Using Big Data to Improve Call
Routing in Customer Service
IndustryView | 2015
Individuals who call customer service are often put on hold and told their call “will be answered in the order in which it was received.” However, more and more call centers are using advanced “big data” techniques to determine which specific agent to route customers to based on that person’s unique customer service preferences.
In this report, we examine these advanced call routing techniques to help businesses find opportunities to improve their own call center routing to improve customer service outcomes.
Many of the software platforms companies use to provide customer service—and nearly all of the call center-specific platforms—offer some type of automated call distribution (ACD). ACD can follow any one of a number of basic call routing rules. The most basic rule is one most consumers are familiar with: simple linear call distribution, where calls are answered in the order they’re received and by the next available agent.
Alternatives to this first-come, first-served approach include skills-based routing and analytics-assisted routing.
Skills-based routing builds upon basic ACD. Agents are divided into groups or departments, depending on their training and specific customer service role. After a caller has identified the reason for his or her call, the ACD routes it to a departmental call queue, where it will be answered by the next available agent who has the necessary skills and training to do so. Skills-based routing is commonplace, and used in most call centers around the world.
Analytics-based call routing is a refinement of—not a replacement for—skills-based routing. It aims to improve call outcomes by intelligently matching callers with specific, individual agents based on the application of analytic models to a selection of customer information datasets.
For example, say a customer calls with a question about their bill. Skills-based routing would route the caller to the billing department’s call queue, to be helped by the next available agent. But analytics-assisted routing would use the caller’s phone number to determine that he is a 55-year-old, high-income male living in Kansas, and that he frequently complains about his bill. It would route his call to a specific agent’s call queue after determining that the agent has the best track record of explaining billing issues to 55-year-old, high-income males in the Midwest.
Analytics-based call routing is based on the premise that records of a customer’s call history or demographic profile can provide insight into which call center agent(s) has the right personality, conversational style or combination of other soft skills to best meet their needs.
In this report, we examine the plausibility of this premise by asking the question: What can be assumed about a person’s customer service preferences just by looking at basic information, such as simple demographics data? To answer this question, we surveyed U.S. consumers on their preferences for customer service calls, and looked for demographic patterns in the results.
The demographic variables our survey focuses on are:
Our data reveals several correlations between customer service call preferences and the caller’s age, income, region and urbanicity. The strongest overall correlation—that is, the best indicator of what preferences a caller might have—is with age.
Our survey questions largely focus on aspects of customer service calls that individual agents, or the service department as a whole, have some control over. For example, one question asks if the caller, with all else being equal, would prefer that a customer service agent work through their call quickly, or take as much time as needed to really listen and understand the problem. The older the caller, the more likely they are to prefer the second choice.
Indeed, most would agree that a person’s age provides clues as to what that person might like or dislike or how they prefer to interact with others. Most companies and their customer service departments are aware of this, especially when their target customer population falls within certain age brackets.
Regarding age, we also asked respondents if they prefer to speak with an agent who is close to their own age. Overall, most (76 percent, not shown in these charts) do not have a preference. However, those who do prefer speaking with an agent in their own age group are twice as likely to live in urban areas.
When given the choice between an agent who sounds “formal and professional” and one who is “more casual and friendly,” we found that older callers typically prefer the former, while younger callers prefer the latter. (As with all the survey questions, we explained that both choices would yield the same result.)
For the same question, there is also a correlation between a caller’s preference and where they live in the country. Callers in the Northeast have the strongest preference for formality; Midwestern callers prefer “casual and friendly” calls. Callers from the West are least likely to express a preference for either.
It’s worth mentioning that, overall, nearly all groups prefer speaking with a relatively “casual” agent. The only group preferring more formality is the 65-and-older bracket.
Managing an inbound call queue can involve more than just directing calls to the right department or the right agent: it can also involve deciding which call center, in which country, to direct a call to. It’s not uncommon for even small call centers to maintain a staff of overseas agents. And while it’s generally understood that U.S. callers prefer speaking with U.S. agents, the cost-benefit analysis of using overseas agents is compelling enough that many companies include them, in part or in whole, in their agent lineup.
