Plain-English definitions of the 25 business intelligence terms that come up in real conversations about reporting, from ETL and KPIs to semantic layers and source of truth. Each definition stands on its own; link to any term directly with its anchor.
Attribution
Connecting a result, usually revenue or a booking, back to the marketing channel or campaign that produced it. Platform-reported conversions and attributed revenue in your own model often disagree, and your model is the one to trust.
Business intelligence (BI)
The practice of turning data a business already collects into reports and dashboards people can act on. In practice it covers the integration, modeling, and presentation layers together, not just the charts.
Cohort analysis
Grouping records by when they started, such as customers who first purchased in a given month, and tracking each group over time. It separates genuine changes in behavior from changes in the mix of who is being counted.
Dashboard
A single screen showing the current state of a defined set of metrics, updated automatically. Distinct from a report, which is usually a point-in-time document; a dashboard is meant to be looked at repeatedly.
Data lineage
The documented path a number takes from its source system through every transformation to the figure on the dashboard. Lineage is what lets you answer “where did this come from?” without reverse-engineering the whole pipeline.
Data model
The structure that defines how your data is organized and related, and what each field means. The model is where business logic lives, and it is the layer that determines whether a dashboard is trustworthy.
Data pipeline
The automated process that moves data from source systems into your reporting layer on a schedule. Pipelines break quietly when a source system changes, which is why monitoring them matters as much as building them.
Data quality
How complete, consistent, accurate, and current your data is. Poor data quality is the most common reason BI projects overrun, because it is usually discovered during implementation rather than before it.
Data warehouse
A central database that holds data collected from your various systems, structured for analysis rather than for running the business. For an SME this is often BigQuery or Postgres, and it costs far less than people assume.
Dimension
An attribute you slice metrics by: location, product, salesperson, month, channel. Dimensions are the “by” in “revenue by region”, and choosing the right ones determines how useful a dashboard is.
Drill-down
Moving from a summary figure to the detail behind it, for example from total revenue to revenue by location to individual transactions. Without drill-down, a dashboard tells you something changed but not why.
ETL (extract, transform, load)
The process of pulling data out of source systems, cleaning and reshaping it, and loading it into a central store. ETL is typically the largest single component of BI implementation effort.
Granularity
The level of detail at which data is stored: per transaction, per day, per month. Aggregating too early is a common and irreversible mistake, because detail you discarded cannot be recovered later.
KPI (key performance indicator)
A metric chosen because it drives decisions. A useful KPI is actionable, unambiguously defined, and owned by a named person. A number failing any of those three is context, not a KPI.
Measure
A numeric value that gets aggregated, such as revenue, units sold, or hours logged. Measures are what you count; dimensions are how you slice them.
Metric definition
The written, agreed rule for how a metric is calculated, including edge cases like refunds, tax, and timing. Undefined metrics are the leading cause of two departments producing different numbers for the same thing.
Reconciliation
Systematically comparing figures across two systems to confirm they agree, and investigating where they do not. Regular reconciliation is what keeps confidence in a dashboard from eroding over time.
Refresh schedule
How often a dashboard pulls new data. The right cadence matches the decision cycle it supports; real-time refresh is valuable for operational views and largely decorative for monthly strategic ones.
ROAS (return on ad spend)
Revenue produced per dollar of advertising spend. Meaningful only when the underlying attribution is sound, which is why ROAS reported by an ad platform and ROAS in your own model frequently differ.
Scheduled report
A report generated and delivered automatically on a fixed cadence, usually by email. Scheduling is what actually removes manual reporting work, because a report that arrives gets read while a dashboard someone must remember to open does not.
Self-service BI
An arrangement where non-technical staff can answer their own questions from a governed data model, rather than queuing requests with an analyst. It works only when definitions are already agreed and enforced.
Semantic layer
A layer sitting between raw data and dashboards that translates database structures into business terms and enforces metric definitions centrally. It is how you stop the same metric being calculated five different ways.
Source of truth
The system designated as authoritative for a given piece of data. Naming one per metric, and writing it down, prevents the recurring argument about whose number is correct.
SQL
The standard language for querying relational databases. It remains the underlying skill in most BI work even when the dashboard tool hides it, and it is worth having someone on the team who can read it.
Threshold alert
An automated notification triggered when a metric crosses a defined boundary. Alerts invert the reporting model: instead of checking dashboards for problems, the problems come to you.
Where to go next
If these terms are new, start with what is business intelligence for how they fit together, then the 7-step reporting plan for the build order. When you're ready to see it working in your business, review the engagements.
