ASSURANT SOLUTIONS sells credit insurance and debt protection products. Maybe you’ve bought a product like theirs. If you lose your job or have medical problems and are unable to make a credit card payment, Assurant Solutions will help you cover it.
Like a lot of insurance products, payment protection is a discretionary add-on often made at the point of purchase. But when customers get the bill and see the additional fee of, say, $10.95 per month for payment protection, maybe they think, “Well, I’ll take my chances” and decide to cancel.
When those customers call, they reach Assurant Solutions customer service representatives, because the company manages insurance activation, claims, underwriting and customer retention (for many industry-leading banks and lending institutions).
It’s in that last piece — that attempt to retain customers, beat the churn and stem a high exit rate — that Assurant Solutions faced a now-universal management challenge. As a call center positioned as the pivot point of all customer interaction for its clients, Assurant had access to hoards of data as well as the ability to create the kinds of rules and systems that any operationally optimized call center would deploy. With skills-based routing, customized desktops with screen pops, and high-end voice recording and quality assurance tools, its efforts were state-of-the-art.
But it wanted to do better. Its 16% retention rate was consistent with the best industry standards, but that still meant that 5 out of 6 customers weren’t convinced to keep their coverage, let alone consider other products. That’s a lot of room for opportunity.
So Assurant Solutions tried something new: deep analytics. And it invented an operations system that capitalized on what the analytics prescribed.
The result? The success rate of its call center nearly tripled.
What Assurant Solutions found was that all the conventional tenets about contact centers “are not necessarily wrong, but they’re obsolete,” says Cameron Hurst, vice president of Targeted Solutions at Assurant. Hurst previously headed up development for HSBC’s Indian offshore Global Technology group and served as HSBC’s group head of contact center technology after HSBC acquired the call center software company he founded in 1992, so he was already expert in getting the most out of data to run call centers. Or so he thought.
But, he says, “we operated under the fallacy — and I believe it’s fallacious reasoning — that if we improve the operational experience to the nth degree, squeeze every operational improvement we can out of the business, our customers will reflect these improvements by their satisfaction, and that satisfaction will be reflected in retention. And that was fundamentally wrong. We learned that operational efficiency and those traditional metrics of customer experience like abandon rate, service levels and average speed to answer arenot the things that keep a customer on the books.” Assurant Solutions was looking for the key to customer retention — but was looking in the wrong place.
So management attacked the challenge from a different angle. They brought in people like mathematicians and actuaries — people who didn’t know anything about running call centers — and they asked different kinds of questions, using analytics to answer them. “We’re an insurance company,” Hurst says, “so it’s in our DNA to be very data-driven. We are able to look at large volumes of historical data and find ways to mine for gold nuggets or needles in haystacks. But this use of analytics was fresh for us.”
What they found surprised them. In a sense, it was simple: They found that technology could assist the company in retaining customers by leveraging the fact that some customer service reps are extremely successful at dealing with certain types of customers. Matching each specific in-calling customer to aspecific CSR made a difference. Not just an incremental difference. A huge difference. Science and analytics couldn’t quite establish why a particular rapport would be likely to happen, but they could look at past experience and predict with a lot of accuracy that a rapport would be likely to happen.
In the interview that follows, Hurst explains how Assurant Solutions figured out the right questions to ask, used analytics to focus on new ways to match customers with reps and figured out the best ways to solve the problem of conflicting goals. He spoke to MIT Sloan Management Review editor-in-chief Michael S. Hopkins.
Different Questions, Different Results
Most organizations already mine their data for insights. How can they apply analytics in new ways that will discover untapped opportunities for value creation?
One of the first questions anyone would have, reading about your experience, is how did you get answers to questions you didn’t even know you should be asking? What triggered the epiphany that caused you to start looking at things differently?
The epiphany occurred because we knew we wanted more. We wanted to retain more customers, and we wanted to get more wallet share by up-selling them.
And so we put the problem to a different group. We went to the decision sciences group, to the actuaries and the mathematicians, and we asked them, “Is there anything you can see that we can do better or that we can optimize more?” They weren’t looking at it from the perspective of “How do I run a contact center?” In fact, these people don’t know anything about contact centers. So I think the first important step was to have a different set of eyes looking at the problem, and looking at it from a completely different discipline.
If they didn’t know how a contact center runs, or what things have been effective, where did they start?
