Skip to main content

Business context

A business context is a named block of Markdown describing how your company talks about its data — definitions, rules, conventions, caveats. It is attached to one or more data sources, and Fanar pulls in the relevant blocks when answering a question against those sources.

Only admins can create or edit business contexts.

Creating one​

  1. Open Settings → Business Context.
  2. Click New, give it a name, and pick the data sources it applies to.
  3. Write the context in Markdown and save.

The name matters — it's shown to the model alongside the content, so "Revenue definitions" beats "Notes 3".

What to write​

Write the things a new analyst would get wrong in their first month.

## Revenue

"Revenue" always means **recognised** revenue, net of tax and refunds.
Use `orders.net_amount`, never `orders.gross_amount`.

Bookings (signed contract value) is a different number. If someone asks for
bookings, use `contracts.tcv` — do not substitute revenue.

## Fiscal calendar

Our fiscal year starts on 1 February. FY26 Q1 is Feb–Apr 2025.
Calendar quarters are almost never what people mean internally.

## Exclusions

Exclude rows where `orders.is_internal = true` from all customer-facing metrics.
These are staff test orders.

Guidelines that work:

  • Be prescriptive, not descriptive. "Use net_amount" is more useful than "there are two amount columns".
  • Say what not to do. Naming the common mistake prevents it.
  • One topic per context block. Blocks are retrieved by relevance, so a block that mixes revenue rules with churn rules gets pulled in for both and wastes room on the half that doesn't apply.
  • Keep it current. Stale context is worse than none — it will be followed confidently.

How blocks are chosen​

When you ask a question, Fanar embeds it and retrieves the closest-matching context blocks for that data source — by default the top 5. Everything else is left out.

This is why several focused blocks beat one long document: a 4,000-word "company handbook" block will either be retrieved in full for every question or not at all.

If the instance has no embedding key configured, Fanar includes all context for the data source instead. That still works, but the more you write, the more diluted each answer's context becomes.