Vol. 5, No. 10; October 2005 Printer Friendly PDF version

A FRAMEWORK FOR EFFECTIVE CUSTOMER ANALYSIS

by Ken Collier, Senior Consultant, Cutter Consortium

Our survey of 106 companies reveals that a large majority of respondents are capturing and analyzing customer data. More than three-quarters (80%) of the respondents link transactional detail to specific customers (see Graph 8), and many are gathering customer data through other means, such as surveys (61%), third-party providers (23%), and shared partner data (22%). In fact, only six respondents (6%) report that they are not collecting any customer data (see Graph 9).

Referring back to Graph 5, in spite of the apparent interest in better understanding customers, only 21% of survey respondents feel that their organization is effective or very effective at leveraging customer data into analysis. The remaining 79% feel that their organization is, at best, only somewhat effective in this area.

While it appears that the emphasis on customer relationship management (CRM) during the past decade has been heard, it also appears that effective use of customer data to drive business value remains elusive. This isn't terribly surprising considering the complexities of implementing a robust customer analytics infrastructure. I don't just mean the systems to support customer analysis; it is also very challenging to establish an effective customer analytics business initiative and to measure the impact and ROI of such an initiative.

Graph 6 reveals that 53% of survey respondents report that projects using customer data analysis are normally championed by business executives. Only 17% of these projects are championed from within the IT department. As Gabe previously discussed, however, the most significant challenge faced in approving such projects is the difficulty in showing acceptable ROI (43%), followed closely by corporate inertia and resistance to change (42%). So while business executives subjectively value customer analytics, they have difficulty objectively justifying, obtaining buy-in, and executing on such projects.

Finally, business decision makers are typically unable to access and utilize customer data in their analyses. Of survey respondents, 41% report that data exists in the IT systems, but business users cannot easily access it (see Graph 7). Another 31% of respondents state that sufficient customer data does not exist, and only 28% of respondents feel that decision makers have adequate access to customer data for analysis.

Our survey results support the notion that there is a significant disconnect between the organizational desire to use customer analytics for decision support and the ability to execute on this desire. There are three key components of an effective customer analytics initiative (see Figure 1). These are each complex components that must be effectively executed to fully realize the benefits of customer analytics. This article examines each of these separately in order to establish a framework for successfully architecting a complete system.

Figure 1 -- Components of effective customer analytics.

ESTABLISH A COMPELLING BUSINESS CASE

What Would You Do Differently If You Knew Every Customer Personally?

As a business intelligence (BI) and data warehouse consultant, I would estimate that more than 80% of my client engagements since the late 1990s have been associated with CRM. I always ask the business sponsors of these projects what they would do differently if they knew each of their customers personally. It's a simple question, but it helps establish a clear purpose. Sometimes organizations become so enamored with the idea of customer analytics that they lose focus on what is really important -- the business actions and decisions driven by these customer analytics.

My good friend owns a bicycle shop in the town where I live. As a cyclist, I am one of his regular customers. When he first opened his shop, he routinely gave discounts to customers that he knew personally. Since we live in a small community, he knows most of his customers personally, and he discovered that he was discounting a majority of the store's transactions. Moreover, new customers quickly learned that it was in their best interest to get to know the owner personally so they could get the "good buddy" discounts.

As a small business owner, my friend was inadvertently cannibalizing his own profits without materially improving the growth of his new business. When he realized the negative impact of his discounting practice, he faced a new problem: how to halt the practice without adversely affecting his customer relationships. His solution: customers now receive a store gift certificate at the beginning of each new year in the amount of 10% of their previous year's total purchases. The "discount" is a direct reward for customer spending, and customers still feel as though they are getting a break. Yet the value of the discount to customers is greater than its cost to the store owner. This is not a new concept, just a unique application of the incentive program in a small business.

Now that the bike shop is capturing and analyzing customer transactional detail, one of the most interesting discoveries is that the "best" customers from a profitability perspective are not necessarily the personal friends of the owner. A few years after implementing the new program, I asked the owner if the benefits justified the cost of tracking and analyzing customer transactions. Not only were annual profit margins approximately 10% higher, but he discovered that many of the avid cyclists who spend a lot of time in the shop don't purchase a lot. His most profitable customers are not his most frequent customers.

