In the beginning, there was mass marketing. Companies promoted new products and services to all their customers, regardless of whether those products or services were appropriate for the individuals on the company's mailing list. When a bank launched a mass marketing campaign, a 1 percent to 3 percent response rate was considered successful. Some success.
If the average direct-mail campaign costs a company about US$98,000 to reach 100,000 customers, according to estimates from the New York City-based Direct Marketing Association, that organisation is netting a huge loss.
Today there's customer segmentation. Instead of devising one campaign for thousands of disparate customers, a company breaks those thousands into smaller demographic segments and tailors its campaigns to each of those segments. Technologies such as data modelling, warehousing and mining, and a whole lot of algorithms enable businesses to look at consumer buying patterns and channel preferences and to slice their segments ever finer.
Sounds great, right? The more you know about your customers, the better you can serve and sell them. Piece of cake.
Too bad it's not. Customer segmentation is hard work. It involves:
-- Consolidation of customer information from a variety of systems that may be integrated or siloed (such as in ERP or legacy systems) -- Brilliant statisticians to define algorithms that will analyse mountains of data in order to sort customers into various segments (for example, good business analysts) -- Collaboration between marketing and IT (because information isn't worth a damn unless you know how to use it) -- A robust infrastructure for aggregating, storing, processing and disseminating information about customers to the appropriate businesspeople (hello, knowledge management) Analysts and practitioners in the marketing departments of the Fortune 1000 view customer segmentation as a great way to reach consumers, but few companies do it effectively or can boast much success.
Barbara Bund, a senior lecturer at MIT's Sloan School of Management, believes that segmentation hasn't served companies well because they haven't thought through the types of segments that would best serve their mission. "You have to understand why you're segmenting your customers a certain way and why that way makes sense for your business," she says.
For example, the MIT teacher and marketing guru says consumer electronics companies tend to segment their customers by age and income. But she doesn't believe that data has much bearing on the way people buy items like CD players and televisions. Bund says they should divide their customers between those who are comfortable with high-tech gadgets and are thinking about the next generation of handheld computers and those who are watching the VCRs in their living rooms blink 12:00...12:00...12:00.
Andy Main, a partner in charge of e-marketing with Deloitte Consulting in San Francisco, believes the difficulties associated with customer segmentation are a matter of business process. "Technology isn't the obstacle," he says. "The obstacle is getting companies to realise that the data really is valuable. A lot of companies log the fact that there's a contact with the customer, but they don't log the fact that there are insights about consumers' behaviour from these interactions. They've got to do a better job of learning from each interaction. And that may mean asking the customer questions about their satisfaction with the product when they call in for help."
Another difficulty, according to Main, is releasing "data from data jail. These warehouses are like big prisons for data." In other words, the challenge for CIOs is connecting the data with the folks in marketing who need to use it. On top of that, they have to make sure the data is consistent and accurate, because it often comes from a variety of sources and systems. Often, says Main, the systems in the sales department are different from those in the customer service department. A customer may be referred to by their full name in the sales system, and their first initial and last name in the service system. The marketing department has to know that Meridith Jill Levinson and M. Levinson are the same person.
Customer segmentation is so difficult to do and do well that companies are extremely hush-hush about their customer segmentation practices and the IT infrastructures required to support them. Happily, three CIO-100 honorees with robust customer segmentation schemes--the Royal Bank of Canada, The Prudential Insurance Co. of America and Charles Schwab & Co.--offered to explain how they do it and describe the role IT plays in enabling them to understand their customers' needs and interests.
At least once a month, business analysts at Toronto-based Royal Bank of Canada use data modelling to segment the bank's 10 million customers according to their credit risk; their current and future profitability; their life stage (whether they're young people in a period of intensive borrowing or seniors approaching retirement, for example); attrition (how likely they are to leave the bank); channel preference (whether they like to use the call center, the branch or the Internet); product activation (getting a customer to use a product they've bought, such as a line of credit); and propensity to purchase another product.
"Gone are the days where we had mass buckets of customers that would receive the same treatment or same offer on a monthly basis," says Shauneen Bruder, Royal Bank's senior vice president for North American markets. "Our marketing strategy is much more personalised. Of course, it's the technology that allows us to do that."
