How Guitar Center rebuilt its data platform on trust

At a dbt Summit 2026 breakout session, Mike Doll, VP of Data at Guitar Center, walked through how his team rebuilt the company's analytics stack from the ground up, and why he thinks reliability is something you design in, not something you buy.

Doll joined Guitar Center in early 2024 to find a data platform in the middle of a failed migration. The team had tried moving off an on-premises Netezza data warehouse to Redshift, and the attempt left behind more than 2,000 lines of stored procedures, no CI/CD, and almost no observability into what was actually running.

Rather than patch the existing setup, Guitar Center moved to managed dbt Core through partner DataCoves and rebuilt around three pillars: controlling costs, improving delivery flow, and building real confidence in the data.

Controlling costs meant rethinking how compute got used, not just how much of it there was. The team ingests data once and shares it down, uses dbt's selectors.yml to scope builds precisely, and splits work across many small Snowflake warehouses and objects so ownership stays clear. That discipline, paired with a habit of continuously optimizing toward a target, paid off directly: optimizing the hourly sales object alone cut Snowflake run rate by 40%, and reducing Airflow minutes cut consumption by another 50%.

Guitar Center now sets a daily build time threshold of more than an hour as a trigger for an automatic round of optimization, rather than waiting for costs to become a problem worth escalating.

Delivery flow was the second pillar. Guitar Center standardized on identical development and production configs inside a containerized DataCoves environment, which let the team evolve a "fallback_source" macro into a more general SourceWrapper pattern and use SQL feature flagging to isolate in-progress changes within shared models. The practical result was that pipelines which used to take months to ship now take days, and the full modernization, from signing with Snowflake and DataCoves to having a working replacement for the old stack, took under a year.

The third pillar, confidence in the data. Stability is not the same thing as correctness, he argued: a pipeline can run reliably every day and still be reliably wrong. Fixing legacy bugs and validating new numbers with stakeholders had to happen alongside the technical rebuild, not after it.

Doll extended that argument directly to AI, saying that AI amplifies trust or distrust rather than creating either on its own. As he put it, AI will not fix an organization's unreliable data. It will scale whatever foundation that organization gives it.

In the Q&A, Doll fielded questions on two topics that come up in almost every legacy-to-dbt migration story: how small teams without executive buy-in should approach outside help, and how to handle PII and access control. On the first, his advice was to use consultancies carefully and never outsource ownership of the platform itself. On the second, Guitar Center designed masking and role-based access into the rebuild from day one rather than retrofitting it later.

Guitar Center's rebuild used dbt's incremental models to control how much data gets reprocessed on each run, and dbt's built-in data tests to catch the kind of legacy bugs Doll's team had to hunt down by hand. For a broader look at what dbt does under the hood, see dbt's product overview.

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