HOW WE ENGAGE / ANALYTICS METHODOLOGY

Methodology
While each client situations is different, LatentView has developed a robust, hypotheses-driven, analytic modelling methodology, building on the industry standard CRISP-DM process model. Our methodology is consistent with our guiding principles to ensure that our clients maximize their return on Analytics investments.

A typical predictive analytics-driven program is divided into four distinct phases.

1 2 3 4 20%   SOLUTION DISCOVERY   Approach / Toolkit   Output   Problem Definition – Precise definition and prioritization of the problems   Mental Models, Cognitive Maps, Analytical Hierarchy Process, Business Analysis   Prioritized list of problems                           Solution Definition – Outline the form of the solution to the problems, the solution usage scenarios, identify limitations and assumptions   Brainstorming, Impact Analysis Interactions with IT & Process owners, third party vendors   Define the form of the output from the Analytics models                           Data Selection – Identify the availability and quality of data from various internal and external data sources.   Data Quality Profiling, costs involved in procuring external data, availability at implementation, integration issues   Data Required for Modeling and ongoing usage History, Granularity, Data Quality Assumptions                           Implementation Specification – Clearly define the solution costs and expected benefits (business case), the limitations and assumptions of the solution, obtain buy in from on the solution usage scenarios, and create the detailed project plan with resources and timelines   Obtain buy-in from sponsors and problem owners   Business Case, Detailed project plan, Solution usage and limitations, Sign off

1 2 3 4 80%   SOLUTION DEVELOPMENT   Approach / Toolkit   Output   Prepare Data – transform raw data into a form suitable for exploration and modeling   Extraction, Transformation, Loading, Summarization, Treating missing values and outliers   Prepared Dataset                           Exploratory Analysis – identify trends, generate and validate hypotheses, understand key drivers   Hypotheses generation and testing, Visualization   Understand business trends, Validate hypotheses                           Build Model – choose the appropriate technique, develop initial predictive models   Sampling, Model development using data mining techniques   First cut models                           Iterative Refinement – iteratively refine the models and select final model(s)   Model stability analysis, cross-validation, instability and uncertainty   Final refined models

1 2 3 4 100%   SOLUTION IMPLEMENTATION   Approach / Toolkit   Output   Deploy Model & Ongoing Scoring – Integrate with IT systems to generate and publish ongoing scores and other business intelligence   Business Rules, IT Implementation   Deployment Plan, Integrate source code for ongoing scoring, Integrate with decisioning systems                           Monitor, Track and Periodic Refresh – develop reports to track model performance over time, across different population sub-segments, and proactively analyze factors that impact model validity   Population Stability, Cross-validation, evaluate instability and uncertainty   Model Performance Reports

1 2 3 4     DECISION MANAGEMENT   Approach / Toolkit   Output   Design Interventions – Business managers identify interventions, based on the output from the model and business objectives. If necessary, decision models could be developed to identify the optimal intervention parameters   Strategy formulation, A/B Testing / multivariate testing, experimentation   Identify specific interventions, Define intervention success criteria                           Segment & Select Customers / Accounts – Target customers for various interventions or campaigns. Create test and control groups for experimentation   Evaluate Filters, Design Test & Control groups   Test and Control Groups for targeted interventions                           Devise Offers and Creatives – Create experiments for identifying the right combination of offers, creatives for different segments of customers   Fractional factorial Analysis and Design of Experiments   Creatives, Offers                           Execute & Track Benefits – execute campaigns, track response from customers, evaluate effectiveness of interventions   Campaign execution & Tracking, Post-implementation analysis   Intervention Strategy Refinement, Feedback to model refinement