Monzo Builds Governed Data Mesh to Support 100+ Teams and 12,000+ dbt Models

Monzo has overhauled its central data platform to handle contributions from more than 100 independent teams and a codebase exceeding 12,000 dbt models, moving to a governed, 'meshy' architecture that changes how datasets are shared and validated across team boundaries. The redesign aims to let teams work independently while preventing regressions and cutting duplicated work that previously drove up warehouse costs. The new architecture enforces four explicit modeling layers: automated landing models that flatten incoming events; generated normalized models that represent entities with full history; logical models that combine entities; and presentation models for downstream use. Cross — team dependencies must go through declared interface models, creating stable, versioned boundaries between teams’ datasets.

To operationalize the design, Monzo introduced Modelgen, a command — line tool that generates SQL and YAML from object definitions, and paired it with continuous — integration checks. The CI pipeline validates model structure, naming, access patterns and metadata before changes reach production, ensuring new contributions conform to the governed mesh. Monzo applies concrete CI requirements to every dbt model: each model must declare a unique key, include freshness tests, run incrementally by default, name an owning team, provide documentation, and follow strict naming and metadata conventions. Those rules are intended to reduce redundant queries and recomputation and to keep datasets performant as many contributors — including AI-assisted code contributors — edit production dbt projects.

Rollout is ongoing: the team reports about 30% completion of a company — wide migration that already covered thousands of dbt models and introduced hundreds of governed interfaces. Early operational outcomes include roughly a 40% reduction in warehouse costs and about 25% faster data landing in some domains, though the engineers caution it is still early and further migration and validation remain.

Industry commentary highlighted that few firms publicly run similar tooling: Monzo’s Engineering Director, Luke Briscoe, said comparable public examples are rare, and founder Mateusz Ulas argued that treating data interfaces as first — class code and wiring standards into CI produces real improvement beyond documentation alone. The piece also notes the growing challenge of AI-assisted coding increases the need for automated guardrails.