Contact Center Hiring & Retention: Using AI to Predict Which Job Candidates Will Be Successful

WRITTEN BY SCOTT BAKKEN, MAINTRAX

Contact Center Hiring & Retention: Using AI to Predict Better Job Candidates
Subscribe to Contact Center Pipeline Magazine

Do you have trouble hiring quality agents? Retaining the superstars? You’re not alone. Surveys suggest agent attrition is a top concern year after year. And no wonder: the ramification of bad hires is expensive. The average cost to recruit, hire and train each new agent starts around $6,500 depending upon vertical. Poor hires mean higher turnover and more time spent hiring (again!), not to mention the negative impact a bad agent can have on the customer experience and your brand.

Hiring the right people is the best formula for employee satisfaction, agent retention and quality within your contact center. To support this effort, a few cutting-edge contact centers have begun using AI technology that automatically ranks which agents are most likely to be successful. Behavioral Predictive Voice Analytics (BPVA) is a new scientific way to improve hiring practices, retain better agents, generate happier customers and grow a healthier bottom line. Furthermore, the ideal personality traits of agents can be relayed to the recruiting department, enabling recruiters to identify future candidates who are most likely to succeed.

Predictive in nature, BPVA analytics can foresee whether a recruiter is likely to and should advance the candidate. Many companies are already utilizing technology that assists HR teams in automatically prescreening resumes to find those that are qualified, experienced and interested in the position. When those vetted resumes finally get to the hiring manager’s desk, they choose who will be called in for an actual interview, which is where behavior analytics takes things a step further.

Once the agents are hired and working in your contact center, these analytics can continue to be an asset in many ways. What’s contributing to their successes, coaching needs, lack of interest in their job? By analyzing post-hires for indicators of agent churn (emotional state), your organization can uncover where training needs to be improved, those agents at the risk of leaving that need extra attention, and who’s going above and beyond in delivering exceptional customer experiences beyond selling spiffs, hitting goals and fulfilling compliance standards.

It’s Not What You Say, It’s How You Say It

We often think of communication as choosing the right words to say in a given conversation. And while that’s the “hard data” that is easiest to interpret post-call, it’s important to start considering non-verbal cues that are happening and how we can utilize them to make smarter business decisions. UCLA Professor of Psychology Albert Mehrabian found that gestures account for 55% of the impact in a conversation, and that tone of voice makes up 38%. Actual words said? Only 7% of meaning in a conversation.

There’s a lot of research behind this concept. Prosody is the study of linguistics that doesn’t involve the words we use. Instead, it considers the intonation, pitch, volume, tempo and rhythm in our speech to convey different meanings. Try saying the word “Great!” five times followed by “that’s really awesome.” Your voice may have gone up or down on different syllables or maybe your voice was louder, faster or slower. Or imagine two candidates you’re interviewing who say the same phrase, “I’m really interested in this job.” Although they say the exact same thing, the way they say it is more telling and more likely to leave the impression on the hiring manager.

BPVA technology doesn’t track personalities, per se. Rather, it tracks personality patterns as measured on scales such as risk-taking, well-being, conscientiousness, positive behaviors and coping abilities. The profiles created reflect aspects such as the temperament of the person; the social behavior of the person; or thinking and acting patterns—do they seem more systematic or more associative? The list goes on—you can predict traits such as dependability and personal integrity, too. Imagine the impact this can have on hiring and retaining quality agents.

BPVA capitalizes on elements of these human voice characteristics and provides a completely new data source from which insights can be gleaned. When you start picking up on tone (is it happy, upset, timid, honest?), inflection (are specific syllables being emphasized in a question or does it sound monotone?), and volume (is shouting controlling the conversation?), you can learn how powerful the voice can be in self-expression and in creating profiles for individual behavior. As the old saying goes, it’s not what you say, it’s how you say it. Ian Jacobs, Principal Analyst from Forrester Research says, “From my point of view, the biggest benefit for this approach is the ability to impact the emotional component of service interactions.”

Voicesense, a company among the BPVA industry leaders who are developing screening tools for prehire analytics, matches behavioral patterns of the best agents and detects early signs of burn out among veterans. Yoav Degani, Voicesense’s CEO, says that “our solution makes applicant screening more effective as well as monitoring for changes in state of mind. We can match key behavioral tendencies to improve better hiring and improve retention. One of our clients is a recruiting firm that reduced screening time by 70% and now only evaluates the candidates with the top 10% of the scores calibrated to the patterns of their best agents.”

Interestingly, because prosodics involves the analysis of speech parameters and not the content of conversations, the same set of algorithms generate nearly the same results regardless of language or culture. For instance, an organization can use it to accurately predict future consumer behaviors without relying on demographic or historical information. That means a contact center receiving calls from both English- and Spanish-speaking customers doesn’t have to rely on English-only transcripts to extract insights.

New Technology Hits Contact Centers

By characterizing patterns in behavioral and personality traits through the innovation of combining psychology, speech analytics, artificial intelligence and signal processing, this new technology can foresee how customers and employees will behave and answer questions in real time such as “which loan prospects will default,” “who likely will pay their debt,” “which of my new agents will burn out,” “which customers will buy,” and “what do the best sales people sound like?”

The difference lies in the unique approach of evaluating many speech parameters, depending on the technology, which aren’t detectable by simply reviewing the words spoken in a transcript. By doing this, a unique profile of a person’s behavioral inclinations can be applied to many situations and industries. If you know what behavior is predicted to follow a conversation, you can make smarter business decisions at a quicker rate. In fact, what’s being said is so inconsequential, testing has shown that call outcomes can be derived within the first 45 seconds of the call.

