Feature: Service Advantage

Banks, brokerages, telcos and insurance companies are no strangers to analyzing customer data to reduce risk and conceive marketing campaigns, crunching the numbers to detect fraud, evaluate creditworthiness, cross-sell and more. But most organizations are just beginning to direct their considerable data mining and analytic prowess to serve customers better.

"Instead of using analytics to predict what customers are going to buy next, we're using them to improve the customer experience, which indirectly boosts revenue, too," says Forrester Research analyst John Ragsdale.

To retain customers in markets prone to churn, such as financial services, firms are using predictive analytics to spot behaviors that foreshadow customer departures and to develop the right inducements to make them stay. To improve customer satisfaction, service organizations are analyzing customer interactions to spot and correct problems. To stay in tune with changing consumer needs, companies are scoring customers by the minute, applying analytics to a steady stream of data from call centers, retail stores, Web sites, e-mails and text messages to customize products and services on the fly.

Making all these efforts possible are emerging predictive capabilities that can examine customer interactions across channels, new analytic tools for inbound call centers and visual analysis tools that give service reps insight into customers at a glance. Read on to learn how the latest technologies are being applied to customer service challenges at mutual fund firm Dreyfus, at U.K.-based mobile phone provider O2, at a leading regional AAA organization and at online insurance provider eSurance.

Predicting Defections

One point at which predictive analytics and customer service logically intersect is customer retention. By applying predictive models to customer behavior, companies can identify patterns that lead to defections and then figure out how to solidify shaky relationships. SAS and SPSS are the leading providers of predictive analytics software for customer data analysis; other specialists include Chordiant Software and Intelligent Results. These products can analyze multiple data sources using several standard data-mining algorithms, often in combination, then provide reports to different groups of decision makers.

When Dreyfus realized its biggest business challenge was redemption--customers cashing out their investments--the mutual-fund company began building predictive models in SAS Customer Relationship Management Solutions, so it could forecast which investors would leave over the next five to nine months. Dreyfus used a "survival analysis" technique that originated in the biomedical industry and was used determine how long patients would be likely to survive untreated or under various treatment regimes.

In financial services, warning signs can include a change in almost any behavior: more calls to the customer-service center, more transactions, fewer transactions, higher or lower balances. At one point, for instance, customers in Dreyfus' Golden Years segment were found to be taking more money out of their accounts than they were putting in; further investigation confirmed these customers were transferring their wealth to younger generations. Dreyfus responded by offering these "majority redeemer" customers trust services and education savings plans. "Once we do that, we extend the customer's life with Dreyfus," says Prasanna Dhore, executive vice president. The overall effort paid off handsomely, bringing the redemption rate down from 25 percent to 7 percent.

Dreyfus's challenge is to keep that defection rate below 10 percent while the industry average hovers around 22 percent. To do so, statisticians run the predictive models regularly to troll for customers with a high probability of redeeming their funds, then test hypotheses as to why customers might leave and what alternative products might keep them in the fold. "We have to keep doing that because people do change," Dhore says. "New customers come in and the segment behavior changes, so models and segmentations become stale and have to be rebuilt."

Dreyfus focuses primarily on what Dhore calls "money in motion." The biggest challenge over the next five to ten years will be predicting how to retain retiring Baby Boomers. "They're looking for ways to invest their money so they don't [outlive] their investment," he says. His team of eight analysts is creating predictive models to determine which customers are on the cusp of retiring, how long each customer is likely to keep working, how long customers will keep their multiple 401(k) plans and how to break the large segment of pre-retirement customers into five to seven smaller clusters that can be served in a more customized way. Surveys and focus groups are being conducted to better understand what these customers want, and service reps are being trained on retirement issues. The company also plans to feed some of the results of its predictive models to partners such as Merrill Lynch, Wachovia and brokerages that sell its funds.

Scoring Customers In Real Time

Everyone wants to feel understood and, in the warmth of a personal chat, even an attempt to sell a product will be more welcome than a series of frustrating requests for the same basic information. What if contact-center reps could know what questions callers submitted during their last visit to the company Web site? And what if reps could automatically run propensity and behavioral models on accumulated data so they could guide the conversation to customer wants and needs? That's just what O2, a forward-thinking British mobile phone company with 16 million customers, is in the process of achieving.

"O2 is a great example of moving beyond the cross-sell concept and saying, 'How do we improve the customer relationship from the customer's point of view, as well as from ours?'" says Gareth Herschel, a Gartner analyst.

O2 is combining data from all channels--call data from three contact centers, billing data, e-mails, chat messages, and detail on interactions at retail stores--into an Oracle data warehouse. Rather than doing a broad customer segmentation that might miss the mark, O2 does predictive modeling in Chordiant software to create offers that can be customized for each customer. Customized recommendation software is also available from ATG/Primus and SSA.

