Delivery Hero's dbt Adoption Was Driven by Customer Teams, Not the Platform Team

At dbt Summit 2026, Delivery Hero's M. Waqas Shahid explained how dbt spread through the company from the bottom up, not from a top-down platform mandate.

Delivery Hero runs a decentralized network of seven regional platforms across roughly 65 countries, processing more than 11 million orders daily. Its data platform holds about 300,000 tables and 60 petabytes of data, with around 250 dbt projects spread across 45 Domain Data Units.

The platform team had built an in-house framework called DataHub, but customer teams found it hard to debug, slow to develop against, and steep to learn. Early exposure to dbt at a GCP roundtable, including Monzo's experience with a fork that slowed past 3,000 models, led the platform team to pause.

Customer teams didn't wait. By 2023 they were testing dbt on their own, and by mid-2023 they were pushing for integration. They built their own Python package, dbtflow, to convert dbt manifests into Airflow tasks. After Astronomer released Cosmos in mid-2023, Delivery Hero retired dbtflow, and the platform team's work narrowed to a CI layer including a Cosmos Injector script.

Shahid framed the company's decisions around four trade-offs: abstraction versus flexibility, standardization versus autonomy, self-control versus central control, and speed versus stability. He closed with two lessons: evolve with intent, and remember that readiness must exist at both the tool and organizational levels.

Source: dbt Labs Blog