Business Definitions — the Certified AI Semantic Layer
Business Definitions turn the AI from a smart guesser over raw tables into an analyst that uses your company's official formulas. Define what Revenue, Active Customer, or Margin means once per workspace, certify it, and every AI answer — in chat, dashboards, auto-generated reports, transforms, and executive summaries — computes it exactly the same way.
Why it matters: without definitions, asking "what was our revenue last month?" forces the AI to guess which table, which column, and which rows count. Two people can get two different numbers. With a certified definition, everyone gets one number, from one approved formula, every time.
What is a definition?
Each definition is a business term with:
| Field | Example |
|---|---|
| Kind | Metric (an aggregated number) or Dimension (a grouping/label) |
| Name | Revenue |
| Meaning | Paid invoices excluding refunds |
| Formula | SUM(Invoices.net_total) WHERE Invoices.status = 'paid' AND type <> 'refund' |
| Synonyms | sales, income, المبيعات |
| Certified | ✓ — the AI must use this formula and may not improvise |
Synonyms work in any language — if your columns or your users speak Arabic, add Arabic synonyms and the AI resolves them to the same certified formula.
Where to manage definitions
Open your workspace → Settings → Definitions. Workspace Admins (and the workspace owner or an Org Admin) can create and edit; all other members see the dictionary read-only.
Three ways to create definitions — no SQL or DAX required
1. AI Suggest (recommended)
- Pick a datasource — or leave All datasources to sweep the whole workspace.
- Optionally pick a single table to go one table at a time.
- Click AI Suggest. The AI reads the real schema and drafts 4–8 definitions with correct table/column names, plain-English meanings, and synonyms (including Arabic when your data is Arabic).
- Review each draft — Add the correct ones (they save as certified), edit before adding, or ignore. Add all approves the whole batch.
Suggestions never duplicate what you've already defined, so you can re-run per table until the dictionary covers everything.
2. Write the formula for me
Click Add manually, type the name and describe the term in your own words — e.g. "total paid invoices excluding refunds" — then click ✨ Write the formula for me. The AI writes the formula against your actual schema:
- SQL datasources get a SQL aggregate expression.
- Power BI (XMLA) datasources get a DAX measure expression — you never need to know DAX.
It also explains in one sentence what the formula does, so you can sanity-check before saving.
3. Fully manual
Power users can type the formula directly.
Certification and the Certified AI badge
Marking a definition Certified makes it binding: whenever a question mentions the term (or a synonym, in any language), the AI must compute it with the approved formula — it is explicitly forbidden from improvising a different column or condition.
Once a workspace has at least one certified definition, its datasources show a green ✓ Certified AI badge in the workspace lineage and list views — a visible signal that answers over this data are governed.
Where definitions take effect
Definitions apply automatically across every AI surface in the workspace:
- Ask AI chat and the dashboard chat panel — generated SQL uses certified formulas.
- AI Auto-Report — generated charts compute defined terms correctly.
- Transform Workbench AI — "add a Margin column" derives it from the approved formula.
- Chart insights, anomaly explanations, and executive summaries — the AI interprets and describes terms with your official meanings.
Users don't do anything differently — they just ask questions and get consistent, certified numbers back.
Availability
- Cloud: authoring definitions (AI Suggest, formula assistant, add/edit) is included in Premium, and in the free first report. Reading and AI enforcement are available to every user in the workspace — once defined, everyone benefits.
- On-Prem: included for all users, no additional licensing.
Best practices
- Start with your top 5 KPIs — revenue, orders, active customers, churn, margin. That covers most questions.
- Go table by table with AI Suggest — it keeps each review small and focused.
- Certify only what finance/ops has approved. Uncertified definitions still guide the AI but signal "draft".
- Add synonyms generously — every word your team actually says, in every language they say it.
- Review after schema changes — if a column is renamed, update the affected formulas (the editor warns you which terms exist).