Three pillars. One delivered system.
TwiceData is a managed model system — not a tool you log into. Every engagement and every Quarter Stack delivery is built on the same three pillars: curated dbt model packs, lineage-aware change management, and a governed semantic layer.
Pre-built dbt model packs.
Start with curated SaaS subject areas instead of a blank dbt repo. ARR, retention, product usage, sales pipeline, revenue operations — each shipped as a versioned model pack with tests, documentation, and assumptions written down.
The packs are not a black box. They are real dbt code, ported into your warehouse, reviewed by your team. Once handed off, you own them.
What's in a pack
- Staging + intermediate + mart layers for the subject area
- Schema tests, freshness assertions, snapshot configs
- Documented assumptions (versioned, change-tracked)
- BI adapter views for Looker / Metabase / Mode / Tableau
- Pre-merge lineage gates wired into your CI
The model pack is table stakes. The knowledge graph and MCP are the product.
Every domain pack ships in three layers. A governed model pack gives you tested, lineage-tracked metrics — dbt-first, or SQLMesh / Dataform / pure SQL to match your stack. On top of it we build a domain knowledge graph that captures the entities and relationships your metrics imply. And if you want your AI agents querying it, an MCP server exposes the governed layer as typed, permissioned tools — but that's a choice, not a requirement. Turn it on or leave it off; the pack stands on its own through plain SQL and your existing BI tools. Agent tooling moves fast and that standard may shift within a quarter, so the MCP layer is deliberately a thin, swappable shim — your governed data never moves with it.
Eleven domain packs ship today — Medical, Public, Legal, Finance, Marketplaces, Medicare, Enterprise, ESG, Sales & Marketing, SaaS, and EU Medical — 400+ governed, tested models between them. Each carries its own graph schema, MCP tool set, and example models you can read before you commit.
What's in a pack
- Governed model pack — staging → intermediate → marts, fully tested, lineage-gated. The dbt foundation, portable to your framework.
- Domain knowledge graph — the entities and relationships your metrics imply (patient → encounter → claim → HCC risk; matter → timekeeper → invoice), traversable by agents and analysts.
- MCP server — typed, permissioned tools over the governed layer, so AI agents answer domain questions without raw SQL or seeing PHI they shouldn't.
- HIPAA-aware handling on the medical pack; row/column governance carried through to every MCP tool.
- Example models shipped in the open — read the SQL before you commit. No black box.
Lineage-aware change management.
Every model update is impact-scored before merge. Downstream dashboards, semantic entities, and dependent models are flagged automatically — so breakage is caught in CI, not on Monday morning when the CFO opens the wrong dashboard.
This is not a lineage diagram you look at in a slide deck. It is wired into your pull-request flow, gated, and required for merge.
What lineage gates do
- Detect impacted dashboards, models, and semantic entities per PR
- Score breaking-change risk on a 0–10 scale, threshold-blocked
- Auto-comment the PR with the impact list and reviewer suggestions
- Required-status integration with GitHub / GitLab / Bitbucket
- Audit log of every merged change + its impact score
Governed semantic layer.
Define certified metrics once and expose them to BI, reverse ETL, and notebooks with role-based access, freshness policies, and audit history built in. The same ARR number that the CFO sees on the board deck is the same number the customer team sees in Looker is the same number the reverse-ETL job pushes to Salesforce.
One definition. Many consumers. Audited every time.
What governance gives you
- Certified-entity model with versioned definitions
- Role-based access (RBAC) per entity and per attribute
- Freshness policies with automated alerting
- Full audit history for every entity read and every definition change
- Reverse-ETL contract: changes propagate or block
- BI adapter exposure: Looker LookML, Cube, Metricflow, custom
Want to see a pack in your warehouse?
30-minute architecture session. We'll match a model pack to your subject area, scope a fixed-price port, and send a proposal within 48 hours.