What is Business Intelligence?

Business intelligence (BI) is the set of technologies, processes, and practices for collecting, integrating, and presenting an organisation's operational data so that people can monitor performance and make decisions: data pipelines that pull from source systems into a warehouse, models that reconcile definitions, and the reporting layer (dashboards, scheduled reports, self-service exploration) where the numbers meet their readers. Its mode is primarily descriptive: what happened, how much, where, compared with when. It is retrospective by nature, organisation-wide in scope, and dependent on the questions that were anticipated when the data was modelled; a BI system can slice revenue by region because someone decided region mattered, and it cannot say why customers in that region churned because nobody recorded a reason.

BI and Research

The two disciplines answer different question types with different evidence. BI reports on the recorded past across the whole organisation: descriptive, precise on counts, silent on motivation, limited to fields that exist. Research investigates specific questions with purpose-built evidence: collected for the question, able to ask why, able to test causes, and limited to its sample and scope. BI is where product questions usually begin (a metric moved; a segment underperforms) and research is where they get answered (interviews with the segment, a study of the flow, an experiment on the fix). The failure modes are mirror images: organisations that treat dashboards as understanding decide on correlations and counts with no mechanism; organisations that ignore their BI run studies to discover what the warehouse already knew. The internal data is also research's best secondary source and its most reliable sampling frame: BI defines the population and finds the segment, research recruits from it and asks.

Reading BI Like a Researcher

1. Interrogate the definitions.
Every dashboard figure rests on a definition (what counts as "active", which events form a "conversion", how time zones and refunds are treated) that usually lives in a model file few readers have seen. Definition changes produce false trends; disputed definitions produce meetings. Treat the metric dictionary as the instrument documentation it is.

2. Ask what the data can't see.
Untracked behaviour, unconsenting users, offline steps, and reasons of any kind are absent from the warehouse; findings describe the recorded, the coverage problem in its most invisible form.

3. Hold BI correlations to the same standard as any others.
Segments differ before they differ on the metric; "accounts with integrations retain better" is a correlation with the usual four explanations.

4. Show uncertainty where it exists.
BI numbers look exact because they're counts, but counts of small segments, short windows, and modelled estimates carry real variability; a small cell's percentage deserves the same scepticism as a survey's.

5. Turn anomalies into studies.
The dashboard's job is to point; when it does, size the question, recruit from the affected segment (the warehouse can name them), and run the research that explains the number, quickly enough that the decision waits for it. A study that collects task performance and recorded answers from precisely the users a BI query identified (a Ballpark study fielded to a customer list) is the natural continuation of a dashboard anomaly.

The Takeaway

Business intelligence is the organisation's recorded past, integrated and presented: descriptive, comprehensive, and confined to what was anticipated. Read its definitions, know its blind spots, keep its correlations in their place, respect the uncertainty its exactness hides, and let its anomalies commission the research that supplies the why. The dashboard is the best place to find a question and the worst place to stop.