The first thing that was interesting about their approach was that rather than thinking about the average speed of answering phone calls, or the average “handle time,” or service level metrics, or individual customer experiences or using QA tools to find out what we did right and what we did wrong — all the things we usually consider when looking at customer and representative interaction — they started thinking of it purely from the perspective of, “We’ve got success and we’ve got failure.”
Success and failure are very easy things to establish in our business. You either retained a customer calling in to cancel or you didn’t. If you retained them, you did it by either a cross-sell, up-sell or down-sell.
So this is what they started asking: What was true when we retained a customer? What was true when we lost a customer? What was false when we retained a customer? And what was false when we lost a customer? For example, we learned that certain CSRs generally performed better with customers in higher premium categories while others did not. These are a few of the discoveries we made, but there were more. Putting these many independent variables together into scoring models gave us the basis for our affinity-based routing.
That broadens the information they were looking for, right?
Definitely. These are data-oriented people, so they just simply said, “Give us everything — all the data you’ve got.” And we had a lot, because we’ve been running this business for years. We had information about our customers that seemed, from the perspective of call center routing, totally irrelevant. We had a lot of data in the contact center about agents’ performance, the time they spend on calls and the like. They took the whole data set and started crunching it through our statistical modeling tools.
The approach they took was to break down our customers into very discrete groups. To see what’s true about our customers. Any bank or insurance company or financial services company that sells products to customers is tempted to cluster their customers into discrete groups. Almost everyone does.
The thing is, it’s not 10 clusters that define your unique customer groups, it’s usually hundreds of clusters. That was the first process, to find out all the different kinds of customers that we have: customers with high balances who tend to pay off early, customers who have high credit-to-balance ratios, customers who have low credit scores. The more variables that go into the creation of a cluster, obviously the more clusters you can have; so, not just customers with high balances who tend to pay off early, but customers with those characteristics who also have low credit scores.
When you’ve got it down to that granular level, you can then look at all the different customer interactions that we had with people in that cluster and say, “How did we do in this particular case? How did we do in that one?”
Wait — are you looking at every single interaction?
Yes. It’s wasn’t on an aggregate macro but on an individual basis, every single interaction that we recorded over the last four or five years. Looking at all of these interactions let the team see patterns that establish that this CSR tends to do well, historically and evidentially, with customers in these specific sets of clusters.
What they also discovered was that the results were completely different from the existing paradigms in the contact center.
Sales as Matchmaking (Because “Variability” Means There’s Someone for Everyone)
What it means to understand — and act on — the critical difference between theoretically inferring why something might be likely to happen and evidentially knowing that it is likely to happen.
Let me stop you. As you’ve said, call centers tend to be pretty statistically driven places from the start. You named a bunch of the metrics that you would be looking at from the customer service side, and I’m sure you would have known when a customer called in what his products were and what his history was, and potentially matched him up with CSRs who had expertise in those particular product lines, yes?
That’s what everyone does in the call center world. When they sit down to write and build their routing strategies for how they’re going to move their macro clusters of customers around to CSR groups, they do it almost 100% onanecdote. We say that CSRs have expertise in an area. The problem is that expertise is a subjective term. When you deal with what we’ll call carbon-based intelligence — that is, inferential judgments made by us humans — versus silicon-based intelligence, or computerized judgments based on analytics, the carbon-based intelligence will say that this rep goes into this segment because they have expertise. They took a test. Or they grade out well in the QA tools.
What the evidence showed us is that the carbon-based intelligence tends to judge incorrectly. The silicon never does. If the model is set up properly and it has the ability to detect performance through whatever way you tell it to detect performance — by noting cross-sell, down-sell, up-sell, whatever — it will always measure a CSR’s performance correctly and in an unbiased way.
So for the first time you’re looking at both ends of the equation in some different ways. You’ve just described the CSR end, where you have this incredible database that reveals patterns about performance with different groups of customers, in spite of what you may or may not have inferred. What happens on the customer side? Are you looking at them in a different way?
Yes. There are obvious characteristics that we can study in our core systems. Think about what a bank or an insurance company would collect about its customers. Credit score, demographics, maybe some psychographics. We might know how many children they have.
You can predict what you think they’re going to do in the future, as long as you have a large enough customer base with enough interactions and enough variability to look at. Because what this whole thing is based on is variability. There’s a high degree of variability in your customer base, and there’s a high degree of variability in your CSR base. We learned to exploit that variability.
It’s the old adage in business: People do business with people they want to do business with. If you are successful at first establishing rapport with your customer, you have a higher probability of selling them, because there’s a trust relationship versus just taking orders.