Royal Bank's customer segmentation practice is so good that the company's response rate to marketing campaigns and sales programs has ranged as high as 30 percent, compared to the banking industry's average of 3 percent, according to Bruder.
The main source of customer data the analysts study is stored in the marketing department's marketing information file (MIF). This data mart contains information on the products customers hold with the bank, the channels they use, past campaigns, as well as restrictions on soliciting customers, and transactional data. It is fed by the enterprise data warehouse, which stores information from every document a new or existing customer fills out. The MIF is also fed by the bank's legacy (checking accounts, credit cards) and operational (ERP, billing) systems.
The business analysts extract customer information from the MIF and the data warehouse to do their analyses. Data modellers download this information directly into SAS Institute files. Royal Bank's systems and technology department has created tools--essentially user-friendly graphical interfaces--that the analysts use to query the marketing database and obtain lists, for example, of customers according to the products they hold.
To query the database, analysts run data models directly against the data in the MIF (as opposed to having to run batch programs overnight). These data models are complex algorithms supported by software programs that the analysts run off their desktops. The models help the marketing department identify the products a customer might need or be likely to buy, and the likelihood that he or she will leave the bank. They are also based on data about customers' attitudes, behaviours and demographics--information that the bank obtains from companies that provide demographic information and from studying its own transactions.
Here's how it works: During a study of its customers, the analysts find that Jim Patrincus's (a hypothetical customer) balance is low, his credit card payments are slow and his deposits have recently become sporadic. All of this indicates that he may be leaving the bank. Using different models, they also find that he has been a profitable customer and that he has the potential to be even more valuable based on his banking patterns in the past and the products he currently holds. The folks in the marketing department know they'd be fools to let Patrincus slip through their fingers and move to the bank's competitor, the Canadian Imperial Bank of Commerce.
His profitability and risk profile aren't the only things the analysts know about Patrincus. They know from his product portfolio that he has a car loan, a line of credit and a checking account with Royal Bank, and they can therefore infer that he is in a phase of his life where he has high borrowing needs. By assessing the channels he currently uses and employing a model the bank has developed to predict future channel preference, they know that Patrincus likes the Internet.
The analysts are also able to determine that he's highly likely to buy a package of banking services (for example, for $9.95 a month he gets Internet banking, bill payment, unlimited access to ATMs and call centres, and a limited number of branch transactions for free each month). The bank has found that customers who hold service packages stay with the bank an average of three years longer than customers who don't and that customers with packages tend to be more profitable. Armed with all this information, the marketing department comes up with a personalised strategy for Patrincus, which includes trying to sell him a package of services.
"[Patrincus] would see that we understand his needs and that we're putting a service package offer out there for him," says Christine Sibley, Royal Bank's senior manager of marketing strategy and performance management for North American markets.
The marketing department enters the actions it wants to take with particular customers in a central database that is fed to the desktops of all its personal bankers and customer service and call center representatives. The entire customer relationship management (CRM) environment, consisting of the enterprise data warehousing system and the marketing information file, are linked via application programming interfaces (APIs) that provide real-time access and updates to customer information, according to Marty Lippert, Royal Bank's vice chairman and CIO. "The Canadian banking system can't take advantage of any float. It's all same-day settlement," he adds, pointing out that Canadian law forbids banks from doing overnight batch updating. The real-time APIs link Patrincus's activity on the site with the call center systems, so if he were to phone the Royal Bank call center to check on a transaction, the reps would know all about it.
Of course, Royal Bank is able to capture information its agents get from clients as they talk to them at branches or over the phone and get the outcomes of those interactions back into the information file.
The reps make notes about the conversation on their desktops. If Patrincus refused an offer the marketing department deemed appropriate based on its modelling, the reps would indicate that. The marketing department uses this information the next time it does its modelling to develop future campaigns. Because all of the systems are updated in real-time, if a branch agent pulled up Patrincus's profile immediately after his conversation with the call center rep during which he declined the offer, the branch agent would know not to offer the same package again. The agent might want to ask Patrincus what was wrong with the package and what the bank could do to make it more attractive to him.
