A dbt "logic mesh" with packages: standard model and metrics across 25 autonomous organizations

At a dbt Summit 2026 breakout session titled "A dbt 'logic mesh' with packages: standard model and metrics across 25 autonomous organizations," Koen Verburg, lead analytics engineer at Greenpeace International, described how a 3-layer dbt package architecture let a roughly ten-person central team support a federation of 25 independently run national and regional Greenpeace organizations.

Greenpeace has operated for 55 years as a federation, with a separate Greenpeace entity for each country or region, including the US, Canada, Germany, the UK, Southeast Asia, and Africa. Each of those organizations runs its own tech stack, so Greenpeace International's data team was effectively working across 25 different source systems and 25 different BI tools.

According to Verburg, the underlying data wasn't especially complex; the organizational structure was. Every organization built its own income report and its own retention report from scratch, so a data team spread across the network was duplicating the same work 25 separate times.

Metric definitions varied just as much as tooling. Verburg said "donor" alone had multiple competing definitions across organizations, with some counting anyone who gave in the past year and others counting only active, recurring donors, which made global comparisons unreliable.

Greenpeace first tried standardizing on a single source system across organizations, which worked in some cases but broke down wherever local context genuinely differed, such as regional payment methods. An ad hoc approach that rebuilt the same transformation logic 25 times for every new metric didn't scale either.

The team adopted dbt in 2025 and started by modeling the business itself rather than the tools underneath it, deliberately treating source systems as changeable and the underlying business activities as stable. Verburg said the hardest part of that exercise was linguistic: getting 25 organizations to agree on what a shared term like "donor" actually meant.

The resulting architecture has three layers. Source packages handle transformations for systems shared across multiple organizations, a thin intermediate layer absorbs each organization's system-specific quirks, and a standard data model, distributed as a single dbt package, carries the shared logic, documentation, tests, macros, governance, and semantic layer definitions.

Local organizations can still define their own metric variants for local reporting, but global reporting requires everyone to agree on one shared definition. Verburg said that trade-off, consistency over any one definition being the "correct" one, was the point.

The stack behind it is Fivetran for ingestion, dbt for transformation, BigQuery on GCP for storage and compute, Hex for consumption, and Terraform to manage deployment across all 25 organizational instances. Rolling out a full data model for a new organization now takes one to two weeks, and 14 of the 25 organizations had completed that rollout as of the talk.

About 75 percent of the logic in any given organization's project now lives in shared packages, with the remaining 25 percent handled as organization-specific customization. Greenpeace ships roughly 50 package releases a month, and those updates propagate automatically to 170 consuming projects, 120 of which require no human intervention because CI checks approve and merge the change on their own.

Verburg was candid about the tradeoffs. Change management and re-educating each organization to trust a shared model took significant time, and the resulting codebase, spread across 25 dbt projects layered with packages, is harder for a new contributor to trace than a single monolithic project would be.

He also said his team has spent the past year testing dbt v2 outside of production, and Greenpeace's unusually package-heavy setup has surfaced bugs that more typical dbt projects wouldn't hit. Even so, he described the net effect as a shift from multi-month delays on global metrics to daily insights, with new metrics now defined once in a package and distributed to every organization instead of being rebuilt 25 times over.

Watch every keynote and breakout from dbt Summit 2026 on demand. Watch dbt Summit 2026 on demand →