For 125 years, Nordstrom has built its reputation around exceptional customer service and experiences. Today, the iconic fashion retailer serves roughly 30 million active customers across its digital business and nearly 400 locations, generating more than $15 billion in annual revenue.
Delivering that experience at scale depends on data. Nordstrom's 150-person data and insights organization supports teams across customer, merchandising, finance, supply chain, stores, and more. More than 1,000 DAGs deliver the data behind everyday decisions, from what products to buy to how teams serve customers.
To make insights more accessible across the business, Nordstrom developed a 3-year transformation strategy focused on scaling data ingestion and transformation, delivering governed and trusted datasets, and enabling self-service insights through conversational analytics.
But one question kept Nordstrom's data leadership up at night: What would it take for data to answer plain business questions and be trusted by every team? AI could produce an answer quickly and confidently, but without consistent definitions and business context, that answer could still be wrong — potentially leading teams to make decisions based on inaccurate information.
Rather than layering AI onto its existing environment, Nordstrom turned to dbt to embed governance into data development. By standardizing definitions, automating quality controls, and documenting business context, the company is creating trusted datasets for both people and AI.
Although still early in its dbt journey, Nordstrom's pilot teams are already developing pipelines 2-5x faster, completing projects in days instead of weeks or months, and preparing governed data sets for AI in a sprint instead of up to 6 months.
Moving from conflicting metrics to governed data by default
One of Nordstrom’s most important metrics illustrates the challenge: customer trips.
The customer organization defined a trip as 1 purchase per customer, per day, per channel. The store organization used a different definition: if that customer shopped with separate salespeople in beauty, shoes, and apparel, the store team counted 3 trips.
Neither definition was necessarily wrong. The problem was that the context explaining the difference lived in people’s heads. Leaders could spend a business review debating whose number was right instead of answering the question that mattered: “How do we drive more customer trips?”
AI made that gap more consequential. When Nordstrom put a bot on top of its data, it returned 1 answer without understanding the competing definitions. Rather than resolving the ambiguity, AI made an incomplete answer available faster and to more people. That experience made it clear that Nordstrom couldn't simply layer AI onto its existing data environment and expect trusted answers. AI needed more than clean data — it needed the business context to understand what that data meant.
Nordstrom adopted dbt to build governance directly into data development, making testing, data quality, contracts, and metadata mandatory for governed models. A model without a description, for example, doesn't merge. Priya Tanwar, Principal Engineer, Data Platform at Nordstrom, cautions that requiring metadata is only the first step: "A required description isn't automatically a good description.” dbt can enforce that documentation exists, but people still need to ensure it accurately reflects the business.
Nordstrom is also aligning its dbt projects with approximately 14 business data models that reflect how the company operates. Rather than forcing every team to use the same definition, Nordstrom is documenting where definitions differ and making that context explicit. A metric called "customer trips" should have a clear meaning, while other variations should be distinctly named and defined so people and AI don't confuse them.
For Nadine Bruxel, Head of Data Insights Platform at Nordstrom, the value is getting teams focused on improving performance rather than debating definitions. With shared metrics and context, she says, Nordstrom can "allow the humans to actually go after new customers" and figure out how to increase trips, instead of spending business reviews arguing about what a trip means.
Giving AI the context to understand the business
Nordstrom learned that clean data alone isn't enough for trusted AI. A table can pass every quality check and still be misinterpreted if the person or agent consuming it doesn't understand what the business means by a field such as “customer trip.”
That makes context — definitions, metadata, KPIs, relationships, and business knowledge — a critical part of the data product itself.
Its medallion architecture progresses from raw data through conformed silver models into governed, business-ready gold data sets built with dbt. Dashboards and agents consume this governed gold layer rather than accessing less-curated data directly.
Nordstrom's high-level data architectureNordstrom pairs those data sets with a centralized context management layer backed by DataHub and GitHub, bringing together dbt and Looker metadata, KPIs, and a business glossary. Business teams help define that glossary, ensuring the context reflects how Nordstrom actually operates rather than requiring the data team to infer business meaning.
"To build the right metadata, it's extremely important to involve the business and have them define it," says Tanwar.
Nordstrom is deliberately supporting 2 paths to conversational analytics. Many teams already work in Looker, while leaders prefer asking questions through Claude. Rather than building separate business definitions for each interface, Nordstrom wants both to draw from the same governed data and context, allowing employees to explore information in the tools they already use.
That opens a more scalable path to self-service. Rather than asking the data team to build another dashboard every time 1 question leads to 10 more, leaders can start with core dashboards and use natural language to dig deeper — getting answers faster without requiring the data team to anticipate every question.
For Bruxel, building trust in data and AI isn't a one-time project: "I think of context and metadata just like we think about customer service at Nordstrom. There is no finish line. We're always evolving, we're always improving, and we're going to have to work on that every day so we keep building that trust."
Moving from months to days and preparing data for AI in a sprint
Governance is often perceived as slowing teams down. Nordstrom is beginning to see the opposite.
By standardizing logic, automating tests, and embedding governance into development, teams can reuse trusted patterns instead of repeatedly resolving the same questions. For pilot teams fully operating in dbt, development is 2-5x faster. Bruxel describes the change as a shift in how quickly the data team can respond to business priorities: "We're doing things in days versus weeks and months. We're starting to move at the speed of the business."
Nordstrom is also using AI to accelerate development itself. A multi-agent framework built on dbt MCP Server can interpret a developer or analyst’s requirements, build models across its medallion architecture, generate tests and metadata, and validate the work.
Nordstrom is deliberately keeping people involved as it introduces AI. As Bruxel explains, "We are also choosing to not fully automate. We do want the human in the loop, and that is a choice." For a company built around customer and employee trust, moving directly to full automation without building confidence along the way would be the wrong approach.
That lets Nordstrom increase engineering velocity without trading away trust, and the benefits continue downstream.
Nordstrom is still migrating legacy datasets into its governed gold layer, but the team is already seeing the potential for much faster AI deployment. Bruxel explained that once a dataset has been curated into gold with the necessary quality and context, work that previously could require another 6 months to prepare for an agent can now be completed within a sprint.
Instead of treating every agent as a new data-readiness project, Nordstrom is working toward making AI readiness an outcome of how its data is built.
Turn data governance into an AI advantage
Nordstrom's 3-year transformation strategy is about making trusted insights more accessible across the business. dbt is helping the company work toward that vision by embedding governance into data development, making business definitions reusable, and preparing governed datasets for conversational analytics and AI.
The early results show why that approach matters. Pilot teams are developing faster, business context is becoming part of the data itself, and Nordstrom is reducing the additional work needed to prepare governed data for AI. But as Bruxel notes, building trust is an ongoing process, not a finished implementation.
For data teams pursuing a similar strategy, Nordstrom's experience offers a practical lesson: build quality controls and business definitions into development from the beginning, involve the business in defining what the data means, and make that context available wherever people and agents need it. With those foundations in place, teams can spend less time reconciling numbers and more time making decisions that improve the business.