We drive rapport and affinity in conversations by finding attributes that we can exploit to match, that create likeness across the CSR-and-customer synapse. It scales to potentially dozens of variables that operate dependently and independently of each other to drive this affinity/rapport relationship.
Having said all this, probably the most significant aspect of our use of analytics to drive conversational affinity was the persistency factor. That is, the length of time that customers remain on the books. We established almost right away that we could save a larger number of customers, as well as more profitable ones, through our new routing engine. But what we wouldn’t learn until later was the fact that we were keeping these customers longer than ever before. This was really exciting to us! As the months went by and we watched the new system operate, we observed an overall higher persistency rate for our saved customers compared to the old system. And since we’re talking about subscription-style products in our business, the longer the customers keep the product, the more revenue we generate. This turned out to be a much more important factor than a pure save or saved fee rate.
Some of this affinity matching is like a version of online dating.
That’s a beautiful metaphor, although there’s one breakdown in it. I would suppose that online dating sites work in a somewhat anecdotal way. It’s driven somewhat based in fact, but it’s also very psychographic.
We also go down to a deep level of granularity. Not body type and hair color like online sites might ask, but we do know that, for instance, certain CSRs perform well with customers that have $80 premium fees, but they don’t do so well with customers that have $10 premium fees. We don’t necessarily know the reason why. Nor do we need to.
And therein lies the difference. In our system there isn’t a lot of science behind why these differences exist, why a rep might be good with $80 versus $10. It’s just evident that that person is good with a certain customer type. So we operate off the fact that it’s true, based on the body of data that we have about the customer base and our past CSRs’ interactions with those customers. On the other hand, matchmaking sites wouldn’t have a lot of historical data about aparticular individual’s interactions with their service (unless, of course, they use it frequently), so they operate off a body of data about people’s general characteristics and what makes them interesting to each other.
So do you see the difference? We’ve become purely evidence-driven: “This CSR always does well with this particular customer type because we’ve seen it happen.”
I would describe it like this: The science does not explain why an affinity will be likely to exist, but it does show that an affinity will be likely to exist.
How Analytics Solves the Problem of Conflicting Goals
What do you do when models predicting things such as best CSR match, willingness of a customer to wait and value of a customer to the company all recommend actions that are in conflict?
It sounds like the kind of information you have about customers is not that different from the kind of information you might have had before this whole process began, and that it’s really on the CSR side that you have all this new data, plus the data about what happens in each specific interaction between a customer and a CSR. Is that what drives your models?
That’s right. There’s one other element that goes into the solution that drives revenue: the predicted economic value of a particular customer. Now, there’s not a lot of new science in that, and we have models that tell us how to calculate that. But it’s important to the solution, because in a call center we sometimes have to decide which customer to focus on. We like the idea that there’s a CSR for everyone, but that’s not always true because of call volumes and agent availability. So if your goal is long-term revenue, you can use these economic predictors to determine which customers we should be focusing on.
There was a problem we didn’t quite know how to solve right out of the gate, and that was the fact that the best matches are almost always not available. In other words, if we have 50 callers in queue and 1,000 CSRs on the floor, we can create 50,000 different solutions, and we make those calculations 10, 15 times a second. One of the 1,000 CSRs is the best match, so that’s the score to beat — the number that shows how often we make that perfect match.
The vast majority of the time, though, those matches weren’t immediately possible because that CSR was on the phone, so we had to factor in another predictive model, and that was “time to available.” That’s not a massively complex model, because the industry has been solving that kind of problem for a long time.
But when you layer “time to available” into the actual scoring engine, you get some interesting results. If an agent’s average handle time is three minutes, 30 seconds, and he or she has been on the phone three minutes, 15 seconds, then we can predict they’re about 15 seconds away to available. Then we can weigh in our prediction of customer tolerance or customer survivability — how long they’re willing to wait in the queue before just hanging up.
We know how long we keep customers in queue. We know what the outcomes are when they’ve been in queue, and we can find out where the curve starts to steepen in terms of abandon rates or bad outcome rates. We connect that information with our CSR’s predictive availability curve. If the optimal match is too far away, maybe 45 seconds or three minutes away, then the score for that optimal match becomes dampened and someone else might look more attractive to us. Because while they may not have perfect affinity, the fact that they’re going to become available sooner certainly makes them look more attractive to us.
When you became more rigorously evidence-based, what did you discover about what might have been wrong in your old assumptions?
The conventional wisdom in the contact center is 80/20 — 80% of calls answered in 20 seconds or less. That’s a promise that most businesses make, because they believe that drives satisfaction.