Our survey found that two demographic variables—age and income—correlate with a caller’s preference for a U.S.-based agent.
While these specific correlations are interesting and could help companies refine or realign their call queue management, the overall lesson here is equally important: Even the broadest demographic variables do in fact correlate with consumers’ customer service preferences. The premise of analytics-assisted call routing—that incorporating some of this information into call routing decisions can improve call outcomes— stands on firm ground.
Our findings show that even simple demographic variables can predict a person’s preferences for customer service calls. Beyond basic demographic data, there are other variables that can also reasonably be expected to hold predictive value. Consumer data for many of these variables is publicly or commercially available, including:
One challenge to successfully implementing advanced call routing is determining what specific mix of variables will be most successful at predicting call outcomes.
“The reality is that you often manage to get millions of pieces of data,” says Chris Farmer, chief marketing officer for SATMAP Inc., a leading provider of advanced call routing technology. ”But the key is in prioritizing which pieces of data you want to incorporate into the [call routing] model.”
This is further complicated by the fact that call centers operating in different industries and that provide different business functions (e.g., sales, technical support, customer retention) each have different goals.
Farmer explains that software such as SATMAP addresses this challenge by using a dynamic modeling algorithm, which is updated daily with all information from that day’s calls. The new information continually refines the model, leading to better pairing and better call outcomes over time.
Changes and improvements to call pairings lead to measurable bottom-line results. One example from SATMAP reflects how advanced call routing helped a major U.S. telecom company improve its call center operations: Average call-handling time fell by 6.4 percent, and the call center’s staffing costs dropped 6 percent.
While there are many sources of data that can be used to inform call routing decisions, not all will have equal predictive value. The best predictors are those that incorporate points from multiple data sets. One way to do this is to create a caller personality profile: a simple categorization of how a person can be expected to behave during a call.
Personality profiles provide a more three-dimensional picture of a caller than any single data point. Likewise, they’re more broadly applicable. Scott Cotter, vice president of marketing for Mattersight (another leading provider of advanced call routing), explains the value of personality profiles:
Whether it’s a service center and you’re trying to reduce call times or reduce callbacks, or you’re focused on customer experience and you want to boost satisfaction or customer advocacy scores, or you’re a sales and retention call center and want to win more orders or retain customers … regardless of what the function is, personality is highly correlative.Scott Cotter, Mattersight
Mattersight holds a number of patents for the behavioral analytics technology it uses to create customer profiles. The profiles are linguistics-based: they’re derived from an analysis of the word and grammatical choices a person uses when speaking on the phone. Cotter gives an example of two different customers calling about a billing question:
“Some people are very fact-based, and they’ll tell you how they came to their conclusions: ‘I looked at line six and it doesn’t seem to be the right calculation. It’s different than last month’s bill. Please explain why it’s different,’” he explains. “Other people are much more emotional: ‘I am just so frustrated with this bill. It’s always wrong. How can your company even explain this?’”
A Mattersight call-analysis diagram, noting the call’s highlights, successes and failures. Source: Mattersight
Mattersight’s analytics engine creates a caller’s personality profile by analyzing the language they’ve used in previous calls. Interestingly, these previous calls may have been to other companies: Data collected from calls to other Mattersight clients is included in the company’s unified database, which identifies consumers by their phone number.
“We start to see patterns over time of which agents do well with which different personality types,” Cotter says.
Mattersight’s analytics engine helps make better call routing and call-pairing decisions. Source: Mattersight
Of course, in many cases, a company will already know a great deal about each customer who calls in—especially when they’re existing customers. In these cases, the challenge is to match the caller with their records, and do so early enough in the call to be able to make a routing decision based on this data.
Bhavesh Vaghela is chief marketing officer for ResponseTap, a call-tracking and advanced call routing solutions provider. Vaghela describes a very basic scenario in which a caller’s phone number is recognized by the company’s phone system. The phone number is matched to the customer’s records—including, for example, information about what products they’ve purchased—and is used to determine to which department or agent to route their call, without the customer knowing or being asked to make a selection.