Business Objectives

The point of this simple example is that an effective customer analytics initiative begins with a clear and compelling business objective. Periodically I work with companies that have a CRM or customer analytics initiative without establishing a set of concrete business drivers to justify the cost. One of my clients justified its initiative by the simple fact that its closest competitor was using customer analytics therefore it needed to do the same to remain competitive. This is an insufficient business justification for establishing an effective customer analytics program.

Customer Analytics Roadmap

As with any BI strategy, customer analytics should follow a progression, detailed below and in Figure 2.

Figure 2 -- Business objectives drive analytics.

1. Business objectives. Customer analytics must be driven by concrete business objectives. These are strategic objectives such as, "Achieve a 1.5% increase in revenue by acquiring new customers who are similar to our current most profitable customers."
   
   
2. Identify action opportunities. Concrete business objectives directly drive the identification of specific business action opportunities of the form, "If we knew X, then we could do Y." For example, "If we knew the common characteristics and behaviors of our most profitable customers, then we could design a campaign to acquire new customers with similar characteristics."
   
   
3. Design analytical strategy. These business action opportunities directly drive the specific analytical strategies needed to "learn X." Continuing the example, we might design an analytical strategy that includes assigning a profitability score to each customer (a significant analysis effort by itself); developing a customer classification model (e.g., decision tree) to group customers by profitability; and developing an actionable profile of high-profit customers.
   
   
4. Data collection and management. The analytical strategy dictates the type and quantity of data required for successful analysis. This topic is directly related to the need to have sufficient technical infrastructure and data acquisition strategies in place. In our survey, only 47% of our survey respondents say they are able to use transactional detail to analyze customers (see Graph 10). Only 36% of business-to-business (B2B) and 35% of business-to-consumer (B2C) respondents are able to capture additional customer detail such as names, addresses, and phone numbers. More difficult still is the ability to capture psycho-demographic customer data. Only 11% of respondents are able to collect consumer demographics; however, 29% are able to collect corporate demographics. The limitations of your data collection abilities may force you to redesign the analytical strategy developed in step 3.
   
   
5. Actionable results. Once you have successfully executed your analysis strategy, you must be able to convert the analytical results into business action. This step is the execution of the "... then we could do Y" portion of our earlier assertion. This action may be personalization of offers to customers, customization of products and services, or highly tailored or targeted marketing campaigns.

Business Case

Ultimately, it is the elements of the customer analytics roadmap that determine whether you have a compelling business case for funding the project. Simplistically speaking, does the value of "doing Y" justify the cost of "learning X"?

Unfortunately, it isn't quite that simple. If the technical infrastructure and data acquisition strategies do not exist, then there is a substantial up-front investment required to implement these components.

A more realistic business case must consider the portfolio of all the business objectives related to customer data analysis. Here we can take advantage of economies of scale. In my work with various companies seeking to analyze customers, I have dealt with finance departments, marketing, sales, product management, and other sponsoring business units. Regardless of the sponsoring unit, I consistently find a high degree of overlap between the data elements needed to achieve the business objectives. In other words, acquiring customer data to serve one analysis objective will almost certainly serve many other objectives. The key to achieving maximum ROI is a focus on concrete business objectives and actionable outcomes.

TECHNICAL INFRASTRUCTURE

As previously stated, 41% of our survey respondents report that customer data exists in their IT systems, but business users cannot easily access it. Based on my experience, I am surprised that this percentage isn't higher. Nearly every company that I work with has executives who are frustrated by their lack of easy access to customer data. Without a sufficient technical infrastructure, customer data analysis is impractical if not impossible. Your technical infrastructure must include four elements:

1. Customer data management system
   
   
2. Business intelligence capabilities
   
   
3. Customer analytics taxonomy
   
   
4. Presentation application

Customer Data Management System

Transactional database systems are a mainstay in most companies today. Unfortunately, many of these companies overload their transactional systems with reporting applications that are inappropriate to the task. Transactional reporting should be operational in nature, not analytical. For example, checking inventory for product availability is a reasonable operational reporting activity; evaluating the seasonal buying preferences by region and time of day for all male customers is not. Transactional systems are not optimized for these types of analytical queries.