In the fall, reps will be able to log why a customer said no to an offer by selecting reasons from a pull-down menu. By mid-2001, the Royal Bank will be implementing technology from Redwood City, Calif.-based BroadVision to link its legacy systems and data warehouse to the Internet. The BroadVision software will enable the bank to track Patrincus's activity on the site after he logs in. So if he's checking the status of his account, Royal Bank will be able to put an offer out to him about a service package right then and there; it will pop up while he's visiting the Royal Bank site.
"The whole idea here is the consistent client experience, whether the client is talking to us via a call center agent or face to face at a branch," says Sibley.
But what if Patrincus wasn't so profitable? What if he were a drag on resources and the bank knew that he was always whining, complaining and tying up tellers? Would Royal Bank say, "Sayonara, Mr. Patrincus"?
"For some clients with whom you have low current value and low potential, your strategies are more around reducing your cost and [incentivizing] them to use lower-cost channels or, in some cases, reducing their credit risk," says Bruder. "The fact is, when 50 percent of your customers, as is the case in all retail banks, are unprofitable, you can do as much to enhance the bottom line by improving or reducing the loss on those relationships as you can in enhancing the profitability of currently profitable customers."
Violet Lang (also a hypothetical customer) is 36 years old. She lives in a small house outside of Hoboken, N.J., with her husband and two sons and earns US $60,000 a year. Her initial purchase with Prudential was for life insurance, which she bought through an agent. Later she bought property and casualty coverage through that same agent, as well as a mutual fund. Six months later, her agent quit his job.
"One issue that plagues insurance companies is volatility in the agent workforce. They come and go," says Peter Lacovara, Prudential's vice president of IS and chief IT architect. "We do a lot to try to ameliorate the loss of an agent on the relationship between the company and the customer."
The information about Lang and the products she owns is stored in the Newark, N.J.-based company's data warehouse. The warehouse takes real-time feeds from 65 systems or operational databases and crunches them on a monthly basis using mainframe DB2 technology for processing and IBM SP2 massively parallel processing technology for storage, according to Luane Kohnke, Prudential's vice president of marketing performance and information. The marketing department uses the data warehouse as a decision support system to help marketing staffers determine how to fashion their campaigns in response to customer needs and to "understand customers' potential behaviour," she says.
The majority of the information feeding the warehouse comes from Prudential's administrative and customer servicing systems, and its customer information directory, an index of all the products Prudential customers own. The company supplements this information with demographic data and information on customers' assets, which it purchases from Acxiom Corp. in Little Rock, Ark., and Pittsburgh-based Innovative Systems.
Analysts in Kohnke's department develop models using SAS Institute's Enterprise Miner to rummage through the warehouse in hopes of finding which customers might leave Prudential or which ones might be likely to buy additional products. Then they sort and rank the customers from highly likely to leave or buy to unlikely to leave or buy.
In the case of Lang, the analysts notice in studying the transaction files from customer service representatives and the payment files that come in with the remittance process that Lang has inquired about the cash value of her policy and that her insurance payments have been lagging, both of which indicate that she might leave Prudential. When they determine who her agent was, they locate all of the departed agent's former clients in order to put programs together to try to keep them happy.
At the same time, the analysts find that based on her behaviour, Lang is likely to be looking into buying another mutual fund. They score her as likely to leave the bank and likely to buy another product.
The systems operations group that supports marketing has access to the coding that Kohnke's group developed to score customers. They pull from the customer warehouse the list of customers most likely to leave and to purchase a mutual fund, and they give it to the marketing program managers who develop programs for Prudential's sundry segments.
Then the systems operations group prepares the files and sends them on a tape drive to the fulfilment house, which produces the materials for a direct mailing. Of course, one of the brochures with information on mutual funds finds its way to Lang's mailbox. She calls the 800 number provided on the brochure. When she gives her name and phone number to the call center agent, he sees that it matches the information in the customer warehouse and knows exactly which campaigns Prudential has mailed to her. He logs Lang's call into Prudential's call center tracking system and notes the nature of their conversation and whether she wants to be contacted by a salesperson or mailed a prospectus. The log automatically triggers a lead, which is sent to the sales department via Prudential's Lotus Notes database.
If the salesperson opens a new mutual fund for Lang, the application gets keyed into the business system for mutual funds, which is one of the 65 feeds that create the customer warehouse each month, so the information about Lang opening a new mutual fund eventually reaches the customer warehouse. The salesperson also files the outcome of his call in a database that the systems operations group extracts to find out what happened with the lead.
