Vinted Go is Vinted's in-house delivery service operating across multiple countries, coordinating a network of carrier partners to manage and fulfil parcel shipments efficiently. Accurate reporting of shipping costs and operational data is essential for financial transparency, optimising carrier agreements, and supporting strategic decision-making.

Our data pipeline integrates carrier invoice data, shipment records, and pricing details from various external sources, making high data quality critical for trustworthy cost reporting and operational insights. Ensuring high data quality while meeting strict Service Level Agreements (SLAs) is a common challenge in modern data platforms.

In early 2024, when we introduced dbt's native testing capabilities into our Vinted Go Finance pipeline, we initially included all tests directly in the build process. Although this improved visibility into data quality issues, it caused frequent pipeline failures on non-critical errors, leading to SLA fulfilment rates between 77% and 87%.

This undermined stakeholder confidence, delayed data availability, and generated alert fatigue. Additionally, the cost of running all tests—including longer runtimes and resource consumption—became unsustainable. To address these issues, we developed a risk-based testing strategy grounded in established risk management and data quality frameworks.

We classify tests based on the financial impact of potential errors, the frequency of failures, and the operational cost of running them. High-impact, frequent tests remain in the main pipeline, while lower-impact or infrequent tests are run separately on scheduled workflows. This approach reduces unnecessary build failures, lowers runtime and resource costs, and minimises alert fatigue. We will share practical frameworks and guiding principles on how to decide what tests to implement and prioritise, balancing rigour with operational efficiency.

We will provide concrete examples of tests we run—covering join key validations, accepted values, invoice completeness, and anomaly detection for invoice amounts. We will also explain how we tag tests, organise workflows, and delegate responsibility for addressing failures between finance and data teams. By adopting this risk-aware testing approach, we improved SLA fulfilment to over 97% within 30 days without compromising data quality