Methodology Diagram

CRS uses a proven methodology for the discovery, design and delivery of analytic and predictive services for our customers, consisting of five key steps:

Discovery

Our first step is to understand your business thoroughly during the Discovery Phase and provide concrete recommendations based on a preliminary analysis using data mining and predictive modeling techniques. The steps in this process:

• Detail business needs, objectives, and goals
• Understand the relationship between the data and the organization
• Baseline current performance
• Identify performance gaps
• Analyze a sub-set of the organizations data
• Priorities opportunities for improvement
• Identify quick wins and recommend changes
• Discuss future project feasibility and benefits
• Estimate analysis, modeling and implementation work effort
• Develop preliminary business case (ROI)
• Create Project Plan to realize benefits

Data Sourcing and Management

The sourcing of data for the analysis can be a challenging undertaking. We first study thoroughly all of the factors affecting the desired outcome. From this list of variables, we then look for the origin of the data and the feasibility of extracting the information on an on-going basis.

When the data does not exist, we look to available surrogates or put in place a process (blessed by you) to obtain the required information. We rarely come-up short in terms of getting to the data needed to proceed with the analysis. The steps in this process:

• Identify factors/variables integral to problem
• Define clear cause and affect relationships
• Acquire relevant data from internal or external sources
• Clean and enhance data when required
• Evaluate need for data warehousing or a data mart
• Initiate new data gathering steps such as surveys, market tests, etc. when needed
• Prepare data for Exploratory phase

Exploratory Analysis

Data is important to any analysis but it is usually highly unstructured and does not give obvious answers to business questions. Exploring the data in the form of “data mining” brings out the important nuggets of information related to patterns and relationship that can be used to predict outcomes and customer behavior. The steps in this process:

• Mine the data for relevant patterns and correlations
• Apply statistical procedures for testing hypothesis
• Make estimates of performance gains and probabilities of success

• Narrow the selection of statistical and mathematical techniques

Modeling and Optimization

Models are built using our Predictive Analytics Platform which contains various algorithms needed to analyze data effectively. This modeling environment enable us to perform data manipulation, data aggregation, data calculations, compare results of different models, and view the results in graphic and tabular form.

A number of models and analyses have been prebuilt to handle common business applications such as campaign and promotion response, customer segmentation, customer churn, cross and up-sell, product affinity, customer loyalty, consumer research and survey analysis, model degradation analysis, etc. Please see the section titled Solutions for more information on the subject. The steps in this process:

• Set-up modeling dataset
• Fine tune model parameters
• Select best algorithm
• Validate and test models
• Select model with strongest results and implement

Implementation

The model does not become useful until it becomes operational. We provide the assistance to help fully implement the predictive models, continue the optimization process, and monitor performance. Scored results are available in multiple formats for use in your execution engines.

• Set-up model management and scoring process
• Assist integrating predictive models into business process
• Support in-market validation
• Provide modeling and data mining guidance during training

• Establish on-going performance measurement process

The process does not end with a single model development and implementation. Our CRS LityxIQ platform enables you to refresh the model and rebuilt it automatically over time on a predetermined schedule or as new data becomes available.