"We have more than 35 products, and we can predict the best way for you to pay, how much text messaging you'll need each month, if you'll use multimedia messaging, if you'll want Internet access in your handset, and the types of music downloads, ringtones and screen savers you're likely to want," says Aly Richards, head of CRM strategy and architecture. It's not always teenagers who want all the multimedia add-ons; men between the ages of 30 and 40 tend to like digital toys, too, she notes.

O2 also uses Chordiant's software to choose contact strategies for new customers. "Within three days of you getting your phone, we can tell you what value band you're likely to be in and, from that and other information, we decide on which of 32 treatments you may want," Richards says. High-value customers might get a phone call from a rep when they start up, welcoming them and offering additional services. A lower-value customer might get an e-mail, text message or snail-mail welcome.

More impressively, O2 continually runs propensity models across all channels to revise customer scores and determine the best course of service at any moment. A customer could go to a retail store or the Web site, buy a new phone and then call the contact center; the rep who responds gets fresh recommendations from the propensity models. Constant rescoring helps O2 maintain relevance, and Chordiant's Strategy Director decision engine applies business rules to the latest models.

"You need that [decision engine] because you can't just look at one model, such as the churn model, to make your decision," Richards explains. "You also need to look at your value model and usage models to say, 'This customer is likely to churn, but what's the best thing to do to keep them and how much money can we spend?'"

Customer-service agents are given access to much of the analytic information and are guided through decisions. They're also given budget guidelines and a list of "save tools"--incentives designed to woo potential customers or would-be defectors--with the most customer-relevant offers highlighted. Agents have some latitude over just how they'll win over customers, but their pay is tied to not only saving the customer relationship but protecting margins. This is a recent change for the agents, who were previously paid extra just for saving accounts.

Since O2's Vision customer-analytics program began, the company has become the No. 1 mobile phone provider in third-party customer satisfaction surveys. The program has also far exceeded the original business case; among inbound calls during which a service rep makes an offer, 75 percent to 80 percent result in a sale. O2 says it now now plans to integrate the decision engine with its campaign management tool, so outbound campaigns and inbound calls can work together--another marketing/customer service Nirvana, where customers get a seamless experience and all messages to, and contacts with, the customer are linked and understood.

Analyzing Speech To Find Clues

As voice recognition technology improves, companies are applying speech analysis to contact center data to help determine what customers want and what turns them off. DMG Consulting forecasts a 120-percent growth-rate for speech analytics applications in the contact center this year, and 100-percent growth in 2007. Vendors in the category include CallMiner, NICE Systems, Verint Systems and Witness Systems.

Improving customer service has been a top priority for AAA Washington/Inland, an independent member of the AAA organization that's currently ranked 12th in emergency road service among 87 clubs nationwide. The Bellevue, Wash.-based club provides travel, insurance, financial and automotive services to 905,000 members, and fields about 7,000 calls per day in three call centers. Last year, the organization deployed call-center analytics software to drill down on transcript data, to spot the most frequently asked questions and most serious customer complaints.

In one application recently developed, the club knew it had a problem with emergency callers in some areas being told that AAA didn't have a local contractor available to send to their rescue, leaving the caller to find aid on his or her own and submit a receipt for reimbursement--a level of service that was unacceptable. The challenge was quantifying the problem.

"I wanted to show our automotive team how often we are referring our members to a pay-and-apply situation, and why," says Janet Ryan, director of call center operations.

AAA Washington/Inland uses Verint's Ultra software to transcribe customer calls and capture customer interactions over e-mail, chat and agent-assisted co-browsing. An associated analytics suite then mines the dialogues to extract actionable intelligence. Ryan ran a query on the words "pay and apply," and turned up many of the cases in which customers had to arrange for their own towing. She then drilled down to spot patterns, in terms of geographic area, time of day and so forth, and the club investigated why contractors weren't responding and how it could improve coverage.

In another case, Ryan suspected customers were being put on hold too much, so she ran a query on the words "thank you for continuing to hold." She quickly noticed that those words appeared in 100 percent of calls to a particular third-party travel provider. Further analysis showed that the average call to that vendor was taking about 16 minutes, whereas customer surveys showed that clients expect to make their travel arrangements within five to six minutes. The information was brought to the attention of both AAA executives and the third-party vendor.

All these analyses took place since users were trained on the software in January. Ryan says results have been quick, and she points to internal surveys that show that 97 percent of customers are satisfied with the service they get from the club. "The software lets us go to deeper, more complex things that are hard for us to see or substantiate so we can create plans to correct those problems," she says.

Keeping Web Visitors Happy

As e-commerce becomes more mainstream, organizations have to pay more attention to how their Web sites are behaving. Yet a recent Harris poll found that 89 percent of consumers experience problems when completing online transactions, and 34 percent switch to a competitor, online or offline when they experience these problems (see Listening Post, right).

