Data Governance with dbt: Feedback from the KPC Webinar

Webinar Wrap-up: Data Governance in the Modern Data Stack, End-to-End Lineage, Data Contracts, and KPC's Practical Feedback.

By 2026, data teams have industrialized their pipelines. Yet, governance remains a separate project, disconnected from transformation tools — resulting in dead documentation, unassured quality, and lack of traceability. This is the observation made by Matthieu Augier (KPC) and Hicham Babahmed (dbt Labs) during their joint webinar on May 26, 2026.

A problem still all too common in 2026​

Whether it is an SME with data scattered across Excel files, or a large corporation with a dense IT system and conflicting definitions between divisions, the observation is identical: trust in data is lacking. AI projects fail due to a lack of controlled semantics, strategic decisions rely on unreconciled figures, and regulatory requirements (GDPR, Solvency II, PII) remain difficult to address without traceability.

On the ground, the same initiatives come up mission after mission: clarifying roles and responsibilities (data owner, data steward, data custodian), implementing quality pragmatically, building a living catalog, and addressing regulatory auditability.

• Untrustworthy data: The question "what is quality data?" remains without an operational answer in most organizations.​

• Under-controlled regulation​: GDPR, Solvency II, PII: constraints are tightening without traceability being in place.​

• AI failing fast​: AI projects lead to a "fail fast" when data semantics have not been stabilized beforehand.

KPC's Approach: The Data Governance Office: a system within the system​

At KPC, the response to these initiatives involves building a Data Governance Office — a dedicated unit that coordinates roles, actors, quality, cataloging, lineage, and compliance. Three organizational models coexist:

• Centralized​: A single DGO unit feeds all divisions. Single vision, consistent governance.
  (The most frequent)

• Federated: Shared foundation + autonomy by vertical branch — balancing central consistency and local agility by domain.​

• Decentralized​: Independent flows by domain. Requires an excellent data culture. Risk of duplication and diverging definitions without a strong DGO. 

"The idea is to capitalize on documentation once, share it with everyone who produces or consumes the data, and make them as autonomous as possible."

Matthieu Augier - Director of Data Governance, KPC.

KPC x dbt Labs Webinar

KPC x dbt Labs Webinar

In each of these models, data players — data owners, stewards, custodians, engineers — manage distinct scopes, but run into the same friction point: without shared and living documentation, everyone ends up rebuilding their own truth. This is precisely where transformation tooling changes the game.

How dbt embeds governance within pipelines

In practical terms, each governance pillar is implemented using native features of dbt Platform — without external tooling. The live demonstration detailed it pillar by pillar.

1. End-to-end lineage and column-level traceability

The dbt DAG (Directed Acyclic Graph) provides a complete visual map: from raw sources down to final usages (Power BI dashboards, Tableau, AI models).

In a multi-project context with central, finance, and marketing teams, each team only sees public models — models not declared public are automatically protected. The lineage goes down to the column level, making it possible to know precisely how each field has evolved through transformations.

"One morning, your dashboard no longer displays the expected data. With dbt Catalog, you go back to the model, check the status of the last orchestration, the test results, and the lineage — without involving the data engineer." 

Hicham Babahmed - Partner Solutions Architect, dbt Labs.

KPC x dbt Labs Webinar

2. Data contracts: governance before production

Defined in the dbt project's YAML files, data contracts run even before the model is created — they block the production release of any data that would violate the contract (incorrect column type, out-of-range value, too long IBAN, etc.). The difference with data tests is structural: the contract sets the boundaries decided beforehand with the business; the test verifies what is observed during execution.

Data contracts are shareable and replicable: a contract validated on one object can be duplicated and adapted to a similar object, accelerating quality deployment at scale. This is particularly powerful in a federated model where multiple teams feed the same data platform.

3. Catalog, metadata, and semantic layer

dbt Catalog centralizes descriptions, types, tests, orchestration status, and recommendations on each model. The semantic layer allows defining business KPIs reusable by all teams — and queryable in natural language via connected LLMs.

Role-based access control allows mastering who consumes the semantic layer, particularly for compute-heavy usages.

"More and more, business profiles are coming to dbt Platform — not to write code, but to verify that the data they will present in meetings is indeed correct."

Hicham Babahmed - Partner Solutions Architect, dbt Labs.

KPC x dbt Labs Webinar

KPC's Perspective: Reducing friction between data and business teams

In KPC's missions, the contribution of dbt comes down to a precise point: the friction between data and business teams. For years, this was due to a communication gap: business requirements were not formalized, the data produced did not meet expectations, and no one had a common source of truth.

dbt then acts as a translator between the two worlds — which is precisely what the platform unlocks. Business users can write their requirements directly in the platform, visualize what the models produce, and challenge data engineers on quality — without intermediaries. Subject matter experts are no longer oversolicited: they document once, and the platform answers continuously.

Another highlighted operational benefit: dbt works as a cross-platform layer across multiple platforms (Snowflake, Databricks, BigQuery, Redshift). In companies that have adopted multiple data clouds, it unifies governance without forcing a migration.

« The Data Governance Office is a system within the system: it coordinates roles and actors — technical, business, or hybrid: data owner, data steward, data custodian, data engineer. And it approaches quality first as a matter of processes, not just tools. » 

Matthieu Augier - Director of Data Governance, KPC.

KPC x dbt Labs Webinar

Key Takeaways from This Webinar

• Effective governance does not live in a disconnected catalog tool — it lives inside the dbt pipelines themselves.

• Automatic lineage + integrated documentation + data contracts = the three pillars for traceable and trustworthy data, right from production.

• Data contracts and data tests are complementary but distinct: the contract sets the boundaries decided with the business beforehand; the test observes what happens during execution.

• dbt adapts to all three organizational models (centralized, federated, decentralized) — the most frequently encountered in client contexts remains the centralized one. The dbt platform acts as a bridge between business and data — subject matter experts document once, and the platform answers continuously.