As conflicts in the Middle East and elsewhere continue to drain the U.S. armed forces and the death toll in Iraq and Afghanistan nears 3,000, recruitment of new servicemen and women is an increasing challenge. The task is particularly difficult for the U.S. Army, which depends on its 8,000 recruiters across the nation to deliver quality candidates. It’s a daunting task: The army ended its 2005 fiscal year 7,000 shy of its recruitment quota of 80,000 soldiers.
In recent years, potential servicemen interacted with recruiters over the army’s Web site, www.goarmy.com, posing questions in an online chat forum. But the chat rooms quickly became too difficult and costly to manage. “It was a free for all,” says Gary Bishop, deputy director of strategic outreach for the U.S. Army. “And what we discovered from looking at the chat logs was that about 60 percent of the questions were being asked on a daily basis.” Bishop realized his team of recruiters needed a way to automate answers to common queries, freeing staffers up for more important tasks—namely, one-on-one meetings with candidates who are close to signing on the dotted line. “The idea here is to shift the work effort so recruiters can assist those who are on the fence about joining up,” he says.
Enter Sergeant Star, an artificial intelligence agent—also known as a bot—created to assist the army by fielding general questions about army life. “Sergeant Starr is bullet-proof,” says Fred Brown, president and CEO of Spokane, Wash.-based Next IT, the artificial intelligence firm that created the gruff drill sergeant, and is working to bring similarly customized bots to companies across the country. Using natural language and a high-powered probability engine to determine what users are asking of him, Sgt. Star is programmed with more than 13,000 responses. If he is asked a question he can’t answer, the automated bot offers to connect recruits to a live agent for more information.
To create Sgt. Star, the army and Next IT assembled focus groups to determine what type of persona would most resonate with potential recruits. “We also spoke to actual soldiers and recruiters to make sure that Sgt. Starr accurately reflected the way a recruiter speaks,” Bishop says. Potential recruits go to www.goarmy.com and click on the Sgt. Star window to launch the program. In a separate window, an instant message box appears where users can type in questions. Sgt. Star answers in text along with a recorded voice message. Meanwhile, the appropriate Web page is called up alongside the chat box so users can read more about their query. Relevant links are also posted in the chat window for users to explore.
Of course, Sgt. Star’s answers are carefully prepared. When posed the question if new recruits are likely to be sent to Iraq, for example, his response is: “In times of war, deployment is likely. The possibility of deployment depends on your MOS (military occupational specialty) and unit of assignment. All soldiers must be strong, trained and ready upon assignment to a unit and prior to any deployment, including training exercises, humanitarian missions or combat operations. In general, deployments last a year or less.”
A softball answer to a hardball question, perhaps. But then, Star was designed to put a friendlier face on the army’s otherwise unfriendly exterior. “People like getting information about the Army from the web, because they aren’t getting pushed by a recruiter,” says Brown.
That may be, but as the danger overseas remains high, can an automated bot really do much to increase recruitment levels? Bishop can’t make any estimates. “It’s really more about helping us to improve communication by separating those with basic inquiries from potential recruits who are further along in the decision cycle,” he says. Other measures, including increasing signing bonuses to $40,000, raising the maximum age from 35 to 42, and loosening medical requirements may have helped the army exceed its recruitment quota by 4 percent during July, though recruitment for the army’s national guard and reserves remain low.
