How the four disciplines differ
| Dimension | Question | Evidence standard |
|---|---|---|
| Reporting | What happened, from available data. | Data availability and consistent extraction. |
| Business intelligence | What a measure means, and how it is presented for a decision. | Agreed definition, one owner and a governed model. |
| Analytics | Why the measure moved, and where the difference sits. | Reconciliation, segmentation and stated assumptions. |
| Data science | What is likely next, or which case deserves attention. | Out-of-sample evaluation, calibration and monitoring. |
What business intelligence actually delivers
BI connects governed data to a decision: a defined measure, a model that holds it, an interface someone opens at a known moment, and an owner who answers for its accuracy. When those four exist, a dashboard becomes part of how work happens. When they do not, it becomes another artefact that decays quietly.
Reporting, BI, analytics and data science
Reporting presents what happened from available data. BI governs the definitions behind that presentation. Analytics investigates why a measure moved. Data science estimates what may happen or which case deserves attention. They are different jobs with different evidence standards, and conflating them is why dashboards get blamed for modelling problems.
The semantic layer is the product
Underneath the visuals sits a model that defines measures, relationships, hierarchies and time logic. That model is the durable part: reports are disposable, definitions are not. Certification, review and a named owner keep it trustworthy as teams and questions change, and they are what stop a new report from re-deriving a metric.
Concretely, the model holds tables at a stated grain, the relationships between them, measures such as net revenue with their filters, a date table carrying the fiscal calendar, and role-based access rules. A report built on it inherits all of that, which is why two reports on the same model cannot disagree about revenue.
- One certified definition per business measure
- Changes reviewed and released like any other change
- Access tested with the identities real users hold
What a usable decision interface needs
A report earns its place when it answers a specific question for a specific audience at a specific cadence. That means showing the definition, the refresh state and the drill-through path to context, and being honest about what it does not cover. A visual without an action attached is decoration.
Self-service with a governed boundary
Self-service is valuable and unmanageable without rules. Decide which datasets are certified, who may publish to which audience, how a new measure is proposed and approved, and what happens when an uncertified report contradicts a certified one. The boundary is what makes self-service safe to encourage.
Illustrative example · hypothetical organisation
Worked example: from a disputed number to a decision view
Take a hypothetical operations team whose weekly meeting starts with an argument about on-time delivery.
Two spreadsheets disagree because one measures against the promised date and the other against the requested date. The BI work starts by agreeing the definition: delivered on or before the date promised at order confirmation, measured per order line, excluding cancelled lines, owned by the head of operations.
The measure is built once in the semantic model. The weekly view shows it by warehouse and carrier with a drill-through to the late lines, and the refresh time sits on the page. The meeting now starts from the late lines, not from the definition.
How BI programmes fail
Three patterns recur: definitions diffuse across the estate until nobody trusts the number, adoption is assumed rather than designed into a workflow, and platform migration happens before the model boundary is understood. Each is avoidable by naming the decision, the owner and the definition before building anything.
Questions and trade-offs
Question
How is BI different from reporting?
Reporting shows what happened using whatever definitions are available. BI is the layer that decides what the measures mean, models them once, and governs who may change them. Without that layer, more reports simply produce more versions of the same number.
Question
Should BI sit inside the data platform or the BI tool?
Wherever it can be owned, reviewed and reused. Definitions held in the platform are inherited by every consumer, which makes the BI tool a presentation choice. Definitions held only in the tool still work, but they make that tool your semantic layer and make any future migration a governance project.
Question
What skills does a BI team need?
Data modelling and SQL to build governed models, a modelling language such as DAX or LookML for measures, enough business knowledge to write definitions people accept, and report design for a specific audience. The scarcest of these is the definition work, because it needs authority as well as technique.
