Data layer
Foundat10n
A spec of ten pass or fail requirements for the go-to-market warehouse. The audit scores your stack against it and gives you the fix list in order.
Most GTM reporting problems trace back to the data layer: history that was overwritten, columns that were dropped, two definitions of ARR. Foundat10n names ten things the warehouse has to do and checks each one.
At a glance
- Covers
- CRM, billing, product usage and the warehouse
- Requirements
- 10, each pass or fail
- Entry point
- Fixed-scope audit
- Audit output
- Scorecard and ranked fix list
- Build
- In your Snowflake or Databricks account
- Hand-off
- dbt project, pipelines and docs
The ten requirements
Each requirement passes or fails. There is no partial credit. Below is what an audit scorecard looks like, filled in with sample data.
- All GTM stack data lands in the warehouseEvery system in the GTM tech stack (CRM, marketing automation, conversation intelligence, dialer, ERP billing) MUST replicate to the data lake on a managed, monitored cadence.Pass
- Raw layers persist every source columnRaw and base layers MUST persist all columns from source. Column selection is a downstream modeling decision, never an ingestion decision.Pass
- The warehouse is the single source of truthSource systems produce data. Consumption tools display data. The warehouse MUST be the only place truth is computed and governed.Fail
- Every non-audit table gets SCD2, plus a "newest" viewAll non-audit source tables MUST be tracked as Type-2 slowly changing dimensions with indefinite retention, paired with a current-state ("newest") view for query ergonomics and agent consumption.Fail
- SCD2 tracks stored state, not computed stateSCD2 change detection MUST operate on stored fields only. Computed fields (Salesforce formulas, roll-ups, vendor-derived scores) are reconstructed in the warehouse from their tracked stored inputs.Fail
- History starts todayHistory capture MUST begin before modeling perfection. Coverage can always expand forward; history that wasn't captured can never be recovered.Fail
- Field names don't change until goldSource field names MUST pass unchanged through raw, base, and intermediate layers. Renaming is exclusively a gold/presentation-layer concern.Pass
- Conformed calendar and fiscal dimensions are non-negotiableA single governed date dimension (calendar and fiscal) MUST exist in the warehouse and be the only date spine any model joins to.Fail
- The gold layer extends to reporting assets, not just source assetsThe gold layer MUST NOT stop at presentation-grade versions of source objects. It MUST include the standard reporting anatomy every B2B SaaS warehouse needs: modeled funnel stages, pipeline point-in-time snapshots, labeled outcome tables (MECE win/loss/slip classification), conformed account & rep dimensions, activity rollups, and the reporting assets built on them.Fail
- The BI layer is a rendering engine, not a logic engineData models MUST extend all the way to final reporting assets. No metric logic, joins, or business rules may live in BI tools, spreadsheet formulas, or dashboard calculated fields.Pass
How an engagement runs
Three steps. The audit stands on its own, and you decide on the build once you have the scorecard.
Audit
Score your CRM, billing and warehouse against the ten requirements.
The scorecard and a ranked fix list, in the order the fixes pay off.
Build
Daily snapshot history, a medallion warehouse and one set of revenue definitions, in your own Snowflake or Databricks account.
Each failed requirement rebuilt until it passes, with tests in place.
Hand off
Your team owns the dbt project, the pipelines and the docs.
A warehouse your analysts and AI agents can both rely on, run by your team.
Start with the audit
A fixed-scope audit scores your stack against the ten requirements and gives you a ranked fix list. The first step is a call.