{"id":1400247,"date":"2026-07-27T17:18:51","date_gmt":"2026-07-27T16:18:51","guid":{"rendered":"https:\/\/www.artefact.com\/?post_type=blog&p=1400247"},"modified":"2026-07-27T17:18:51","modified_gmt":"2026-07-27T16:18:51","slug":"data-quality-by-design-dbt","status":"publish","type":"blog","link":"https:\/\/www.artefact.com\/de\/blog\/data-quality-by-design-dbt\/","title":{"rendered":"Data \u201eQuality by Design\u201c: Warum sich Ihre Data-Governance mit den Ursachen und nicht mit den Symptomen befassen muss"},"content":{"rendered":"
\n

Co-hosted by Artefact<\/a> and dbt Labs<\/a>, the webinar “Data Quality by Design: Treating Causes, Not Symptoms<\/em>” aimed to address a critical strategic alignment that remains underutilized across enterprises: the convergence of Data Governance and Data Engineering.<\/p>\n

Florence B\u00e9n\u00e9zit<\/a>, Partner & Global Manufacturing Lead at Artefact, and Eric Ortiz<\/a>, Partner Development Manager EMEA at dbt Labs, presented a comprehensive methodological framework for embedding data quality directly into the architecture and design phases of data products\u2014a foundational approach to delivering measurable, long-term ROI.<\/p>\n<\/div>\n

Why Reactive Quality Management Fails<\/h2>\n

In most enterprise environments, data quality is handled reactively: errors are identified downstream during consumption\u2014such as in dashboards or executive BI reports\u2014and patched on a case-by-case basis at the end of the pipeline.
\nThis approach suffers from three major drawbacks:<\/p>\n

    \n
  • Significant Operational Overhead<\/strong>: Data teams spend the majority of their time on corrective maintenance and manual data reconciliation.<\/li>\n
  • Erosion of Business Trust<\/strong>: Contradictory or incomplete data hinders analytics adoption and executive decision-making.<\/li>\n
  • A Barrier to Scalability<\/strong>: The accumulation of ad-hoc hotfixes complicates pipelines and inflates technical debt.<\/li>\n<\/ul>\n

    Data quality cannot be relegated to post-hoc cleanup or treated as a purely documentation-driven exercise within a data catalog. To be effective, quality must be engineered and executed “by design” within the codebase and architecture, spanning from ingestion to consumption.<\/p>\n

    The 6-Step Roadmap: From Source to Consumer<\/h2>\n

    To resolve root causes rather than symptoms, Artefact structures quality management across six links of the data value chain:<\/p>\n

    [ A: Source Quality ] \u2794 [ B: Monitoring & Remediation ] \u2794 [ C: Data Product Design ] \u2794 [ D: Code Quality ] \u2794 [ E: DataOps & Run ] \u2794 [ F: Service & Consumption ]<\/div>\n
      \n
    1. Source Quality (A)<\/strong>
      \nThe primary imperative is to intervene at the upstream source systems: refining business input processes, adapting enterprise architecture, and enforcing strict input formats within IT applications.<\/li>\n
    2. Source Monitoring (B)<\/strong>
      \nImplementing automated checks on the raw data layer to detect anomalies upon ingestion and trigger immediate remediation workflows.<\/li>\n
    3. Data Product Design (C)<\/strong>
      \nFormalizing strict data contracts between business and engineering teams prior to development. These contracts define schema specs, business semantics, and acceptable data quality thresholds.<\/li>\n
    4. Code Quality via dbt (D)<\/strong>
      \nIndustrializing unit and integration tests across transformation pipelines using dbt. Powered by automated CI\/CD checks (dbt build), non-compliant code is blocked prior to deployment into production.<\/li>\n
    5. Operational Monitoring (E)<\/strong>
      \nContinuous monitoring of runtime performance: tracking data freshness, execution latencies, and volume spikes to prevent service disruptions.<\/li>\n
    6. Downstream Observability (F)<\/strong>
      \nSurfacing data health metrics directly within BI and reporting tools (e.g., Tableau, Sigma) to alert end-users in real time when issues are detected.<\/li>\n<\/ol>\n

      Business and Engineering: Covering the 8 Dimensions of Data Quality<\/h2>\n

      A Data Quality by Design initiative must simultaneously address business requirements and engineering constraints:<\/p>\n

      Business Dimensions (The 3 Cs + Accuracy):<\/strong><\/h3>\n
        \n
      • Completeness<\/strong>: Absence of unjustified missing values.<\/li>\n
      • Conformity<\/strong>: Adherence to standardized formats and business rules.<\/li>\n
      • Consistency<\/strong>: Logical coherence across tables and columns.<\/li>\n
      • Accuracy<\/strong>: Strict alignment with real-world ground truth.<\/li>\n<\/ul>\n

        Technical Dimensions:<\/h3>\n
          \n
        • Uniqueness<\/strong>: Absence of duplicate primary keys.<\/li>\n
        • Timeliness<\/strong>: Data availability aligned with scheduled SLAs.<\/li>\n
        • Freshness<\/strong>: Cadence of data refreshes matched to operational needs.<\/li>\n
        • Integrity<\/strong>: Lossless data preservation across ingestion and transfer pipelines.<\/li>\n<\/ul>\n

          The Value Proposition of the dbt Labs Platform<\/h2>\n

          Leveraging dbt translates these theoretical dimensions into executable code specifications. Checks are declared in human-readable YAML configuration files<\/strong>, version-controlled via Git, and executed automatically with every model update.
          \nBy combining dbt\u2019s data lineage, data contracts, and agentic AI-driven assistants (such as dbt Wizard), engineering teams can pinpoint the root cause of an anomaly in minutes while maintaining full end-to-end transformation lineage.<\/p>\n

          The Critical Impact on Agentic AI<\/h2>\n

          The emergence of autonomous AI agents fundamentally escalates the risk profile associated with poor data quality. Unlike traditional dashboards designed to inform human decision-making, AI agents<\/strong> directly execute operational workflows inside core enterprise systems (e.g., ERPs, CRMs).<\/p>\n

          In this paradigm, flawed data no longer merely produces flawed reporting, it triggers erroneous automated transactions. A modest 1% error rate across 10,000 daily transactions yields<\/strong> 100 faulty automated operations per day. Autonomous AI therefore demands uncompromising standards for data quality, low latency, and semantic precision.<\/p>\n

          Action Plan for Data Leaders<\/h2>\n

          To transition from a reactive posture to a Data Quality by Design framework, data leaders should prioritize the following operational imperatives:<\/p>\n

            \n
          • Prioritize Strategic Use Cases<\/strong>: Establish data contracts for high-impact business data products rather than attempting to overhaul the entire enterprise data asset base at once.<\/li>\n
          • Automate Testing in the Development Lifecycle<\/strong>: Embed schema validation and anomaly detection directly into dbt pipelines within the CI\/CD workflow.<\/li>\n
          • Remediate Anomalies at the Source<\/strong>: Leverage observability insights to fix upstream input logic and enterprise IT flows, driving sustainable, long-term quality improvements.<\/span><\/li>\n<\/ul>\n
            \n

            Watch the webinar on YouTube (French)<\/strong>
            \n

            <\/div>