Companies wishing to have ready access to their customer data should consider a data management system that is optimized for analysis. A data warehouse or data mart is well suited for highly interactive, exploratory customer data analysis. A properly architected data warehouse provides data validation and cleansing opportunities. It also houses substantial historical detail to support trend analyses and other comparison analyses. The data warehouse provides the basis for a broad spectrum of analytics including multidimensional exploratory analysis, data mining, quantitative analytics, data visualization, and ad hoc querying.

However, the development and maintenance of a data warehouse requires a specialized set of skills that is not present in all organizations. Proper design and implementation of a data warehouse is a systems integration and customization effort that generally requires assistance from consultants with experience in such projects.

It can be difficult to justify the cost of a large data warehouse or data mart. In a Cutter Consortium Business Intelligence Executive Report [1], Cutter Business Technology Council Fellow Jim Highsmith and I describe agile data warehousing (ADW). ADW is a method for delivering early and incremental value in the development of a data warehouse. Not only does this approach help ensure that the project will meet the needs of end users, it also enables you to realize early ROI and tailor the system to maximize your ROI.

Alternatively, specific CRM applications have appeared on the market over the past several years. These tools are designed to incorporate the operational activities of sales and marketing, such as lead generation and campaign management. These tools also incorporate some degree of customer data analysis and reporting. Most of these tools are designed to integrate with existing transactional systems to capture customer behaviors.

Many companies I work with have an assortment of unsanctioned and semiofficial customer data "repositories." These are the Excel spreadsheets and Access databases stored on business users' workstations. They are built by enterprising business decision makers when they can't easily access the official customer data that resides in IT systems.

I recently worked with a company whose finance department had a large number of these type of Excel spreadsheets. Each finance department member had his or her own customized spreadsheet -- each with similar content but used for different decision support purposes. As we began evaluating all of these unofficial spreadsheets, we uncovered a potentially serious problem: the business rules used to calculate gross revenue, gross profit, and net profit were not well documented and therefore not consistently implemented in each of the spreadsheets. In many cases, one finance department employee's figures were substantially different from another's. It was evident that the company needed a common shared system to provide this type of customer information. It opted for a finance data mart, which established the technical infrastructure needed for the various customer and financial data analysis needed by all users in the department. The development of the data mart necessitated a clear definition of the finance business rules so that everyone had a shared understanding of revenue, profit, and other financial measures.

For obvious reasons, these customized spreadsheets and unofficial databases should be discouraged. However, these should be harbingers of the demand that exists for systems that enable easy access to customer data.

Business Intelligence Capabilities

The analytical capabilities present in the organization are a critical part of your technical infrastructure. These specialized sets of skills may include multidimensional data modeling, data mining, data visualization, statistics, and report designers. Not every company has, or needs, expertise in all of these areas. However, as you develop your analytical strategy (phase 3 of the customer analytics roadmap), you must evaluate the capability of your analysts to execute the strategy.

Customer Analytics Taxonomy

Our survey reports that 56% of responding companies utilize customer data for campaign management; 52% to analyze customer profitability; 26% to analyze customer churn; 20% for propensity scoring; and 17% for other types of analysis (see Graph 11). Remember, however, that less than 25% of the respondents state that their organization is effective at leveraging customer data into analysis.

The following outlines a fairly comprehensive customer analytics taxonomy. This taxonomy is useful for assessing the capability of your organization to conduct the desired analysis. First I'll define the metrics and then outline the methods and skills needed to generate each one.

Customer Behavior Metrics

These are customer analyses that evaluate customer patterns and behaviors:

• Timeliness. What is the likelihood that a credit customer will pay her balance each month?
  
  
• Risk score. What is the credit risk of a customer?
  
  
• Purchase behavior. Who bought what products? What products are selling best (by region, by field location, by marketing event, by channel, etc.)?
  
  
• Affinity analysis. Which items tend to be purchased together (cross-sell/up-sell analysis)?
  
  
• Propensity analysis. Who tends to buy what products? What is the profile of customers who buy office furniture and don't buy paper? What is the profile of customers who shop by catalog versus call center?
  
