Data Assessment
The Edj team collaborates with key stakeholders within your organization to align on goals and desired outcomes for the engagement. Once overarching goals are established, the two organizations partner to discover relevant datasets that are available for access and determine the appropriate approach for secure data access. Edj will assess the overall quality and properties of the data with respect to formatting consistency, breadth and depth of information available, and usefulness in accomplishing the goals of the engagement. The Data Assessment Report outlines the data explored, the variables which appear to be the most complete and useful for further analysis, and opportunities for improvement related to data collection and formatting. For example, the report will outline which fields have high missingness or inconsistent formatting which need to be addressed.

Descriptive Analytics
Edj’s team of analysts will prepare the data for analysis by cleaning, transforming, joining, and processing the data as required. The Descriptive Analysis phase is typically designed to explore historical trends or discover patterns within your datasets. A variety of approaches may be used to help uncover and present these findings including segmentation, principal component analysis, and data visualization. Depending on the organizational goals, we may focus on identifying groups of individuals who have similar characteristics to one another to provide focus for future strategy or communication initiatives. Alternatively, we may be narrowing in on summarizing changes in yearly revenue by product line and searching for potential underlying causes – to name a few examples. The final deliverable from the Descriptive Analysis phase will present findings in a visually appealing way with charts that enable your organization to quickly discover key insights.
Predictive Analytics
Predictive modeling leverages patterns in historical data to discern which outcomes are most likely to occur in the future. Developing a predictive model first begins with a process called feature engineering which involves creating new variables from the existing dataset, often to summarize more granular data points into more useful information. Next, the data is split into training, testing, and validation partitions to aid in model development and final performance evaluation. Various iterations or versions of initial models are tested to see which methods offer the greatest chance of generating accurate predictions. The model performing the strongest is then improved upon to maximize accuracy, before finally being tested on unseen data to determine how well it predicts the actual outcomes observed. Once the model is finalized, the methodology, key insights into the variables which most impact predictions, and example predictions are presented in a final report. Depending on your unique needs, the final model may then be integrated into your existing workflow to help inform decision making in an ongoing basis.
