1	Marketing Excellence Through Data and Analysis The Mercatus Story
2	Overview Integration of Data and Processes Data Accuracy for Reporting and Analysis Consolidation and Cleansing of Customer/Prospect Data Direct Marketing Campaign Optimization Data Mining
3	Integration of Data and Processes Typical Practices Prior to Mercatus Needed data spread across multiple sources in many disparate formats, both internal and external to the company. Customers, prospects, inquiries Order entry systems Specialty Marketing Lists Demographics Marketing strategies developed independently by various marketing channels. Coordination and integration difficult, because their activities are supported by separate, often conflicting, systems. Direct Mail / Retail Stores / Web-based Most clients have been forced to plan and execute their retail, catalog and web-based marketing efforts as totally separate entities. They have little understanding of whether their catalogs are bringing customers into their stores or their web-sites augment or detract from catalog sales. In most cases, they don’t even know whether these channels address the same clientele.
4	Integration of Data and Processes Mercatus-Assisted Solutions Mercatus has designed and developed Marketing Data Warehouses for many of their clients, customized to their clients’ business requirements. The warehouse is a routinely updated central repository of data optimally organized to support strategic marketing activities. Data sources include the client’s order-entry, billing/shipping and other operational systems as well as external sources of marketing information. Operational data is massaged to create a single representation of business concepts out of the different mechanisms supplied by operational systems. Mercatus works closely with BOTH the Marketing and IT Groups to provide valuable insight into a clients’ business goals and the data available and appropriate to support those goals. We identify and obtain the proper data. Errors or inconsistencies in past data collection efforts are identified and rectified prior to warehouse updates. Marketing goals often suggest changes or additions to data collection efforts.
5	Integration of Data and Processes Mercatus-Assisted Solutions (cont.) Mercatus has built customized systems that give their clients a single coherent view of marketing activities across all channels. A client could now know that a customer should be considered “high-value” even though she has never purchased products on the web-site. Client could now know that a customer is purchasing products in a retail store immediately after receiving a catalog, even though he has never responded to a catalog offer directly. They are able to coordinate events – perhaps sending a follow-up email after a direct mailing, or using the direct mail piece to indicate the location of the nearest retail outlet. Web-site interactions can lead to customized direct-mail pieces.
6	Data Accuracy for Reporting and Analysis Typical Practices Prior to Mercatus Many clients are able to produce only rudimentary, often inaccurate, Sales and Marketing Reports Many order-entry and billing systems are structured in such a way that accurate aggregation of sales information across customers or customer groups, territories, product lines, etc., is very time-consuming. Use of transaction processing systems to support such reporting often leads to unacceptable slow-downs in order-taking or billing processes. Inaccuracies and inconsistencies in operational systems and data collection procedures. Most if not all of these issues can be corrected as part of the marketing database update process . Operational systems are built for tracking day to day transactions. They are not optimized to collect historical data. Key marketing information such as promotional history often is not maintained at all.
7	Data Accuracy for Reporting and Analysis Typical Practices Prior to Mercatus (cont.) Misinterpretation of data in their operational systems One client we worked with had a table in their operational systems named “Inquiries”. They would send this table along with their customer base to their direct mail processing house. When ‘inquiries’ were mailed, pieces were sent to every name in the table… neither the client nor the processor realizing that the table was being used for “No-Call/No-Mail” records as well as catalog requests. Most Marketing Groups have a very good sense of their market and client-base. However, without accurate, quantitative information about the purchase behavior of that base, and the response behavior of their prospect universe, their success is limited to their level of intuition. Intuition augmented by factual understanding leads to greater success.
8	Data Accuracy for Reporting and Analysis Mercatus-Assisted Solutions A major focus of Mercatus’ efforts is on data mining activities to extract information about customers and prospects that can be leveraged in support of the creation and execution of strategic marketing and sales efforts Investigate, validate, and cleanse the data in order to ensure accuracy Are the fields always populated? Will missing data be a problem? Are field values legal and reasonable? Do dates fall within date ranges? Are data elements and records being used appropriately? Example – a client that enters coupons as product line items. These must be handled separately to avoid erroneous counts of items ordered, dollars purchased, etc.. Data is transposed to appropriate levels of granularity and aggregation Orders made by multiple members of a household (or employees of a firm) can be aggregated, or not, as needed. Purchase behavior can be examined at weekly, monthly or annual levels; by region, state or territory.
