Analytics Governance – Business Intelligence

Analytics Governance

We explained and discussed data governance in Chapter 6, “Data Democratization, Governance & Security,” in detail.

However, as we discussed about data governance, people sometimes get confused that it deals more with enterprise data management governance: policy and procedures for analytics pipelines governance. As it is similar to data stewardship, analytical stewardship need to be clearly specified. Policies and procedures for data practice are related to the data analytics/BI life cycle.

Data Analytics Life Cycle

There are six phases of data analytics that must be followed in data science projects. The framework for data science projects is simple and cyclical in nature, both backward and forward, but has to be one step after another. The steps are as follows:

1. Data discovery and formation: It starts with defining a goal, objective, and benefits the organization wants to achieve. Next, it evaluates and assesses required data and comes up with high level goal with objectives. The next step is to create an evaluation and assessment of the data. Create a basic hypothesis related to the objective.

This stage consists of mapping out the potential use and requirements of data: what data is required, where to find the required data, and how to get this data.

Mandatory activity in this phase is structuring the business problem in the form of an analytics goal and formulating the initial hypothesis to test and start learning from the data.

2. Data preparation and processing: This stage consists of anything related to storing and processing the data.

Data is ingested from external sources/internal systems and sources. After ingestion, data is prepared and transformed. Sample data is prepared to using business logic. This phase takes the longest time in the life cycle to make sure that data requirements match with business requirements.

3. Designing a model: After mapping out the business goals, collect structured, semi-­structured, or unstructured data. These steps include the team to determine the best methods, techniques, and workflow to build the model in the subsequent phase. Model building initiates with identifying the relationship between data points to select the key variables and eventually find a suitable mode.

Datasets are developed by the team to test, train, and produce the data. In later phases, the team builds and executes the models that were created in the model planning stage.

4. Model building: In this step, after designing a model, we develop, test, and train a dataset. Experts build the model that was built in step three. Experts use different modeling techniques using coding or user interface tools. Some of the examples of algorithms are linear regression analysis, logistics regression, neural networks, etc., as per requirements for building and executing the model. This step may have multiple iterations within and makes sure that a model is tested, fit for purpose, finalized, and ready for production.

5. Communicating and publishing: After model is ready for production, communication and collaboration are started regarding the success or failure as per predefined criteria. Business value is re-evaluated.

6. Measuring effectiveness: The data is moved to a live environment from the sandbox and monitored to observe if the results match the expected business goal defined in previous stage. If outcome deviates from the goal set out in first phase, you can move backward in the data analytics lifecycle.

BI and Data Science Together

The implementation of data warehousing (DW) and business intelligence (BI) has eleven guiding principles, as follows:

 1.  Executive commitment, sponsorship, and support is paramount.

2. Secure business SME (subject matter expert) support in understanding, correcting, and user acceptance testing UAT) of data, and validation of KPIs.

 3.  Priorities to be driven by business.

4. Maintaining high-data quality is mandatory and critical to ongoing business intelligence and data science program success.

 5.  Build and show incremental value.

6. Customize according to business domain requirement. One size does not fit all.

7. Provide transparency and self-service along with context to provide value and satisfaction to business.

8. Design and architect with global standards, but build with local business rules in mind.

9. Collaborate and integrate with other data initiatives across business for synergy—and avoid duplication.

10.  Start with a clear objective and goals in mind.

11. Summarize and optimize in the end. Build on the granular data and add aggregates of summaries needed for performance.

Companies regularly interact with customers; develop and design products; and create strategic plans and make decisions according to market forces. Modern data warehouses have modern expectations, like being well governed, consistent, and fast in delivery of quality data. Value can be brought to organizations through a well-governed strategy for getting required data from customer, supplier, partner, application, and systems.