"Companies are shifting from viewing the online channel as a way to deflect customer calls and save money, to seeing it as a valid means of working with customers," says Forrester's Ragsdale. "To do that, they have to provide a great customer experience in every channel."

The ideal Web site is tailored to you so you never see content that doesn't apply to you, your interests or the products you purchased, Ragsdale says. "You don't get offers that have nothing to do with you, and you don't search a knowledge base for a solution and get meaningless returns. Even the language may be appropriate for your age group."

Web analytics tools provide insight into customers' online activity so companies can move toward a more customized experience. ATG, InQuira, Kana Software, Knova Software, Sento and TeaLeaf Technology are among the e-service analytics vendors.

ESurance, an online provider of auto insurance, strives to deliver fast, easy self-service. When a problem is reported on the Web site, they use TeaLeaf's software to find records of the faulty session by searching for an e-mail address, time stamp or another identifier. Analysts can then replay that session to determine what led to the problem, and they can analyze all Web session data to determine how many other customers may have been affected.

"We also use TeaLeaf to benchmark key metrics on our site," says Marj Davies, director of Internet operations. "For example, we can measure page load times before we make changes to our site and capture the same measures after the change to make sure that performance has not degraded." Faster load times speed up the quoting process, please customers and boost sales. The metrics also help Davies spot and eliminate problems in the quality assurance phase--before new pages are released to the live production environment--minimizing self-service problems.

Acing Customer Service

So the tools are out there to scientifically decide how to serve customers better. But don't forget lessons learned in more established analytic pursuits. "Go in with a well-defined business problem," advises Dhore of Dreyfus. "Don't go in with the idea that you're going to do some data mining and figure out what needs to be done. Ultimately, that's a waste of time and money."

And, despite the move toward making business intelligence more accessible to business users, leave the deep analysis and interpretation to professionals who know the difference between, say, correlation and causality. "I recommend that a statistician or other knowledgeable person use these tools," Dhore says.

In other words, when relationships with your best customers are at stake, don't send your customer service amateurs in to play the analytic equivalent of Wimbledon. Let the pros do what they've trained so hard for.

CRM Vendors Need to Learn Respect

You'd think CRM technology providers would be masters of online customer service, but you'd be wrong, according to The Customer Respect Group. The consulting firm evaluated site usability among a wide range of users; communication, in terms of response to specific customer questions; and trust, as in responsible policies on personal data. The CRM community earned an average rating of 5.6 out of 10--versus 5.7 for all industries. The group was particularly poor at communicating with customers, with a score of just 4.1 out of 10: Twenty-seven percent of e-mail inquiries were ignored, none of the vendors consistently responded with helpful responses in a timely manner (within one day), and only 31 percent of e-mail inquiries were acknowledged with an autoresponse. Just half of the inquiries that did receive a response were answered within a day.

Visualize Better Service

Although data-mining experts can spend their day diving into data and finding gems of insight, the downside of delivering analytics to the call center is that it can present too much information to service reps. "They'll spend all their time trying to understand and digest the information, instead of having the dialogue with the customer," says Gartner analyst Gareth Herschel. "The key is to distill what you give to the agent into the smallest amount of information possible that will enable him to continue the appropriate action."

Customer-service reps need to get to the salient facts without putting customers on hold. A picture is worth a thousand words, so tools are becoming more visual--with a new breed of software graphically illustrating key performance metrics. Large red bubbles, for instance, can be used to represent customers with the highest risk of defection, or highest value to the company, showing reps at a glance that these are people that need to be treated with velvet gloves.

ELoyalty, FYI, Panopticon and Visual Analytics all offer visual analytic tools for contact centers. By turning analytics results into visual displays, these tools give service reps and others a way to instantly grasp the most pertinent facts and deliver appropriate customer service.

Executive Summary

Business intelligence has long been used in fraud detection, cross-selling, price analysis and other applications, but rarely have analytics been applied to simply serving customers better. That's changing, as customer service data gets more plentiful and as new technologies and techniques are applied to the task.

Predictive analytics are being used to determine which customers are about to leave, and to come up with offers that extend the relationship. Real-time scoring and decision engines are matching service and upgrade options to customer activity--even if they bought a new product at a retail store just 10 minutes ago. Voice-recognition and speech-analysis technologies are being used to spot and correct customer-service problems. Data-visualization tools are giving service reps at-a-glance views of customer behavior and value. And Web analytics tools are mining clickstreams for clues as to how to improve the e-commerce and online service experience.

The technologies are there; all it takes is vision, creativity and management buy-in to make them work. This article describes the practices and lessons learned at Dryfus, mobile phone supplier O2, a major regional AAA organization and online insurance provider eSurance.