Michael Ringman, CIO at contact center and IT services provider Telus International, has his hands full.
Ringman, who reports to the CEO, oversees the company's internal IT operations, with a team of about 300. He also helps lead a team of 3,000 who provide expertise to customers in the throes of digital transformation efforts. And he oversees a majority of Telus International's data scientists, who are experimenting with advanced analytics techniques such as machine learning.
Here, Ringman describes a machine learning use case IT has taken on to curb call center agent attrition rates -- a major issue in the call center business.
Editor's note: This conversation has been edited for brevity and clarity.
As the leader of internal IT operations, how are you bringing artificial intelligence into the enterprise?
Michael Ringman: Our internal IT role is to create as efficient a call center with the practices within our organization as we can. So, we've started to apply some of the, I'll call it statistical analytics, to things like our team member attrition. Can we better help identify that an employee may leave our organization earlier? And can we start to red flag that across the organization so we can have a positive impact on that team member and convince them to stay?
In the call center business, that has huge implications. One of the biggest costs is team member training. In most organizations, they're lucky if a call center team member stays in that role for seven months. It's viewed as a transitional-type role, regardless of where you're delivering it from. At [Telus], because of some of the ways we go about embracing our team members and looking at things like how we can leverage machine learning across our HR portfolio, we can better identify those team members. We now have an agent attrition rate that is half that of the industry average.
I wouldn't say the 50% less attrition was all due to machine learning and data analytics, but we are definitely looking at those types of things internally to better help us understand how we can reduce our overall cost and add value back to our customers.
What's a challenge your team of data scientists faces with this particular machine learning use case?
Ringman: There are a ton of data sources that we could look at in regards to team member attrition -- everything from their resumes when they come on, to how were they onboarded, to how did they go through their initial training, to who is their supervisor -- all of this sort of data that resides out there.
So, we go to the HR team, who says, 'Wow, this is great. Let's really dive into this.' We're trying to find out how we can be sure that our analytics are properly identifying the team members who are going to attrit. So, we quickly need to narrow our focus and get to a small subset [of employees who could attrit]. Do we start to look at a particular training class, and can we more narrowly define some of the potential indicators? [People know] with AI that it isn't necessarily going to be easy, but you better be prepared to fail a bunch.
As you start to go through and look at what you think might be early indicators for attrition, you may not find that team member training has anything to do with team member attrition or maybe it has a tremendous amount of impact. So, you need to go through, narrow down and put your arms around the types of data sources needed to identify something as complex as a team member attrition rate. At the end of the day, you may be able to identify within a certain parameter, but you're never going to get 100% of it right.
And in this particular machine learning use case, how do you identify the parameters?
Ringman: So, the way we've done it is to start very narrow -- we create a hypothesis, which is really the IT hypothesis that we can help identify team member attrition through all of the data and analytics that we have. So, we put the analytics data sets in, the data scientists leverage mathematical algorithms to start to -- based on those analytics -- predict what that subset [of potential employees who could attrit] is. They start to look across those analytics sets to create a model that can predict team members who attrit, and they compare the output to what actually happened. So, if the model looked at 100 team members, flagged 10 members and seven of those members actually attrited, that's a pretty high success rate.
That is a [rearview analysis]. Now, we start to apply those findings against new data sets and try to get more and more proactive. And we continue to grow the data set. So, now we've got something that says, 'OK, 70%, 80%, 90% probability that we can identify those team members.' Now, we can get back to the HR team with some clear indications -- these are strong indicators that you're going to have a higher propensity for team members to leave -- and start to identify some of those causalities that just a person looking through all of this data would have difficulty uncovering.
You start out small, you start to grow your data sets and suddenly that machine learning, maybe it doesn't produce 90% or 100% accuracy, but now you can see better trends across the organization.