"It's a closed loop," says Kohnke, referring to the new world of CRM. "It's much more dynamic than in the past. It's building your systems and processes around the customer. It's a much different relationship between marketing and other parts of the organisation, and it's really enabled by technology."
Of course, closing this loop is no easy feat. Nitty-gritty work has to be done each step of the way. The hardest part, according to Prudential Senior Vice President and CIO William D. Friel, is scrubbing the data. Chief IT Architect Lacovara agrees. He explains that as the systems feeding the data warehouse are updated (as often as once a week), "they change to some extent the data that is coming in." IT has to find ways to translate different file layouts across different systems and get terms like balance and premium to mean the same thing so that when they are put into data models, analysts can get accurate results.
"We use rules about the structure and translation of files and set a rigorous process for people to inform us of changes in the data sources," says Lacovara. "It's a job. There's no magic, just a lot of work. You need to impose discipline. You can't leave things up to wizardry."
Dolores O'Riordan (another hypothetical customer) was not a Schwab client when she first called the company to inquire about an IRA. During that initial contact, the call center agent captured her name, address and request in Schwab's central database (a.k.a. the core production system) so that the marketing department could send her information about retirement funds. The call center rep also asked her if she owned any other stocks or equities and whether she considered herself a beginning investor.
All business transactions, including the call between O'Riordan and the agent, are processed and stored on the core production system, according to Beth Devin, senior vice president of retail technology for the San Francisco-based brokerage. Devin's team customised applications from Siebel Systems to track Schwab's business development activity with individual investors on this production system. The next time O'Riordan phones the call center, the agent knows exactly what took place during her first flirtation with Schwab and has an idea of her needs based on the information the analysts pulled out of the data marts and brought back up to the production system. Similarly, a sales manager can find out if one of her staff is following up with O'Riordan's request for information on IRAs.
The information about O'Riordan, and all of Schwab's 7.1 million active accounts, is replicated in the company's data warehouse, which serves as the data distribution house for the database marketing systems in retail. Devin explains that analysts use the data warehouse for reporting, analysis and what-if trending. It is fed by multiple production systems, including the company's one customer production database. The data warehouse feeds different data marts inside each of Schwab's business units. Marketing analysts work off these data marts to do their modelling and sort Schwab's customers into different groups according to gender, affluence, trading activity, channel preference and the frequency with which they deal with Schwab.
When the modellers analyse the customer data, they also try to make inferences about customers' needs and preferences using so-called inference algorithms, a tool Devin considers "way powerful." If they find that O'Riordan's call for information on IRAs was echoed by hundreds of others who share similar characteristics (for example, they are all in their late 20s and earn between US $40,000 and US $60,000), the marketing department will ramp up its efforts to market IRAs to that segment. The information about which customers fall into various segments is fed into the production system so that all of the different touch points (Internet, call center or branch office) have access to it. These different channels interface with the core production system, making the transfer of data from one system to another and from business analysts to the call center reps virtually seamless.
"It's a loop, really," says Devin, explaining the way information about customers is aggregated, stored, processed and disseminated. "As customers do business with Schwab, we capture information about these interactions in our one central customer database. We feed this information down to the data mart, and it gets in the hands of the marketing analysts, who have tools to analyse the data. After the analysis, they create new data and feed it back up to the database weekly or nightly, depending on how much information we have, completing the loop up to the production system. Because of the way our systems are architected, all of our channels interface with the same production database.
"We have some basic infrastructure in place and a number of initiatives under way that are going to make our customer segmentation strategy even more robust," says Devin. "It's not just about data marts. It permeates all of our operational and touch-point systems."
All three financial service providers agree that although technology plays a fundamental role in enabling customer segmentation, the key to really making it a success is believing that, as Royal Bank's Lippert puts it, "understanding the customer is something you can translate into value."
Staff Writer Meridith Levinson can be reached at firstname.lastname@example.org.
Honourees in the Story
Royal Bank of Canada, Toronto www.royalbank.com The Prudential Insurance Company of America, Newark, N.J. www.prudential.com Charles Schwab & Co., San Francisco www.schwab.com.