Integrations

Every source, one schema.

Upload a weekly file today, or connect your sources as they open. Every source lands in the same schema and passes the same check before a model runs.

From source to model

Every path passes the channel registry and the contract check

Weekly CSVLiveAd platformsGoogle, Meta, TikTok, LinkedInCRM, commerce, paymentsSalesforce, HubSpot, ShopifyWarehouseSnowflake, BigQuery, DatabricksChannel registryOne name perchannel, every sourceContract checkEnough weeks,no gaps, real revenueYour modelFit only when thedata qualifiesGA4 · Search ConsoleLiveAnalyticsTraffic and search, read beside the model, not into it

Solid paths are live today. Dashed paths arrive with the launch connector set.

How it works

Connect, name, check. Then fit.

  1. 01 Connect or upload

    Upload weekly rows, or connect a source. Its status reflects the contract check, not the upload.

  2. 02 Name your channels

    Known names map on their own. An unknown name asks you to create or map it, so a new spelling never becomes a new channel.

  3. 03 Meet the contract

    Each requirement says whether it blocks a fit or only warns. Modeling starts when the data qualifies.

Source families

Spend on one side, revenue on the other.

The model needs both. Each family below fills one side of it.

  • Weekly CSV· live

Weekly CSV

Date, revenue, one column per channel

  • Google Ads
  • Meta Ads
  • TikTok Ads
  • LinkedIn Ads
  • Microsoft Ads
  • Amazon Ads
  • Pinterest Ads

Ad platforms

Weekly spend per channel, with the history each platform still holds

  • Salesforce
  • HubSpot
  • Shopify
  • Stripe

CRM, commerce, payments

Booked revenue by week, so the model explains the number finance reports

  • Snowflake
  • BigQuery
  • Databricks

Warehouse

A weekly table your data team already keeps

  • Google Analytics 4· live
  • Search Console· live

Google Analytics 4, Search Console

Traffic by channel group; queries, pages and devices

The contract

Checked before any model runs.

You hear what your data is missing before a fit is attempted, not after.

  1. Blocks a fit

    Under 52 weeks of history, no channels, missing or zero revenue, or a gap in the weeks.

  2. Warns you

    Under 104 weeks, a channel with under 26 weeks of spend, or a channel whose spend barely varies.

  3. One channel registry

    One canonical name per channel across every source and week. A rename relabels; it never refits.

  4. Kept apart by organization

    Row-level policies keep each organization's sources, registry and datasets separate.

Out of Bayesin

Your results, in the AI tools you already use.

Connectors bring data in. The Bayesin MCP server takes results out: connect Claude, ChatGPT or another MCP client, sign in, and read your reports there. It reads only.

Also in the product

What your data turns into.

Your first fit can start from one weekly file.

Early access is by application. We set up each team ourselves.