  
• Timeliness profiling. What is the profile of customers who take 60 days to pay their balance versus less than 30 days?
  
  
• Risk profiling. What is the profile of our greatest credit risks?
  
  
• Event behavior. What is the customer's buying behavior during a sales/marketing event? How does this compare to normal behavior? Did the event cause higher spending per order? Was there a change in post-event behavior?
  
  

Customer Value/Profitability Metrics

These metrics evaluate the customer's monetary value to the business:

• Current profitability. What is the net value added by the customer considering monetary value, profit margins, and service costs?
  
  
• Potential profitability. What is the greatest profitability expected from customers like this one?
  
  
• Future profitability. What is the predicted future profitability based on past behaviors of similar customers?
  
  
• Profitability profiling. Who are our most profitable customers? Are they who we thought they were? Who are our unprofitable customers? Are we investing resources in servicing them?
  
  
• Profitability conversion. Are our customers reaching their profitability potential? Which ones aren't? Is there a natural grouping of these customers?
  
  
• Wallet share. What percentage of the customer's budget for our products is being spent with us?
  
  

Lifetime Value Metrics

These metrics forecast the customer's expected long-term value to the business:

• Lifetime value (LTV). Based on the customer's profitability (present and future) and his estimated purchasing life expectancy, what is the estimated total net value of the customer? What is the behavior profile of customers like this one over time?
  
  
• Potential lifetime value. What is the maximum LTV of customers like this one?
  
  
• Lifetime value profiling. What do high-LTV customers "look like"? What about low-LTV customers? Are they the same as the high- or low-value customers?
  
  

Loyalty Metrics

These metrics describe the customer's loyalty to the business:

• Recency. How recently has the customer purchased?
  
  
• Frequency. How often does the customer buy?
  
  
• Monetary value. How much does the customer spend?
  
  
• Churn. Which customers are likely to shift their loyalty to our competitors?
  
  
• Acquisition. Who are our customer prospects? Which ones are likely to have higher value?
  
  
• Retention. Who are our defectors, defection risks, diminishers, business risks, and customers who are "maxed out"?
  
  
• Growth. Who are our newly acquired customers, the "middle grounders," and the loyalists? How can we migrate the first two upward?
  
  
• RFM1 profiles. What do high-RFM customers "look like"? How about low-RFM customers? Are there any natural groupings of customers with respect to RFM?
  
  
• Retention profiles. What do defectors "look like"? How about customers who are "maxed out"?
  
  
• Growth profiles. What do our next best customers "look like"? How do they tend to respond to conversion efforts?
  
  

Responsiveness Metrics

These metrics describe the likelihood of customers to respond to marketing events and triggers:

• Response score. How likely is each individual customer to respond to an upcoming event? To another marketing trigger? In general?
  
  
• Response modeling. Who is most likely to respond to an upcoming event based on this event last year? Based on other similar events?
  
  
• Responder analysis. In general, which customers tend to respond to marketing triggers more than others?
  
  
• Response profiles. What does the typical responder "look like"? What is her buying behavior? How profitable is the typical responder?
  
  

Campaign Metrics

The following is a list of measurements and models that can be developed from event data:

• Event ROI. What was the profit per dollar spent on this event?
  
  
• Estimated value add. What was the added value of this campaign compared to doing nothing?
  
  
• Event lift. How much of additional purchasing activity can be attributed to this campaign?
  
  
• Event effectiveness. How effective was the event compared to non-event purchasing trends prior to the event, after the event, compared to the plan, and this time last year?
  
  
• Profitability cannibalization. Did this event reduce sales profit from another event, category, brand, etc.? (Note: if movement to a higher LTV brand or channel occurs, this may be optimal for the organization.)
  
  
• Event cannibalization. Did this event take sales away from another channel, event, region, etc.?
  
  
• Event profitability. What was the profitable sales growth that is attributable to this event?
  
  
• Cross-channel effectiveness. How well did the event marketing effect purchasing in all channels? Was the lift similar across channels? Was the effectiveness similar across all channels?
  
  
• Media effectiveness. How effective is each medium at increasing customer response? Which medium is most effective? Which is least effective?
  
