Selecting Data Mining Tools to Increase Customer Value
The art of mining data to develop insightful and actionable results that will effect the bottom line in terms of improved efficiency and increased profitability requires a mix of skills and tools. Each data-miner has varying levels of statistical, programming, database, and business-process skill. The available skill sets are a primary driver in dictating the type of data-mining tools that should be used. Other drivers include the primary data-mining objectives, available data and format, and model-deployment timing.
Before deciding on the appropriate data-mining tool organizations should spend time understanding how the underlying data systems will effect the power and application of the subsequent analyses. Many operational systems are not built to support data-driven analytical applications. The data and, ultimately, the current database systems are at times the bottleneck that no data mining-tool enhancement can overcome. Resolution of these issues should be initiated before embarking on an analytic platform decision.
Some would like us to believe that one analytical tool fits all, but nothing can be farther from the truth. Data-mining software must be evaluated based on data available, personnel skills, objectives, expectations, and price. If expectations are to deliver a full suite of near-real time, event-driven models, the first question should be, Do we have the personnel to complete this mission? The data-mining tool would then be selected based on maximizing available skills, given obvious budget constraints.
Once available skills, objectives, and expectations are realized the next series of questions in the selection process should be:
What data is available?
How will the models be implemented and how often?
What are the data mining objectives and what analytical techniques may be leveraged?
Without data attributes to segment customers and transactions data-mining technology may not be worth the investment. The data lays the groundwork for the proposed segmentation; data-miners maximize the effectiveness and timeliness of the segmentations. If a significant amount of transaction data must be processed then the analytical suite must have the tools to complete this process.
Since data preparation and processing prior to the analysis typically requires at least 80 percent of the time and may be computer intensive, it is imperative that the tool be able to deliver as promised. Many promise that they can handle any data, but most can only handle the data once the majority of the processing is complete.
Two basic questions that may qualify the software as processing-capable might be, is the software capable of matching data sets and can the software aggregate monthly transactions up to a unique customer?
If the answer is no
and significant processing is required, the processing must be complete in the database environment. Several issues are involved with completing the processing in a database. The general overhead that comes with undo and archive capabilities make databases suboptimal batch-processing environments. Also, precious resources are used when processing data in a database. If possible, try to transfer as much of the processing to an environment made for that purpose. Many of the analytical suites, as well as other tools, are efficient data-processing environments.
An additional consideration is data connectivity. Select a tool that provides a native connection to the marketing database. This allows for quicker and more efficient data transfer, model build, and model implementation. If it is estimated that the model implementation (model scoring) may require a significant manual commitment, then the tool selected should offer an automation technique or macro programming language.
Next, develop a brief road map of the proposed processes to be analyzed, which may include campaign efficiency and contact strategy, cross-sell, retention, and customer profitability opportunities. The appropriate statistical and data mining techniques should be identified for each of the opportunities. The tool selected must provide the data mining capabilities identified in the road map.
Most of the analytical techniques may be divided into two main categories: segmentation and prediction. Segmentation techniques place customers in buckets with similar attributes. The segments may be used to determine the appropriate offer or channel as well as to enhance the predictive models. Predictive techniques are used to enhance targeting efforts. Targeting includes enabling offer delivery to those most interested as well as determining the appropriate offer timing.
Data-mining software should be evaluated in parallel with all knowledge-driven marketing efforts. The marketing processes, database(s), and analytic software should deliver a seamless, holistic view of the customer that may be used to deliver an enhanced customer experience. Purchasing an enterprise level-analytic suite when the only data available is in old legacy systems or when a clear customer marketing strategy is not in place, in most instances, will not deliver significant value. However, analyzing and upgrading each part of the knowledge-driven marketing process, in union, will maximize analytical efforts as well as realize the ultimate goal: to increase customer value.
About the Author
Roman Lenzen is a senior analytical consultant at Quaero, a provider of knowledge-driven marketing services. With several years of analytical CRM experience, Lenzen has delivered value added analytical processes in several industries including financial, hospitality, telecommunications, circulation, pharmaceutical, insurance, and retail. His significant analytical, technical, and business process experience provides a unique perspective on improving process efficiency and customer profitability. Contact him at lenzenr@Quaero.com