Think your own intuition coupled with speech analytics technology could effectively determine which combination of prosodic characteristics might impact the results? We once did, too. Years ago, we at MainTrax used the data available at the time to formulate a hypothesis of who would be the best sales agents. Our initial hypothesis proved to be flawed. Had BPVA been available at the time, it’s likely that our first assessment would have been closer to the mark.

BPVA has the potential to significantly impact other use cases:

Improving NPS

Forrester Research claims that emotion is the largest driver of a customer’s perception of their experience with a brand. According to Jacobs, “Customer service leaders have worked hard to solve customer problems quickly and easily over the past several decades, but they haven’t given as much thought to how the service experience makes those customers feel. The behavioral cues offered by these systems allows agents to influence emotion during the interaction. Humana, for example, uses this approach and has found that customers who had agents handling their calls using these emotional cues gave a 28% higher Net Promoter Score (NPS) than those who didn’t. That’s a really significant benefit.”

Reducing Customer Churn

Imagine two customers calling in to cancel your organization’s services. If they each say, “I need to cancel,” analyzing that request could offer completely different messages if one is speaking in an angry voice after being dissatisfied with the service, and the other is relaying their message in an even tone due to simply not needing the service anymore.

Identifying Medical Conditions

Considering how people interpret non-verbal cues, it’s not surprising that prosodic speech analysis caught the attention of large medical companies like UnitedHealth Group and Mayo Clinic to predict medical conditions such as Alzheimer’s and depression. CompanionMx analyzes patients’ voices for mood and behavior to detect depression and bipolar disorders.

BPVA: Complement or Alternative to Traditional Speech Analytics

What’s the differentiator between your successful and struggling agents? Is it found in tangible data such as call duration, average handle time or words spoken? Traditional speech analytics is riding the tail of the big data revolution, which focuses on speech-to-text transcription. Amazing strides have been made in NLP and traditional speech analytics, but the limitations of those analytics have also been revealed. Different languages, accents, sarcasm and joking are some of the challenges related to using contextual speech analytics. By going beyond the transcription, the insights developed from elements in the voice can also generate customer experience insights that aren’t available by looking only at the words spoken.

Then where does traditional speech analytics technology come into play? After all, traditional tools continue to be the workhorse of contact center quality managers, data scientists and marketers. And for good reason. Speech analytics have proven to be extremely effective when it comes to identifying:

  • Agents’ lack of knowledge
  • Complaint handling
  • Unresolved issues
  • Unprofessionalism
  • Escalation requests
  • Soft skills
  • Agent frustration
  • Customer frustration

Or does it lie in the nonverbal elements of a conversation? Remember, 38% of impact in a conversation lies in tone of voice. Superstar agents likely possess exceptional voices. They please your customers and because they perform well, you want to retain them longer. Those who are less adept at vocal communications and problem-solving usually perform poorly. Customers are dissatisfied and may leave. These reps must be replaced. Looking at their word choice alone might not reveal these issues as quickly.

It should be noted that the traditional speech analytics industry hasn’t ignored prosodic markers. In fact, many vendors offer “sentiment analysis,” which supposedly factors in non-contextual cues. But we at MainTrax have found the results of sentiment metrics to be mixed. Furthermore, some assumptions based on sentiment analysis can be flawed.

  • Fast pace could demonstrate positive energy, but it can also demonstrate agent nervousness and defensiveness.
  • A flat, monotone voice could convey boredom, but it could also convey a measured and steady response to an agitated customer.
  • A high-pitched voice may indicate enthusiasm, but it may also demonstrate anxiety.
  • A strong tone could signal confidence, but it could also be a sign of anger or conflict.

But, unlike conventional sentiment scores, BPVA takes specific business objectives and call outcomes into consideration when models are built and lets the data determine the appropriate scoring algorithms. In cases such as predicting agent success and identifying fraudulent callers, that’s all that may be necessary.

Speech analytics and BPVA are strong on their own, but when used together some extremely powerful insights can be uncovered.

In other cases, traditional speech-to-text technology helps to understand why. Daniel Ziv, VP of Customer Analytics at Verint, says, “Improvements in noise-canceling headsets and recent advances in speech technology including deep learning have brought new attention to these unique acoustical cues. Speech analytics vendors, including Verint, are now looking more closely at incorporating additional acoustical elements and combining them with linguistic speech analysis.”

Scott Kendrick, VP of Marketing at CallMiner, suggests that, “Acoustics and prosody are beneficial primarily in understanding the strength of an emotion, but not necessarily the sentiment. Semantics have been proven to be the most accurate method of measuring one’s sentiment. The tonality or way in which something is said helps us understand the degree or level of that sentiment. Combining the two will give us the most accurate method of measuring emotion.” Kendrick added that context is important to consider, too.

Conclusion

  • While far from perfect, traditional speech analytics technology remains the bedrock of contact center insights and for good reason.
  • Emerging technologies such as prosodic sciences fill an important gap because data shows that it’s not always what you say but how you say it.
  • Behavioral Predictive Voice Analytics by itself is proving to be an effective way to hire agents, retain superstars, improve customer experience, predict loan default, improve debt collection, identify fraudulent calls and more.
  • Contextual patterns identified using traditional speech analytics technology combined with BPVA cues from the voice patterns can generate even more powerful insights.

About MainTrax
MainTrax is a visionary team of experts uniquely qualified to help companies maximize speech analytics technology and transform customer data into smart, actionable business solutions.


Scott Bakken, Founder and President of MainTrax, is a highly respected independent voice in the speech analytics industry. His company provides professional services that help end users use their existing speech analytics tools to deliver actionable business intelligence. Bakken was named an Ernst & Young Entrepreneur of the Year finalist.