  

Analytics tend to be either descriptive or predictive in nature. That is, sometimes we use analytics to evaluate what happened (descriptive), while other times we use analytics to forecast what will happen (predictive).

Analytical methods range from basic to advanced in terms of the skills required to conduct the analysis and interpret the results. Every business decision maker is familiar with the predefined tabular reports and graphs that are ubiquitous in today's business processes. Most business users are comfortable building spreadsheet reports. These types of analytical methods are more basic. While some decision makers may have a rudimentary knowledge of statistical analyses such as logistic regression, most do not have the skills necessary to build effective data mining models. These are more advanced methods requiring specialized skills. Figure 3 depicts this analytical spectrum.

Figure 3 -- Business intelligence techniques.

Table 1 rounds out our customer analytics taxonomy with a guideline of the general analytical method typically associated with each metric; a viable technique for implementing the metric; and the type of skills required for executing the technique. Note that these methods are not exclusive to these metrics; there are alternative approaches that may be equally effective for many of the metrics.

Table 1 -- Customer Analytics Technology

We can generalize the ADW methodology described earlier to the broader spectrum of BI development efforts including data mining, statistical analysis, and reporting. By implementing some of the more basic, descriptive analytics first, you can deliver early value to business users while you establish strategies to deliver more advanced and predictive analytics.

Presentation Application

The fourth key component in your technical infrastructure is the means by which decision makers access the customer data and analytical results for business action. There is a broad range of options for presenting customer data to business users for evaluation and action.

Perhaps the simplest of these options is the generation of spreadsheet reports that can easily be delivered to users via e-mail or a document management system. This mechanism is often favored by power users since it offers the flexibility they need to conduct their own custom analyses or hypothesis testing. Powerful software such as Excel offers the ability to develop customized spreadsheet applications that connect directly to the customer data management system ensuring that users are always viewing the most recent data. This presentation mechanism supports basic reporting, charting, graphing, and multidimensional OLAP reporting (using pivot tables and pivot charts).

Web browser-based applications and BI portals offer a presentation mechanism that is well suited to less sophisticated end users as well as power users. These applications typically require more substantial development and usability design than spreadsheet applications. Not only can these applications present reports, graphs, and multidimensional OLAP reporting, they can also be designed to present data mining and quantitative analytical results as well as advanced data visualization. In addition to the custom development using any of the various portal technologies, there are many BI software vendors that offer browser-based client applications as a presentation option.

Finally, there are many BI software vendors that offer powerful reporting and presentation tools. These tools provide powerful out-of-the-box analytical functionality while supporting substantial customization flexibility to tailor the presentation to end users' needs. These tools can be quite costly, but they substantially reduce the deployment effort required.

DATA ACQUISITION STRATEGY

As Graph 7 illustrates, 31% of our respondents report that sufficient customer data does not exist for analysis. Graph 8 goes on to show that 20% of respondents say their organization is unable to link transactions to specific customers, and another 26% say they are only sometimes able to link transactions to specific customers.

The third leg in our customer data analysis triad is an effective strategy for acquiring customer data. The obvious starting point is to consistently link customer transactions to specific customers. Several years ago I worked with a large book reseller seeking to better analyze customer buying patterns. My client specifically sought to analyze customer propensity for buying from its new online store versus the physical stores. It had previously established a loyalty card program as an incentive for customers shopping in the physical stores. This enabled the company to link approximately 60% of in-store transactions to customers. Since online transactions require customer sign-on, the company was able to link 100% of these transactions to a unique customer ID. Unfortunately, the online store did not have a provision for customers to use their loyalty card, so there was no clear linkage between the in-store customers and online customers. Clearly, the company could not accomplish its customer analysis objectives until this problem was reconciled.

Linking Transactions with Customers

Establishing a noninvasive mechanism for linking transactions to specific customers can be challenging. Your customer data acquisition strategy should effectively link all transactions to the right customer regardless of the method of payment, varying customer nicknames, changing addresses, and so on. The ideal strategy is to establish a unique identifier for each customer and require that identifier to complete each transaction. Ideally, the ID should not be a name, address, or phone number. The challenge is providing the necessary incentive to customers to provide their identifier. Grocery stores have become highly effective at this by providing on-the-spot discounts to those who use their loyalty card. Airline frequent traveler benefits are another example of a highly effective method of linking all transactions to the right customer.

