Article : Predicting Agents Most Likely to Pass Exam
Early in my experience with call centers, I encountered a curious mystery. This tightly run service center had an ocean of standardized cubicles, punctuated by manager turrets. Each service agent had the same computer screens, the same scripts to follow, the same goals and the same incentives for success. They handled high call volume with little variety.
Yet - the success of these agents was remarkably varied. Some were unstoppable high volume call machines, others handled less volume and were mediocre, others quickly burned out and had many dissatisfied customers. These agents were each saying the exact same words to a wide sample of callers, but showing very different results.
This was what economists call a "natural experiment" - everything was set to be equal, except for some "it" factor inside the agent themselves. And, that "it" factor consistently delivered results dramatically affecting service levels, volume, compliance, errors, single call close and attrition metrics. What was "it?"
Other business domains use advanced analytics to find and predict these kind of factors. They use these factors because they are so predictive of success.
Marketing, in particular, has made a science of sorting out consumer signals into personas, to optimize outbound messaging and offers. Medicine, finance, elections, and industry all use similar predictive approaches to predictive how people will behave.
We can apply the same analytical methods to find and foster this "it" factor for call center agents. What would this look like?
Example: Preventing Early Dropouts
Most contact centers spend weeks or months training agents. Some, for example in the financial, insurance or other industries, must complete difficult certification exams before an agent can even take an initial phone call or touch a headset.
Many of these centers find a disappointing and expensive number of agents who terminate during training, or within weeks of its completion. This is a worst-case scenario for the center - all expense, no value.
A predictive approach would evaluate behaviors, aptitude, and other factors for candidates who made it to this "three weeks past training" window, versus those who didn’t. We would build and validate a rigorous model based on these factors that is able to quantify the "it" factors that tend to repeatedly pass training and successfully migrate to phone work.
The most powerful way to use a predictive model is for pre-hire candidate selection. Each new candidate would be tested to the model, which would calculate a predictive score. In this case, the score would be the probability of passing training and getting onto the phones. So, one candidate may have a 73% chance of passing the exam, and another 32%. (Or one could predict a candidate’s Net Promoter Score, or likelihood of staying in the role for at least a year - all pre-hire). Usually, we see "bands" of acceptance based on this number, which guides the recruiting team along with other factors.
By selecting candidates that are more likely to pass training, we get more agents past that hump and onto the phone where they begin to provide value to their employer. Fewer candidates waste their own time in a career that they ultimately don’t want. The contact center spends less money acquiring, recruiting, training agents. Performance goes up. Engagement goes up. Attrition goes down. Customers feel it.
A Learning Model
The predictive modeling process brings ongoing feedback to the recruiting and hiring process. It connects hiring with what happens when they land in their call center role. Math and modeling methodologies keep everyone honest; if something is predictive, we keep it. If it is random and doesn’t predict, we remove it.
A quality model also informs recruiters of the qualities that matter and others that don’t. They can even inform advertising and sourcing decisions.
This approach can be additionally useful for entry-level candidates with little job experience - there isn’t much to go on except for that "it" factor.
Of course, we have bigger goals than just passing training. We want agents to have high customer satisfaction, to sell, to follow rules and process, to have low call times, and single call closes and to stay on the job for years. Your center will have more.
Models can, and should be built to address these KPIs. Some will be predictable, others not. The goal is to use predictive tools to screen in both a) high potential candidates, and b) to screen out low potential candidates. Your approach will vary with call center demand, job market conditions, and changing business imperatives.
The smartest users of predictive models have a portfolio of predictors for each candidate, so that hiring professionals can intelligently balance potentials and business needs.
Call centers are ideal for this kind of work because of the performance data they gather today. Almost all of what is needed exists somewhere and can easily be extracted. It’s gold waiting to be mined - so you can solve the call center hiring mystery.
Today's Tip of the Day - Back To The Basics
More Editorial From Talent Analytics, Corp.
About Pasha Roberts:
Pasha Roberts is chief scientist at Talent Analytics Corp., a company that uses data science to model and optimize employee performance in areas such as call center staff, sales reps and other high volume roles. He regularly writes on predictive workforce topics and speaks at major predictive analytics events like Predictive Analytics World for Workforce, INFORMS and many others. Follow Pasha on twitter @pasharoberts.
About Talent Analytics, Corp.:
Talent Analytics, Corp. uses advanced Data Science to predict employee flight risk and KPI performance, both pre- and post-hire. Our predictive scoring algorithms are easily deployed into the employee sourcing, recruiting, hiring, and operations process via our award winning cloud platform, Advisor(™).
Published: Friday, January 8, 2016