What we learned is that satisfaction has almost nothing to do with that. Obviously the faster you answer, the better, over a larger body of interactions. But we found most customers are willing to wait much, much longer, on the order of 39 to 49 seconds, before annoyance affects outcome.
So our observation was, if customers are willing to wait, why are we trying so hard to force them into that 80/20 or 80/25 window? The longer we’re willing to wait, the better the match is, the better the outcomes, the more revenue generated.
We’ve done tests that push all the way out to 60/60 — 60% of calls answered in 60 seconds or less. At some point there is a negative effect on abandon rates. But what we were surprised to learn is that there is no negative effect on abandon rates until you start approaching 60 seconds. Which obviously means we’ve got that time to work with in order to find the most ideal customer/CSR match. It leads to a very, very direct impact on revenue. A direct correlation between time and revenue.
To see this work so obviously is amazing, because to go from 80/20 and then jump it to 80/40, and then within a few days to see immediate results in terms of save rates and saved fee rates, it’s stunning. It makes you wonder why the rest of the world doesn’t get this.
Summarize the results you’ve seen. The problem was that you were at a 15% to 16% retention rate despite operating in a fairly optimized state-of-the-art way. What’s happened since?
We’ve seen our retention rates, our actual save rates, go as high as 30% to 33%. But that’s not the end of the story. For us, we’re more focused on saved fee rate. Save rate is if two people call in, save one, lose one, that’s 50%. But if two people call in and one is worth $80 to you and the other is worth $20, you save the $80 one, you’ve got an 80% saved fee rate, because you saved $80 out of a total $100 eligible.
This relates back to what you said earlier about having to make choices about which customer to serve during busy periods?
Yeah. We use those predicted economic availability models to help us focus on the more valuable customers. That’s not to say we discard the less valuable ones, because diversity in our customer base matches the diversity in our CSR force, so if a $20 customer calls in, we’ve got a $20 CSR to match him to. But our focus is on revenue, so saved fee rate is more important to us.
So while our save rates went into the 33-ish range, even as high as 35%, our saved fee rates went into the 47% to 49% ranges. We’ve seen days where we’ve been in the 58% range. Effectively that means that 58 cents of every dollar that was at risk has been saved. Those are very substantial numbers for us in our business.
Just so we can do the apples-to-apples comparison, what was the saved fee rate before?
The same as the overall save rate, 15% to 16%. And that’s actually a very exciting point to us, that our saved fee rates went up so much more than save rates, because we were focusing on saved fee as opposed to just saved customers alone.
If It Makes So Much Money, Why Doesn’t Everyone Do It?
What are the impediments to adopting evidence-based analytics? What can organizations do to overcome them?
Why don’t more people see the bottom-line impact of this sort of analytics?
In my own space, in the contact center world, I still am amazed when I come across very, very large Fortune 50 organizations that are still running very, very old technology. They don’t have the appetite to adopt it yet. Their current system is basically working, it’s been doing fine.
You scratch your head, saying, “Yeah, but don’t you want all the benefits that you can get from analytics?” And the answer is sort of subjectively, yes, we want those benefits, but next year.
There are early adopters and there are adopters. I wouldn’t call what we do an early adoption of a technology; it’s using very state-of-the-art tools just in a little bit of a different way. I think our creativity is in how we deployed it.
The program you developed at Assurant — called “RAMP,” for “Real-time Analytics Matching Platform” — is now available to other organizations that have to manage inbound calls. What do you see in organizations that makes it hard to apply analytics in this kind of an effective way?
The first one is, “I don’t have the IT resource to go do this right now.” You have to go compile the evidence, and that’s not a trivial task for most IT departments. It’s all data that they have, but in these days everyone’s stressed and pushed for projects and IT time.
Another objection is the perception that this is just a skills-based routing solution and that we already have skills-based routing. That’s an interesting one to overcome because, first off, this use of analytics is not skills-based routing. It’s evidence-based or success-based routing. We don’t really care about a CSR’s skills as defined by a skills-based routing system, and in fact we tell you that the skills that you assign a CSR are practically irrelevant.
Those are legitimate objections. What do you say to get someone started down the path that could enable them to get results like yours?
Well, we have proof that it works. But hearing about 187% improvement over baseline at Assurant is hard to believe at times. So we say, let us prove it to you by giving us some teaser data.
We can show you, based on your data, that you are not fully optimized and that you are relatively randomized in your routing — because effectively that’s the premise statement here. We are taking randomness and chaos and making order out of it.