What is business intelligence (BI)?
Business intelligence (BI) is the set of practices and software used to collect data from operational systems, organize it for analysis, and present it as reports, dashboards, and queries. A BI stack usually has three layers: pipelines that copy data from the enterprise resource planning (ERP) system, the customer relationship management (CRM) system, and other sources into a data warehouse or lakehouse; a semantic layer that defines business terms such as net revenue and active customer once; and a presentation tool for building and viewing reports.
BI answers questions about what has happened and what is happening now, which places it in descriptive analytics, with diagnostic analytics close behind when users drill into why a number moved. Predictive and prescriptive work builds on the same foundation with different methods.
Why BI matters
Many companies run on spreadsheets exported by whoever knows the system best, and the numbers disagree between departments. BI replaces that with governed definitions and a single place to look. When finance, sales, and operations all read margin from the same semantic layer, the Monday meeting stops being an argument about whose number is right.
BI is also the foundation generative AI builds on. Natural language query and text-to-SQL tools work well only when the semantic layer already defines the metrics, because the model then translates a question into known terms instead of guessing at raw tables.
BI in practice
A regional hospital group runs an electronic health record system, a separate billing system, and a staffing tool. Each produces its own reports, and none agree on patient volume. The group builds a lakehouse that lands all three nightly, defines encounters, length of stay, and denial rate once in a semantic layer, and publishes a few dashboards for department heads. Denials by payer and coding reason become visible for the first time, and the revenue cycle team uses the view to prioritize appeals.
In Kirk Tech Solutions engagements, a BI foundation of this kind is usually the first phase of a data platform program, because the same governed tables later feed forecasting models and agents that answer staff questions.
What to watch for
- Fix the definitions before buying the tool, since every BI product can draw a chart and none can settle what a customer means.
- Watch for report sprawl, because hundreds of unused dashboards usually mean the semantic layer is missing and people rebuild the same logic in each one.
- Confirm who can see what, since row-level security on the data model is simpler and safer than separate reports per audience.
- Set the refresh schedule from the decision it serves, since a daily load is fine for financial reporting and useless for a dispatch board.