9	Data Accuracy for Reporting and Analysis Mercatus-Assisted Solutions (cont.) Additional tables to augment the data available in the operational systems Custom aggregation tables to support specific reporting needs Third-party enhancement data - demographics, financial data External prospect lists Time tables to include both calendar and fiscal time periods Derived variables generated for decision support and reporting needs Categorical variables - product classifications, regions, etc. Seasonality and purchase timing - date ranges, seasons, holiday periods, etc. Geographic variables – distance to nearest store, reporting by ZIP regions, etc. Rankings, intervals, etc.
10	Data Accuracy for Reporting and Analysis Mercatus-Assisted Solutions (cont.) Specialized aggregations (tables) Gender-specific analysis and reporting Product type, classification Purchase timing, seasonality Geography – sales territories, regions RFM
11	Consolidation and Cleansing of Customer/Prospect Data Typical Practices Prior to Mercatus Large amounts of return mail, because of inaccuracies in names/addresses extracted from order entry and billing systems. Inaccuracies in the long distance telephone companies’ billing systems in the 80’s and 90’s forced them to bill through the Baby Bells (local phone companies). NCOA (National Change Of Address) processing not effectively integrated into billing systems, or subsequent direct mail efforts. Duplicate catalogs sent to a given household. Prior to Mercatus’ involvement, most clients have not yet set up business rules to define households with respect to their products/services or their marketing efforts. ‘Canned’ merge/purge processing – as performed by many Direct Mail processing companies – does not address client-specific ‘quirks’ in data collection systems or policies. Example, a client that cannot delete a name and address from their customer base. So when a company moves or there is a name change, a ‘new’ account is created - with a freeform reference to the new account placed into the old record.
12	Consolidation and Cleansing of Customer/Prospect Data Typical Practices Prior to Mercatus (cont.) Some clients have mailed to external (rented) mailing lists without correctly suppressing their customer base Many clients have no way to properly track customer/prospect Event History. Previous marketing ‘touches’ unrecorded, and thus unavailable for analysis and utilization in future marketing efforts. Multiple members of a household treated individually (sometimes appropriate, but not always). Thus, marketing history to that household is not aggregated. Marketing event history should survive across a household/individual move to a different address – rarely done.
13	Consolidation and Cleansing of Customer/Prospect Data Mercatus-Assisted Solutions Name/Address information is subjected to significant processing prior to placement in the data warehouse and its subsequent use in reporting and marketing extracts Multiple accounts rolled together where appropriate (as in the issue of new addresses being made into new accounts, with old record pointing to the new). All name elements are parsed to determine first/middle/last name components, business components and gender Can be extended to include ethnicity, religious preference, etc.. Address elements are standardized, corrected where appropriate, and geo-coded where possible County code, MSA, lat/long, census block/tract Corporate vs. residential Household relationships are identified according to customized rules developed with client to address their specific industry and marketing needs
14	Direct Marketing Campaign Optimization Typical Practices Prior to Mercatus Many clients simply performing mass mailings to their entire customer base Buyers, inquirers, gift-recipients, etc. all treated identically. No differentiation between high-value, ordinary, low-value customer records Lack of targeted messages Costly over-printing and mailing No product-specific targeting No campaign tracking or analysis No testing of different promotions, because there is no way to determine who responded to a promotion. No way of knowing how well external (rented) mailing lists perform even though large amounts of their budget are spent every year to use such lists.
15	Direct Marketing Campaign Optimization Mercatus-Assisted Solutions Our clients have the ability to trace their Customer Lifecycle from prospects becoming customers through customers churning out (and returning) Promotional event history permits identification of ‘new customer’ attributes. Supports customized contact/treatment of new customers Customer spending behavior can be segmented and characterized – by dollar amount, seasonality, product-preferences, etc.. Trends in behavior leading to service cancellation can be identified. Supports targeted retention efforts Clients can target specific customer segments in their base - offering different promotions for each group Identify the top 20% (or other figure) of high value customers on their customer base. Enhance efforts targeting high-value customers, eliminate mailings to the low-value portion. One current client has saved $700K in 2003 alone by simply targeting their high-value customers, and not mailing their low-value customers. Consumer vs. corporate offerings and promotions.
