01 October 2004 Features

 

Analyse this

With increased system performance and a need to know more and more about customers to win and keep their business, Elizabeth Lumley finds data mining back in vogue

Data mining, or predictive analysis, has had a problematic reputation with banks. Often thought of as an IT issue, it has been generally seen as a black box process that gives little transparency and rarely follows business logic or rules.

However, changes in global banking environments have led many to re-examine how they can better use their large stores of historical data. In addition to years’ worth of transactional data, banks also store ever-increasing customer data from internet usage, ATMs and call centres. According to one software provider, 20 years ago, a bank might store up to 10 variables per customer record. Today it will store over 2,000 variables per customer.

New regulations and guidelines such as the upcoming Basel II risk directives and the Sarbanes-Oxley and Patriot Act in the US are calling on banks to have ready access to clean and comprehensive client information. Calls to expand profits and minimise costs also push banks to look for new tools to support customer relationship management (CRM) and fraud detection.

The goal of data mining, or predictive analysis, is to discover new correlations, trends, patterns, relationships and/or categories by digging through large stores of data armed with business tools based on statistical and mathematical techniques. These patterns can be used to forecast customer behaviour, detect fraud, monitor product performance and assess risk.

“Historically, data mining was much more of an art than a science requiring highly specialised programming and analytical skills,” says Kevin Cavanaugh, vice-president of technology at Unica.

Data mining used to be seen as an IT-driven project, according to Elizabeth Gooch, managing director of EG Solutions. IT may collect data, but others decide “what is the data and is it the right data”, she adds. EG Solutions provides data mining tools and consulting.

However, improving on CRM, considered by some to be the Holy Grail of data mining, requires a business process that can be augmented and understood by bank staff not versed in mathematical techniques or IT requirements. CRM staff once conducted manual exploration of customer records to discover patterns. The amount of varied content in most data warehouses now makes manual processes impossible.

Lack of confidence in the quality of the customer information is one of the problems associated with trying to use these huge stores of historical data for CRM purposes. “The data in operational systems was not collected with data-intensive applications such as data mining, business intelligence and CRM in mind and as such is not high enough quality to be successfully deployed in these applications,” says Tom Golden, vice-president of marketing at Similarity Systems.

Similarity provides Athanort, a suite of data quality software products that identify and correct data quality problems based around an end-to end data quality management and improvement process. Customers include Rabobank and Bank of Ireland.

Poor quality data can include duplicate information, incorrect or incomplete fields or flagging information that needs to be corrected by call centre staff during customer interactions. “Cleansing, standardising and consolidating data in this way is more amenable to analysis,” he says.

In addition to poor data quality and data availability, predictive analysis projects tends to run into problems when statistical models are not updated frequently enough to be relevant and data warehouses omit various customer variables such as post codes or telephone numbers.

Statistical models used in data mining must be timely and specific in order to offer consistent and confident predictive analysis. “Banks often use risk models that are three to four years old and are not relevant,” says Joerg Rathenberg, vice-president marketing and communications at Kxen. Traditional risk management models do not differentiate between variables that are high risk and those that are low risk. “Risk and non-risk are not the same.”

“Traditionally banks can take up to four weeks to bring a model to production,” says Rathenberg. With Kxen that process can take “minutes or hours”. Customers include HBOS and Disbank in Turkey.

Successful data mining can produce information that has a number of benefits for banks. It can help banks keep existing customers profitable as well as acquire relevant new clients. Customer service can become quicker and more comprehensive by giving staff one, overall view of the customer. Marketing campaigns can be brought to market quicker than the competition and new products can be targeted to specific groups. Fraud and risk detection, as well as identifying new entrants and changes in the market can be achieved. Sensitivity to interest rate changes can be monitored between various groups.

After Disbank in Turkey issued its millionth credit card, the need for growth drove it to take additional risk in order to gain more clients. Using manual processes, the bank was not catching fraudulent credit card applications, prior to issue. The bank is now using a data mining system from Kxen to monitor client behaviour. By using the Kxen system, Disbank identified up to one to two per cent of its credit card applications as fraudulent and is now catching up to 80% of these cases before the bank issues a card.

