By Rachel Stewart – Before you start answering questions, you need to know exactly what you’re trying to find out. The starting point of most any dashboarding project should be a whiteboarding session with the end-users; the dashboard becomes a collection of visuals that hold the ability to answer their questions.

For every single visual you create, make sure you’re answering a specific question. Each graph needs to be intentional and purposeful, and it’s very important to have your KPIs clearly defined well before you start building. If you don’t include your stakeholders from the very beginning, you’ll almost certainly have a lot more reworking to do after initial production is complete.

Generating meaningful visualizations is nearly impossible without a good data foundation. Unclean data means holes and problems will need to be patched and fixed further down the pipeline. Many BI tools have functions that can format/prepare your data and generate some level of relational modeling for building your visualizations. However, too much modeling and logic in the tool itself will lead to large performance issues, and most BI tools aren’t specifically built with data wrangling in mind. A well-modeled semantic layer in a separate tool that handles all the necessary business logic is often essential for performance and governance.

The semantic layer is the step in preparation where the business logic is performed, joins are defined, and data is formatted from its raw form so it’s understandable and logical for users going forward. For Power BI users, for example, you would likely generate tabular models within SSAS. With a strong semantic layer in place before you even get to the BI tool, there will be little to no data management to be done in the tool itself. This means there is less processing the BI tool needs to handle and a much cleaner governance system. Read On: