When the Hiring Boss Is an Algorithm 245
Hugh Pickens writes "Joseph Walker writes at the WSJ that although personality tests have a long history in hiring, sophisticated software has now made it possible to evaluate more candidates, amass more data and peer more deeply into applicants' personal lives and interests. This allows employers to predict specific outcomes, such as whether a prospective hire will quit too soon, file disability claims, or steal. For example after a half-year trial that cut attrition by a fifth, Xerox now leaves all hiring for its 48,700 call-center jobs to software. Xerox used to pay lots of attention to applicants who had done the job before. Then, an algorithm told the company that experience doesn't matter. It determined what does matter in a good call-center worker — one who won't quit before the company recoups its $5,000 investment in training. By putting applicants through a battery of tests and then tracking their job performance, Evolv has developed a model for the ideal call-center worker (PDF). The data recommend a person who lives near the job, has reliable transportation and uses one or more social networks, but not more than four. He or she tends not to be overly inquisitive or empathetic, but is creative. 'Some of the assumptions we had weren't valid,' says Connie Harvey, Xerox's chief operating officer of commercial services. However, data-based hiring can expose companies to legal risk. Practices that even unintentionally filter out older or minority applicants can be illegal under federal equal opportunity laws. If a hiring practice is challenged in court as discriminatory, a company must show the criteria it is using are proven to predict success in the job."
Not an algorithm... (Score:3, Informative)
Since it doesn't guarantee success, it's a heuristic...and I wouldn't trust anybody trying to sell me one, who doesn't know the difference
In favor of algorithms (Score:2, Informative)
I know there is a general backlash to the increasing use of algorithms in determining major decisions such as hiring. However, from a quantitative standpoint interviews have been shown to be extremely inaccurate as a judge of future job performance. There are simply far too many opportunities for bias on the interviewers part and so they tend to be neither reliable nor valid. Irrelevant characteristics such as appearance end up having far too much weight due to the halo effect. If you want the best result, depending on faulty human judgement is often the wrong choice.
For example, the Apgar score [wikipedia.org] for judging the stability of newborn babies was designed to combat biases on the part of delivery room doctors. Prior to the use of this score, doctors rated how healthy newborns were based on a wide-range of criteria, and each doctor did it differently. When the Apgar score was introduced, it standardized the process by rating newborns on five categories: skin complexion, pulse rate, reflexes, muscle tone, and breathing. The result was that the error introduced by human bias was reduced and countless babies have been saved by quick intervention.
I work at Evolv (Score:5, Informative)
I'm an industrial/organizational psychologist at Evolv. I help build assessment content and I work closely with our predictive algorithms. A few clarifications from the WSJ article & responses to /. comments:
Yes, creativity and empathy are important for some positions, even in call centers! We're not looking for hateful drones who will hang up on you when you call in. In addition to staying longer, our recommended hires perform better as well. That means increases in both customer satisfaction and efficiency (we call it "average handle time"). But it's a curvilinear relationship - somebody who is too inquisitive is going to tend to waste your valuable time (and their employer's) while trying to resolve your issue. There's a balance.
Most test vendors put a test in place and walk away. At Evolv we take all the post-hire data from our clients and continually feed it back into our algorithms. The content, scoring, and weighting adjust over time to be more predictive.
At Evolv, we don't pair obvious responses when we create questions. So no "I like to steal office supplies" vs "I always show up to work on time" questions. Coupled with the continual refresh & validation of the content, there is no "answer key" that will get you a job. One of the neat things about this approach that we've found is that people applying to entry level positions often don't know what they're good at. Either they've bounced around a few jobs or they're just out of high school. So when somebody applies to a call center job that's hiring for both customer service and sales positions, and we can recommend the position for which they're likely to be "fitter, happier, and more productive"... that's kind of cool. Their employer will make more money off a more stable employee, and the employee ends up doing something they will enjoy just a little bit more. I know some folks will see it from the Radiohead point of view, as creepy (and I respect that), but we think it's better than dumping somebody into a position they're not going to enjoy just because they had the right keywords on their resume or they BS'd their way through an interview.
Science & statistics help eliminate some crazy gut-based hiring decisions. Some hiring managers want to ask call center applicants what they'll be doing in 10 years with an expected response of "I'll be working at this call center". But let's be realistic - while some people enjoy them and thrive, call center jobs are typically not where you plan to be in 10 years. We've also found that resume experience for entry level positions is less important than basic skills and attitude. It's easy to look at that and say "duh" but you'd be surprised how many people hiring & screening for these roles want to exclude applicants who don't have prior experience. So we can cut things out of the interview and hiring process that just don't mean anything.
Evolv doesn't just do employment screening. We periodically follow up with people after they're hired. We find out what information wasn't communicated well during the hiring process, get their feedback on how their training is going, their thoughts on their supervisor, that sort of thing. We feed all of this back in to improve the process. In some cases, that means identifying the trainers whose students perform poorly when they start working. Other times it could be flagging a tenured stellar performer whose numbers are starting to dip for a new position to help reinvigorate them. We strive to improve profitability across the workforce, and do so in an employee-friendly way.
Last but not least, we're still expanding through Xerox, so if you've called their customer service and had a bad experience it must not have been one of our hires. Joking aside, agents are people too, and even our top recommendations have a bad day. We're working hard to to make it better though!
Hope that helps! Yes, there definitely are risks with employment testing, but we try to avoid them and build solutions that make everybody's life a little better.
Cheers,
Tim
Re:Tell me about it (Score:4, Informative)
LOL I'm sure they'd get right on it!
State legislators, particularly Representatives, tend to be a whole lot more responsive to their constituents than do their counterparts at the national level, for the simple reason that they represent a lot fewer people. For example, in Colorado, we have about 5.1 million people and our House of Representatives is 65 people, which means each Rep has about 78,000 constituents, of whom about a third are actual voters (going from turnout figures in recent elections). Those are numbers small enough to get some real attention when a constituent has a problem, and I know several people who have done just that.
Re:Reason is simple: U.S. Workers are stupid (Score:2, Informative)