In addition to customer marketing, new global guidelines and regulations are forcing banks to better manage and use their data. Sarbanes-Oxley in the US requires banks to demonstrate the availability, integrity, non-repudiation, and confidentially of financial information. Global risk management regulations based on the Basel II reforms, as well as the Patriot Act in the US, advocate the use of statistical and pattern analysis to better understand banks’ performance and customers.

All of these factors are pushing banks to create an overall view of the customer to allow marketing staff, customer service staff and fraud detection staff to make decisions based on real-time analysis of client records.

Banks are now gathering information from internet usage, web traffic, call centres, ATMs, and portals, says Marcel Holsheimer, vice-president of vertical product marketing at SPSS. All of this information must be made available, as a single view, in a data warehouse. This allows bank staff to make customer analysis “at the moment of interaction”, says Holsheimer.

SPSS provides predictive analysis tools to support the real-time understanding of multi-channel financial information. Customers include ABN Amro and Rabobank.

However, often the data stored from avenues such as the Internet and call centres tends to be unstructured and ill-suited to mathematical analysis. Call centre logs, written by bank staff, as well as customer emails, are organised in a way that make it difficult to find patterns, says Jason Goodwin, head of customer intelligence solutions at SAS.

SAS provides data mining tools that can analysis call data, mail-order addresses, sales histories, POS data, web transactions and free-form text notes as well as more structured client data. Customers include GE Capital and ING Direct.

Data mining projects have had “bad press” in the past because operational CRM projects have had no “return on investment, because there was no business intelligence interested with the client”, says Goodwin.

SAS recently conducted a survey that concluded customer service, above price and relevant products, had the most influence on client loyalty. Quality customer service “requires a deep level of understanding of who the customers are”, says Goodwin, “and you only get that understanding with data mining.”

Data mining tools must be integrated with operational environments, such as marketing or call centres, in order to adequately action the data analysis, he says. Because of this the data mining tools need to be easy to use, provide a wide range of analytical techniques for various results and allow for model and document sharing.

“Traditionally, various client information had been stored in separate data warehouses,” says Duncan Ross, assistant director, advanced business analytics, Teradata, a subsdiary of NCR. Deploying that data back into a separate data analysis models is a “long process”, he adds. Teradata provides data warehousing, analytical CRM and business intelligence tools. Customers include Barclays, Lloyds TSB, AIB Bank, The London Stock Exchange, BNP, Bank of America, Wells Fargo and Capital One.

“Today it’s not whether a company can benefit from data mining, it’s about how effective companies are at leveraging their data assets by applying predictive modelling and then how quickly they can act on this information,” says Cavanaugh at Unica.

Unica provides the Affinium system which performs marketing analytics campaign design and execution, lead management and marketing resource management.

Interestingly, it is the larger, tier-one banks, with their enormous databases of historical data gained from years of acquisitions and growth, that are having the most success with implementing data mining projects.

“Banks with assets more than $10 billion have some sort of data warehouse used for customer segmentation purposes that store market, customer and transaction data,” says Steve Pentin, product manager, Fincentric Corporation. Banks with assets lower than $10 billion, “barely touch” customer and transaction data because the data is not in a useful form, and to covert that data into something useful is very difficult, he says. This data is typically stored in product-specific silos, however to get the overall view of a customer, banks need to analyse client data across those silos. “Matching one, let alone a million customers, is very difficult,” says Pentin. “It takes a lot of time and money.”

Tom Stock, senior vice-president, FTI, agrees with the current view of data mining. FTI provides the Street Model system and Street Consolidator system for data modelling and analysis.

“From a customer service standpoint, customers touch different pieces of the organisation,” says Stock. Banks need one single view of the customer whether they are retail or commercial.

“Banks tend to underestimate data analysis,” says Stock, “it is actually very complicated.” However, for data mining projects to work effectively, data must be clean and “stored in a way that is easy to access”.

It can never “provide reliable results with poor quality data — it’s the old garbage in, garbage out argument”, says Golden.