Skip to content

Si: AI agent for BigQuery

A team that has gathered its data in BigQuery still needs a way for people who don't write SQL to get numbers. Connect BigQuery and DataxLand loads the tables you choose (full or incremental); Si proposes how to read them and maps them to the standard model, you approve, then the whole team asks Si about net revenue, orders, contribution margin, ad spend and blended ROAS. Every answer shares its metric definitions with dashboards and reports and carries a confirmation level.

Free plan with no time limit · No card required · Sample workspace ready in minutes

Ask Si about BigQuery

Example questions Si answers from the metrics below, once the source is connected and the model approved.

  • “What is net revenue by channel this month compared with last month?”
  • “Show blended ROAS (net revenue over ad spend) by week this quarter.”
  • “What was contribution margin by product last month?”
  • “How much did ad spend grow this month versus last month?”
  • “What is average order value by month this year?”
  • “Why did blended ROAS drop last week?”

Si answers with these metrics

MetricDefinitionUnit
Net revenueLine value after line discounts and allocated order vouchers, minus refunds to buyers. Refunds are deducted only here.Currency
Valid ordersOrders that were not cancelled (standard status other than cancelled).Count
Average order valueLine revenue after discounts divided by valid orders.= line_net_total / Valid ordersCurrency
Contribution marginNet revenue minus COGS, marketplace fees, payment fees, shipping and return costs (before advertising).= Net revenue - Cost of goods sold - Marketplace fees - Payment fees - Seller-paid shipping cost - Return costCurrency
Blended ROASActual net revenue divided by ad spend for the same channel and period.= Net revenue / Ad spendNumber
Ad spendTotal advertising cost by report date. Filtering by channel or platform uses the destination channel the ads drive traffic to.Currency

How Si answers

A semantic layer, no free-form SQL

Si picks metrics, dimensions, filters and periods; DataxLand builds the query from versioned definitions; Si runs no SQL of its own when answering.

A confidence level on every number

Standard (pack metric), Approved (workspace metric checked before use), Unconfirmed (exploratory answer, clearly labelled).

One tool set everywhere

In the app, Slack, Telegram, Microsoft Teams or MCP, Si uses the same read-only tools and your permissions in the workspace.

Where to ask Si

In the app

Ask inside the workspace in the language you use; Si answers in the same language, with the metric source and confidence level.

Slack, Telegram and Teams

Link your account with a one-time code; ask in a private chat or a group bound to the workspace.

MCP

Create an expiring, revocable token (plans with MCP) so Claude, Cursor or your own agent call the same tools.

Get started

  1. 1

    Connect BigQuery

    Choose Add source → BigQuery, enter the project, dataset and a read-only service account key, then pick the tables to load.

  2. 2

    Review Si's proposal

    Si proposes how to read your tables and maps them to the pack's standard model; you fix the highlighted fields, answer the multiple-choice questions and approve the build.

  3. 3

    Ask Si

    Ask in the app, connect Slack or Telegram with a one-time code, or create an MCP token for Claude, Cursor or your own agent.

Frequently asked questions

Do questions run up BigQuery query costs?

Not while you ask: Si reads the workspace's built warehouse. BigQuery is read only during loads on the refresh schedule.

Does Si change anything in BigQuery?

No. Si only reads the workspace's approved metrics; there is no tool that writes to BigQuery, and Si never runs SQL it wrote itself on your data when answering.

Can I use Si with Claude or Cursor?

Yes, through the DataxLand MCP server with a token that expires and can be revoked. MCP tokens come with the Small Business plan and above; on Starter you ask Si in the app, Slack and Telegram.

How do I know Si's numbers are reliable?

Every answer names the metric it used and its confirmation level: Standard for pack metrics, Approved for workspace metrics that passed their checks, Unconfirmed for exploratory answers.

Dashboard templates that use this source

Connect BigQuery →BigQuery + Google Ads →Use Si over MCP →

Ask Si about BigQuery today

Free plan with no time limit · No card required · Sample workspace ready in minutes