How EF Education First built a cost and value layer for dbt models with a Snowflake Cortex Agent

At a dbt Summit 2026 breakout session titled "Does your data pipeline deserve to exist? How EF Education First built a cost and value layer with dbt and Cortex Agent," Adrien Ledoux, who leads the data platform team at EF Education First, described how a scoring system built on dbt and Snowflake turned an unmanaged cost problem into a self-service tool for the company's decentralized domain teams.

EF Education First, whose mission is "opening the world through education," sends thousands of students on tours around the world every year. Ledoux's platform team supports a data mesh of domain teams that each own their own marts and business logic.

The trouble started when EF migrated from a legacy SQL Server setup, where spend was fixed and the real constraint was throughput, to Snowflake's consumption-based compute, split into separate warehouses per domain team.

That move gave domain teams the flexibility to run pipelines on their own schedules, but it also meant every dbt model run now carried a real cost. Ledoux's team became responsible for allocating that spend fairly across a decentralized organization.

Three symptoms showed up quickly: costs climbing steadily, some pipelines with barely any downstream usage, and a handful of expensive projects nobody was clearly accountable for.

Ledoux's team responded by building a scoring system for every dbt model and test, based on execution time, credits consumed, and rows processed per second. Scores are normalized as a percentile rank against similarly sized peer models, so a large fact table isn't penalized unfairly against a small one.

Models are sorted into three tiers: top-tier models are left alone and highlighted to the rest of the company as success stories, mid-tier models get periodic review, and bottom-tier models trigger a review conversation with the owning domain team.

The scoring system draws on Snowflake account usage transformed with a Kimball-style approach, plus a custom pipeline that ingests Power BI activity logs. Ledoux noted that Fivetran shipped a native Power BI connector that could replace that custom ingestion going forward.

Those sources feed a governed semantic layer in Snowflake, wrapped in a Cortex Agent that domain teams can query. The agent is enriched with context like domain, user, and team, so it already knows who's asking and what they're responsible for.

The effort paid off in measurable terms: some of EF's heaviest pipelines saw a 40 percent efficiency improvement, one of the biggest models dropped from 10 minutes to 2 minutes, and some Power BI workspaces saw up to 70 percent of reports deleted once low usage became visible.

Ledoux was candid that the project's biggest ongoing challenge is adoption of the agent itself, even though building it pushed his data engineers to work more like analytics engineers and surface gaps in their own CI/CD.

He also drew a sharp distinction between usage and value: a report checked a thousand times isn't necessarily valuable, and a report checked once a week by the right person can be. His team is now building a reverse ETL loop that pushes actionable insights into Salesforce, assigns them to people, and re-ingests the outcome to close the feedback loop.

Other next steps include making the agent more autonomous by feeding it Confluence documentation through Fivetran, connecting it to Snowflake workflows that can automatically open Jira tickets for domain teams, and pushing regular usage digests directly to domain teams who might not otherwise know the tool exists.

Ledoux's closing point doubled as a warning: a good technical idea does not adopt itself, and leaders still have to invest the time in change management.

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