Reporting the difference between a modelled attribution and what Meta or Google claims for itself is the number most marketing stacks avoid producing, because it is uncomfortable. Having it as a callable primitive changes the conversation from 'which dashboard do we believe' to a figure. The pnl namespace is the other serious part — contribution margin net of shipping, returns and discounts is the only revenue number worth optimising against, and almost nothing computes it per SKU.
A marketing data layer exposed as MCP primitives across six namespaces: attribution, pnl, cohorts, journeys, benchmark and agents. The endpoint is per-account, so the URL comes from their setup.
- Attribution — multi-touch and incrementality models, plus the delta against what each platform claims for itself
- Pnl — contribution margin by channel, by SKU and by cohort, including shipping, returns and discounts
- Cohorts — acquisition cohorts, RFM segmentation, LTV and repeat curves
- Journeys — the full customer path across ads, Shopify, third-party checkouts, shipping and post-purchase surveys
- Benchmark — CAC, LTV and repeat rate against industry and cohort
- Agents — composable ones for budget reallocation, anomaly watching and audience building
Their onboarding, since the endpoint is per-account. On privacy, the architecture is described as keeping customer PII behind a boundary by default rather than as a switch you turn on.