Another difficult challenge for B2C businesses is called "householding." Householding refers to the ability to link multiple customers from the same household with one another. This is not essential for customer analytics, but it can provide some valuable additional insights. One company I have worked with is a nationwide home improvement products retailer. A family involved in a home remodeling project might conduct multiple transactions related to the same project. Moreover, these transactions might be executed by different family members using different credit cards or other methods of payment. Since the company's products are targeted to entire households rather than individual consumers, it was important to link transactions to the same household. Furthermore, many third-party data providers like Acxiom, Experian, and TransUnion offer psycho-demographic data at the household level; for example, Acxiom provides data on such characteristics as household annual income and household size. Householding enables your company to take greater advantage of third-party data.

Similar to householding but for B2B companies is the need to link together multiple transactions generated from the same company. In my experience, the problem of linking transactions to specific customers is not as challenging for B2B companies since these transactions typically utilize invoicing systems or the like.

Data Quality

Regardless of your ability to link transactions to customers accurately, you must also consider the issue of data quality. There are a number of data quality problems that significantly inhibit effective data analysis including the following (see Graph 12):

• Sparse data -- customer data is missing important information and has a negative effect on analysis. This is a common problem with manually entered data or data entry requirements that are not well enforced. Slightly more than half of our survey participants (51%) report this as a problem, which is consistent with my experience.
  
  
• Duplicate data -- customers appear in the database multiple times with minor variations. I recently logged in to my online American Express account and discovered that it has two separate accounts for me: one is associated with a credit card issued by a former employer that was cancelled more than five years ago, while the other is associated with my current card. Since there is no linkage between the two accounts, I suspect American Express sees me as two different customers. A full half of our survey respondents report that this is a problem for their company.
  
  
• Inaccurate or outdated data -- customer data is inaccurate, untrustworthy, or out of date. This is often a problem for online companies that rely on the customer to input their personal data. Spurious data yields spurious analytical results. For example, the American Express account associated with my former employer is full of outdated information about me. Nearly half of survey respondents (47%) report this as a problem for their customer data analysis.
  
  
• Missing data -- entire customer records are missing, or the inability to link every transaction to a customer. The book reseller I discussed earlier had at best a 60% rate of success linking in-store transactions to customers. I've worked with other companies whose linkage rates are well below 50%. Our survey results show that 26% of companies presently have this problem.
  
  

Although there are others, these are the primary customer data quality issues you should be prepared to address. Your technical infrastructure should include data cleansing to resolve these and other data quality problems, and your data acquisition strategy should be designed to preempt these problems as much as possible.

Data Stewardship

Your data acquisition strategy should also address data stewardship. Data stewardship includes validating data quality, championing data acquisition initiatives and data improvement activities, and evaluating the impact of data issues on analytical results. A customer data stewardship council is an effective mechanism for addressing this need. The stewardship council consists of one or more business stakeholders, one or more IT stakeholders, and at least one member with knowledge of the underlying customer data model and customer database.

CONCLUSIONS

The responses to our survey on the business value of customer data suggest that there is significant value in this data for most companies. Unfortunately, there is also significant cost involved in effectively analyzing this data to realize the benefit potential.

My colleague Gabriele Piccoli presents a very powerful set of tools and concepts for assessing the value of your current customer data and the cost of analyzing it. My contribution to this topic is a framework for implementing effective customer analytics in your organization. I have developed this framework after working with many different companies at varying stages of maturity in their customer analytics.

If you are starting from scratch, there are many significant decisions to be made, and the cost and time can be overwhelming. In this case, I urge you to take a highly incremental approach that will deliver an embryonic system early and then work toward maturing that system over time. If you currently have customer data analysis in place, I urge you to evaluate the components of this framework that you may not have considered yet.

REFERENCE

1. Collier, Ken, and Jim Highsmith. "Agile Data Warehousing: Incorporating Agile Principles." Cutter Consortium Business Intelligence Executive Report, Vol. 4, No. 12, December 2004.

ABOUT THE AUTHOR