Blog · Alternatives and comparisons
Business intelligence (BI) is the practice, and the software, of turning a company's operational data into reports, dashboards and queries people can trust. This page defines it, sets out the four layers of a BI system, traces one revenue question through them with the numbers, and covers who runs BI, how it differs from analytics and data science, and where it goes wrong.
Business intelligence (BI) is the practice, and the software, of collecting a company's operational data, storing it in one place, applying shared definitions and presenting it as reports, dashboards and queries, so people can see what happened and why. BI means one agreed answer to questions like "what was revenue by region this quarter?", traceable to the rows behind it.
A BI system pulls data out of the systems that run the business, such as the ERP, the CRM and billing, puts it in one store, defines each measure once, and serves the results to the people who make decisions. The term is older than the software: H. P. Luhn used it in "A Business Intelligence System", published in the IBM Journal of Research and Development in October 1958, about routing information automatically to the people in an organization who need it. Today the word covers both the discipline and the tools.
| Layer | What it holds | Example |
|---|---|---|
| 1. Sources | The systems where transactions happen | ERP invoices, CRM opportunities, billing, spreadsheets |
| 2. Integration and warehouse | Data extracted, cleaned and loaded into one store (ETL or ELT) | A table of invoice lines, refreshed nightly |
| 3. Semantic layer | Measures and relationships defined once | Net revenue = invoiced amount excluding tax, after credit memos |
| 4. Reports and queries | What people read and ask | A regional revenue dashboard; an ad hoc query |
Sources. Each system records its own events in its own format, and none of them alone answers a cross-system question.
Integration and warehouse. ETL (extract, transform, load) cleans data before it lands; ELT loads it first and transforms it inside the warehouse. Warehouses are usually organized as fact and dimension tables. In the Kimball Group's definition, a fact table holds "the numeric measures produced by an operational measurement event in the real world", one row per event at the lowest grain: an invoice line, a shipment, a payment.
Semantic layer. This is where "revenue" stops meaning different things to different people. It holds the definition of each measure, the relationships between tables and who may see which rows. Microsoft, for example, describes Power BI semantic models as "a source of data that's ready for reporting and visualization". Keeping those definitions under version control is the subject of versioned metric definitions.
Reports and queries. Scheduled reports, interactive dashboards and ad hoc queries all read from the semantic layer, so they agree with each other. Which of the three gets used, and for what, is covered in dashboard vs report vs list.
The question: revenue by region, this quarter against last.
Rows in. The fact table holds one row per invoice line. Summed by region and quarter:
SELECT region, quarter, SUM(net_amount) AS revenue
FROM invoices
GROUP BY region, quarter
Definition applied. The semantic layer says revenue is the invoiced amount excluding tax, after credit memos, so net_amount already has tax removed and credit memos netted in.
Answer out.
| Region | Q1 revenue (USD) | Q2 revenue (USD) | Change (USD) | Growth | Share of total change |
|---|---|---|---|---|---|
| Northeast | 120,000 | 138,000 | 18,000 | 15.0% | 63.2% |
| South | 95,000 | 91,000 | -4,000 | -4.2% | -14.0% |
| Midwest | 64,000 | 80,000 | 16,000 | 25.0% | 56.1% |
| West | 151,000 | 149,500 | -1,500 | -1.0% | -5.3% |
| Total | 430,000 | 458,500 | 28,500 | 6.6% | 100.0% |
Growth = (this quarter − last quarter) / last quarter
Contribution to growth = region change / total change
Revenue rose $28,500, or 6.6%. Northeast and Midwest together added $34,000, more than all of the growth; South and West took $5,500 back.
The check. The regional changes sum to the total change: 18,000 − 4,000 + 16,000 − 1,500 = 28,500, so the contributions sum to 100%. And the Q1 and Q2 totals must match revenue in the ledger for each quarter. If they do not, the dashboard and the books disagree, and the dashboard loses.
| Main question | Typical output | |
|---|---|---|
| BI | What happened, by which slice? | Governed reports and dashboards |
| Analytics | Why did it happen? | An analysis of drivers, a variance bridge |
| Data science | What will happen, and what if? | Statistical models fitted on the data, tested for error |
| AI analytics | Can I ask in plain words? | An answer to a typed question, with an explanation |
The lines overlap, and most teams do some of each. Self-service analytics and AI analytics covers where AI analytics differs from self-service BI.
Data engineers build and maintain the pipelines into the warehouse. BI developers build the semantic layer and the core reports. Analysts answer questions with it and turn findings into recommendations. Business users read the reports and, increasingly, build their own. Self-service BI moves report building toward business users, which only works if the semantic layer holds one definition per measure: otherwise each user builds their own version of revenue.
A full BI platform is the right fit when many people need governed, refreshed reports from many systems, and someone will own the warehouse and the semantic layer. For a team that needs answers to its own questions from files it already exports, there are lighter routes, set out in BI tool alternatives for teams without a data team.
Covirage skips the warehouse build for a team's own questions: it takes the exported files, applies one versioned definition per measure, and its tools compute the answer with the rows behind it; the external AI model explains it and never does the arithmetic. See how Covirage compares with BI platforms, and when each is the better fit. To compare the platforms themselves, see business intelligence tools, and for the cube model behind fast slicing, see OLAP.
Product names are trademarks of their owners; Covirage is not affiliated with them. Facts about Microsoft products were checked on October 1, 2026 from Microsoft's own pages.
A monthly sales dashboard that pulls invoices from the ERP, applies one definition of net revenue and shows revenue by region, product and customer against budget, with the ability to drill into the rows behind any figure.
BI usually describes reporting on what happened, from governed data. Business analytics is the wider term and includes explaining why and predicting what next. In practice most tools and teams do some of both.
Excel can do BI work: Power Query connects and cleans data, the Data Model holds relationships between tables and PivotTables report on them. Dedicated BI platforms add central governance, scheduled refresh and access control for many users at once.
Builds and maintains reports and dashboards, writes queries, defines measures with the business, checks that figures reconcile, and answers ad hoc questions. The role sits between data engineering and the people who use the numbers.