16	"Through Match-Back Reports our clients..." Through Match-Back Reports our clients can now evaluate the performance of specific promotions Each person/household receiving a given marketing piece (mail offer, catalog, telemarketing call) is known, as is their membership within some well-defined segment of customers or prospects. The purchase (response) behavior of these recipients is tracked during a time-frame relevant to the marketing campaign. Thus, responders are identified, and categorized with respect to the size, type and other aspects of their purchase/response. This tracking can be done by placement of segment-codes and/or tracking numbers on the mail piece… these numbers being collected during order-entry. Often, however, relatively few responders are identified in this way. Match-Back via name/address matching (as part of routine data warehouse update) identifies many more responders.
17	Direct Marketing Campaign Optimization Mercatus-Assisted Solutions (cont.) Match-Back Reports also allow evaluation of third-party mailing lists Example - A catalog client that sells educational supplies and equipment. They were able to see that they were very successful with certain portions of a rather large third-party mailing list and not so successful with others. They are now realizing significant savings by leasing only those segments of the overall list that are most successful. Evaluation of promotion alternatives is now quite straight-forward Differing catalog appearance and content Effectiveness of targeting differing segments Comparative response rates resulting from specific offers (coupons, discounts) provides insight into subsequent marketing efforts
18	Data Mining Typical Practices Prior to Mercatus Many clients have little understanding who their customers are Unable to characterize their customers – demographics, lifestyle, etc. No way to understand their purchase behavior or tenure – seasonality of purchases, product-preferences, etc. No mechanism to discover segments or niches in their customer base What understanding DOES exist is non-quantified, and thus difficult or impossible to act upon Most clients do very little sophisticated data-driven marketing, with poor results Extracts built by an IT staff are usually inappropriate for analytical efforts. Incorrect or inconsistent data elements are included, other vital elements are overlooked. They are rarely sampled correctly, which severely impacts the relevance and accuracy of the statistical techniques used. Particularly when the analysis itself is done by a vendor working with such data extracts, the results can be misleading. Back-end validation of the analytical results is often overlooked. Thus, not only is the effectiveness of the analysis itself unknown, but valuable insight into future efforts is not obtained. Effective data-mining requires personnel whose experience includes data systems, statistical analysis and strategic marketing – a rare combination.
19	Data Mining Mercatus-Assisted Solutions After gaining a thorough knowledge of our clients’ marketing goals and the various data sources to support them, Mercatus performs many different types of analyses in support of targeted marketing efforts. Predictive modeling – statistical process for predicting future behavior Based on randomly selected sample of data describing past or current activity Model results are applied to the client’s entire data universe to predict probable prospect or customer behavior Multiple models to predict responsiveness… specific to customer segments, lifetime value, churn, etc. A cost-effective way to expand the customer base. Enables targeting of optimal prospects for specific campaigns and products
20	Data Mining Mercatus-Assisted Solutions (cont.) Profiling and segmentation Descriptive profiles of distinct customer segments Customer segmentation permits targeted efforts for improving follow-on sales and cross-sales We work closely with our clients to determine the most appropriate analysis to perform for the task at hand. The end result for our clients is to target the right offer to the right person at the right time. The bottom line to our client is optimizing their budget, by discovering the people that are most likely to respond to an offer, and the offer to which they are most likely to respond. We perform back-end analyses on all modeling efforts. This gives our clients a way to evaluate and refine their marketing efforts. Detailed tracking of target-marketing success rates Sales and promotion analyses suggest new approaches and product lines
21	Data Mining A Cautionary Tale Most clients that perform some Predictive Modeling or Niche Identification with vendors obtain a single all-purpose model… which they are then left to implement on their own. Mercatus prefers to work closely during all aspects of an analytical effort – to ensure the data used as input is appropriate, the analysis itself done properly, and the resulting conclusions are effectively and properly applied to the marketing task In the early 90’s, MCI had tremendous success with a Telemarketing Response Model for Domestic Long Distance service offerings (built by Mercatus, as were all of MCI’s models and segmentation schemes during this period). Soon ALL the different Marketing Groups wanted to use it - Direct Mail, International Calling, Friends & Family, etc.. However, the people that were likely to respond to a long distance telemarketing promotion were not the same people that were likely to respond to a direct mail campaign. Further, international segments of a Long Distance companies’ customer base are very different than those customers that make domestic calls. Finally, responders to telemarketing were uniformly low-usage customers, indicating that a product line specific to them should be developed (it was). By understanding the details of the client’s business as well as those of the analytical effort itself, Mercatus was able to ensure that MCI’s predictive modeling activities remained highly effective, leading to a increase in market share from 8% to 16% over the course of Mercatus’ involvement with MCI.