This, however, is a fairly elementary implementation of advanced call routing.
“The process becomes more interesting if you are identified before you call,” Vaghela continues. “In this case, your online customer journey will determine [how your call is routed]. If you’re on a bank’s mortgage page when you call, chances are, you are looking for a mortgage. Why not route directly to that team? It can get much more sophisticated when looking at the entire online journey, including multiple visits to a company website, and gauging things like interest and intent to purchase.”
Vaghela gave other examples, including how many times a customer has visited the company website (or a specific page on the website) since they last called.
Since companies have different customer service objectives and different call center goals, they look to analytics-assisted call routing to improve different metrics. While many support-based call centers seek to reduce average handle time (AHT), sales-based call centers often seek to raise close rates or increase average order size.
Improvements in AHT across various industries after implementing Mattersight’s behavioral routing (Note: ATT = average talk time)
It’s not always easy to measure results from changes to call center operations. They are complex environments with many variables at play, and thus it’s not always clear which changes are responsible for which results, or if those results are actually related or just statistical noise. Changes to call routing plans, however, are generally much easier to evaluate.
For example, the SATMAP call routing system proves results to users with simple A/B testing. Once the system is implemented, it can be turned on and off automatically at 15-minute intervals. This on/off cycling, explained on SATMAP’s website:
“Allows contact center managers to measure the impact of SATMAP while it is running compared to when it is not.”
Whichever data sources a company uses to improve its call routing and service it uses to implement them, the insights provided by the data remain valuable—even after the call ends. Ideally, this information will be retained in the company’s customer relationship management (CRM) or help desk software, to be used again (and added to) in the future. Using software that is highly customizable and that integrates with external data sources (such as HappyFox and Zendesk) is one way to ensure these data sources are available to improve post-call service.
HappyFox, for example, includes two functions that allow external data to be incorporated into its customer support system. The first, CEO Shalin Jain explains, is a function that allows customers to be placed into groups. For example, those in Group One might be high-value customers, those in Group Two might be new customers, while those in Group Three might be calling from overseas. These groupings are then used as the basis for call routing and ticket management decisions.
The ability to add custom fields and tags to individual customers and their support tickets can also help integrate external data sources into customer service operations.
“A lot of companies promise that you’ll have a dedicated account manager, or priority support, but often they still manually look through the system for tickets coming from that particular customer. This can lead to problems if the call or ticket gets picked up by a lower Tier One agent,” Jain explains.
Manually looking through tickets is also an inefficient use of time. By adding a custom field for “priority support” and then tagging all of the priority support customers, companies can increase efficiency and ensure that support requests are always automatically routed to the correct department and handled by an agent with appropriate skills.
Customer service centers have traditionally used ACDs to route calls on a first-come, first-served basis. Now, many are switching to more advanced data-driven methods to optimize their call routing for better customer service outcomes.
Our survey shows that within large populations, demographic variables do correlate with customer service preferences. Even broad demographic variables, such as age and income, provide indications of what style of customer support a customer will likely prefer.
There is a very wide variety of data sources available—basic demographics and other publicly available data, records of previous calls and a company’s own CRM files with histories of customers’ online and offline interactions, to name a few. The challenge businesses face is determining which data to use, and which solution will integrate it best with their call center and customer service operations.
To find the data in this report, we conducted a three-day online survey of 10 questions, and gathered 394 responses from randomly selected consumers within the United States. We worded the questions to ensure that each respondent fully understood their meaning and the topic at hand. Additionally, we spoke with 20 businesses that use or provide specialized call routing to learn about their experiences. Software Advice performed and funded this research independently.
Results are representative of our survey sample, not necessarily the population as a whole. Sources attributed and products referenced in this article may or may not represent client vendors of Software Advice, but vendor status is never used as a basis for selection. Expert commentary solely represents the views of the individual. Chart values are rounded to the nearest whole number.
If you have comments or would like to obtain access to any of the charts above, please contact email@example.com.