Skip to main content Skip to footer

The problem had been bubbling for a few years at our company: First-year turnover among a dozen or so customer-facing roles in our call centers was high. The irony wasn’t lost on us. If we want to deliver a stellar experience for customers, we also need to deliver a stellar experience for our employees who interact with our customers every day.

We thought we knew the answer.

We didn’t. This is the story of how we found it.

A problem that wasn’t getting better

Call center turnover is high everywhere, and it was no different for us. However, that’s not where our story ends, as we weren’t willing to accept that as inevitable. As we reviewed several years of artifacts that had accumulated in our efforts to tackle the turnover problem, we realized our hypothesis had been the same each time: It was a talent acquisition problem. Each time, we’d set out to fix what we thought was wrong with recruiting.

The familiar pattern repeated itself numerous times: When turnover spiked, we’d make a tweak somewhere in the recruiting process and then walk away. But that approach wasn’t working - the high turnover in our call centers kept reappearing. To make real, sustainable change, we knew we’d need to take a new approach.

I made a suggestion that, while logical, was contrary to everything I’d been taught as a researcher and statistician: Let’s go in without a hypothesis. It was a record-stopping moment. Most of us grew up learning and teaching the scientific method, where you develop a hypothesis and leverage data to prove or disprove it.

While we’re a data science environment, we weren’t using data science to its full extent to get to the root cause of our problem, often relying instead on assumptions.

We decided to toss out the con- ventional playbook and pursue a largely untested, AI-driven path. It was unpopular, but fortunately, we found a partner who said, ”No hypothesis? That’s OK.”

Working without a net – or a hypothesis

Our teams got to work. We developed a problem statement and then performed an extensive qualitative analysis that included talking to our employees across the country, reaching more than 700 employees and managers in customer-facing roles.

We pored through project archives from the past decade to find out where earlier efforts had gone wrong. To uncover best practices, we benchmarked companies with lower call-center turnover through phone interviews and site visits.

We partnered with Cognizant to conduct the quantitative analysis. The team deployed innovative machine learning models to analyze over 250 data elements, drawing from our own data as well as third- party sources. Sifting through the information, we identified retention and turnover indicators for customer-facing roles.

The models we created predict retention and turnover with 94% accuracy. To find key themes and employee sentiment, the team leveraged text analytics from multiple employee listening posts and job search sites, including Glassdoor and Indeed, as well as over 100,000 comments from our annual employee survey.

It was the first time The Hartford utilized employee-related information as input into AI algorithms. Our examination of retention data from a data science perspective was eye-opening: We created new variables, including looking at why people stay. When we started looking at the results, the data that popped the most – regarding career development, remote work, rewards and recognition, and digital badging – was information we weren’t always looking at on an ongoing basis.

We’d been missing critical indicators for retention and turnover that, individually, generate a 30% to 50% lift in employee retention.

By working from a blank slate, we could let data guide our decisions – not intuition or gut feelings. And as it turned out, it wasn’t about recruiting; it was about the experience once the employee arrives.

Taking steps toward a culture shift

Following this discovery, 2018 was a year of analysis and understanding at the local level. In 2019, we’ve been focusing on retention, recruitment and taking action.

One of the first changes was to our dress code. In place of a list of do’s and don’ts, we adopted a simple statement: Dress appropriately for your day. It sounds minor, but what we’d heard from employees is that while they believe in The Hartford, our brand and their co-workers, the work is complex, and the policies are sometimes rigid, like our former dress-code policy.

As a result, we’re shifting our culture. We’re becoming more contemporary, and the dress code change is an important visual indicator that everyone can see. We’re incorporating the changes not only into our employee experience, but we’re also making them part of our training classes, onboarding teams, first-day experiences and internal communications.


Figure 1

We’re also moving toward empowering managers to make local decisions. For example, managers can now use their own discretion to reward employees as part of our “on-the-spot” recognition program rather than requiring approval. It sends a message of trust to our employees and managers, and it speeds up the recognition process.

We also created retention risk models – a framework that, as an insurer, is very familiar to us. These models provide a helpful lens through which to view potential turnover issues and help us pinpoint the variables that indicate an employee may be looking to leave.

Operationalizing the models

We decided to operationalize the models we developed, taking the insights and data gained through the analysis and creating an interactive dashboard for leaders to use. The dashboards display attrition and retention indicators at both department and team levels. The data, which is refreshed every month, is sourced from predictive models in our analytics computing environment.

The dashboard is an important part of sustaining progress; we’re taking action on root causes and will continue to monitor and assess attrition risk and retention.

We’re taking a rolling adoption approach for the dashboard and deploying in small pilots that will give us an opportunity to refine the training and tackle some of the more complex cultural changes, such as how to ensure managers with higher-risk employees are taking the right actions and not giving up on their employees.

The results have exceeded our expectations. Taking action on indicators such as career development, digital badges, on-the-spot recognition and remote work has already improved retention over 20 points. We’ve also seen improvements in our Glassdoor and Indeed comments and scores.

Letting go of our long-held hypotheses has given us new opportunities. Using innovative, AI-driven tools, we can now identify ways to improve employee retention and, perhaps more importantly, model for our employees the experiences we envision for our customers: people- focused, engaging and positive.

Author

Michelle Deitchman is the Head of Employee Experience & Insights at The Hartford. She has over 20 years of experience in the areas of process improvement, data & analytics and program leadership. Prior to joining The Hartford, Michelle held leadership roles across multiple industries and functions including manufacturing, engineering, technology, operations and financial services. Michelle earned her BA in international business & management and psychology from Dickinson College, her MS in industrial & organizational psychology from Springfield College and her MBA in operations management & marketing from Rensselaer Polytechnic Institute. She can be reached at https://www.linkedin.com/in/michelle-deitchman/.

To learn more, read the full Cognizanti journal, Navigating the digital age: What senior leaders worldwide have learned from pushing the boundaries of change, or contact us.