Design & Test Concepts – POC Concepts
Though every analytical problem is different, and each is addressed within every businesses’ unique circumstance, having a roadmap of tried-and-true steps to follow allows us to focus on solving the problem and not getting lost in the noise surrounding it.
Over 10 years, we have developed many tips-&-tricks, best practices and ways of addressing each step so that we move through the problem towards the analytic outcome as quickly and as effectively as possible.
For the First Analytics team, these steps and methods have become second nature, yet we review each analytical problem individually and use these steps as guideposts, which is especially helpful when working jointly with your analytic teams.
Frame the Problem
What are the objectives?
What decisions will the analysis support?
What is the impact of each decision?
What is the process flow?
What non-controllable factors affect the outcome?
What are the constraints?
What data could be available?
Assess the Data
How suitable are individual data sources?
We use proprietary tools to facilitate initial review and quality data checks.
Build Core ABT
Align the data and bring it all together.
We use proprietary tools to build the Analytics Base Table or ABT – the data foundation of the project.
Explore relationships among previously disconnected data elements.
In the process, assess strengths and weaknesses of the combined dataset. Begin to form ideas on approaches.
Identify alternative approaches and determine which to pursue.
Draws on the full range of skills, experience, and training of the entire FA team.
Build Modeling Datasets
Create approach-specific datasets with business-problem-specific feature engineering.
Draws from ABT.
Explore the new modelling datasets.
Leverage data visualization tools for insights into relationships that might not be apparent from tables of numbers or model results.
Do the Analysis
Execute the analysis plan.
This begins an iterative process. Typically, initial findings lead to more discussion with business stakeholders and additional data feature engineering,
Summary of findings
Go / No-go recommendation. If “Go”, specific design recommendation