetSage as played by different actors. Finali tapes the testers’ reaction to each one to analyze it, which leads to the final selection of which actor will play the NetSage for that project, and a final decision on which digital photo images will be used to support each piece of online communication. Consumer tests guided the developers to wardrobe changes for one NetSage, for example, when testers registered the opinion that a Sage wearing a shirt with a corporate logo was less credible than the same character in a plain shirt. Consumers also had a preference for certain angles, feeling more in control of the interaction when the Sage was depicted at a slight downward angle than when it appeared the Sage was looking down on them.
And though the consumer tests are late in the process, the results can still lead to dramatic changes as the 11th hour approaches. Take the pilot for Dell’s NetSage, who’ll be identified on the site as “The Product Adviser.” The character was originally conceived by the development team as a Texas-based ex-college athlete with a strong interest in computers. The developers presented the test panel with a couple of versions of that character and threw in a few that didn’t fit that description just to see the reaction. The testers’ overwhelming favorite was one of the “extras” that did not fit the original profile.