How WWT Modernized Its Own Analytics Without Waiting for Perfect Data

The challenge: AI was moving faster than data access

Earlier this year, WWT brought together engineers from across the business for an internal AI innovation event. The demonstrations showed how quickly teams were building new AI solutions, but also exposed a common challenge: AI was outpacing data access.

Many teams were relying on manual workarounds to bring data into their AI workflows, scraping information from dashboards and exporting spreadsheets, then copying the results into prompts. It worked, but it added friction, increased token consumption and created governance challenges.

Like many large enterprises, WWT already maintains analytics data across its core business systems, including Oracle ERP and Salesforce. Historically, we curated that data in Snowflake and Oracle data warehouses and then surfaced it through reporting platforms such as Tableau and Power BI. However, that infrastructure had been built to help people read dashboards, not to help AI tools access data on their own.

As we explored how employees were using AI, members of the enterprise architecture team saw an opportunity to put existing data to work by connecting it to AI workflows through a governed approach. Rather than waiting for a large-scale transformation project, the team moved straight into building it.

Building the solution, Strata

Several principles would guide the solution:

To meet those goals, they used GraphQL as the foundation for exposing analytics data. Unlike traditional APIs designed around transactional requests, GraphQL excels at retrieving and relating data across multiple business domains, which closely aligns with how analysts work. That mattered because the value of enterprise data often comes from connecting it across systems (e.g., pairing sales opportunity data with quote data to see the full picture, a task traditional transactional APIs aren't built to handle efficiently).

The result was Strata, an internal platform built on three core capabilities: data access, business context and governance.

The first core capability, data access, creates GraphQL endpoints directly from Oracle and Snowflake data sources, so teams can work with existing data without migrating or restructuring it before connecting it to AI tools. That access runs through the MCP, the emerging standard for how AI systems interact with governed data, which means the approach isn't a one-off integration but is built on the same open standard other organizations are adopting.

The second capability is a data catalog and advisory layer that captures the business context and knowledge data stewards already possess. Dataset definitions, usage guidance, field explanations and query best practices are stored in markdown, traveling alongside the data whenever an AI tool might need to access it.

This advisory layer proved especially important because enterprise data is rarely as straightforward as it appears. The same customer may be represented differently across systems, naming conventions vary from team to team, and business rules often exist only in the minds of experienced analysts. Human experts account for those nuances instinctively; AI systems require that knowledge to be documented.

The third capability is governance. Rather than relying on ad hoc report exports or dashboard scraping, access is managed through an API layer that gives teams visibility and control over how AI systems consume data.

Results are already taking shape

Users across the business are already demonstrating how AI-connected analytics can change the way work gets done.

One example comes from WWT's labs organization.

A director of production operations was asked about a decline in inspection pass rates that had occurred roughly a year earlier. Historically, answering that question required multiple reports and manually connecting information across different datasets.

Using Anthropic's Claude, connected to Strata, the director composed a simple query about the decline. The platform drew on manufacturing data and business context from the advisor layer, then used the relationships between datasets to trace the issue through defect records, a specific program, and service templates. The analysis ultimately identified a change in label requirements, which triggered a large relabeling effort, as contributing to the decline.

This investigation took less than 30 minutes. Previously, it would have required running four or five separate reports and manually connecting them, work that could easily stretch into multiple hours.

Our Strata platform is also influencing how teams think about reporting itself.

Marketing teams, for example, use Strata to define exactly how common questions about Amplitude website analytics should be queried, so Claude pulls consistent answers every time rather than rebuilding the analysis from scratch. Some questions that once required a formal dashboard can now be answered conversationally, and users are experimenting with faster ways to build live reports connected directly to those data sources. Platforms like Tableau and Power BI continue to play an important role, but employees can increasingly start with a question instead of a report request.

That shift also improves efficiency. Governed APIs reduce repeated exports and manual uploads, and as AI adoption scales, this approach strengthens governance while reducing unnecessary token consumption.

How AI-native engineering played a role

One of the most telling parts of the project involves who brought Strata to life: two internal IT professionals whose backgrounds were in analytics, reporting and databases rather than traditional application development.

That background shaped how they worked. Using AI-powered engineering tools, they moved from concept to working pilot in roughly two months, accelerating everything from architecture exploration and code generation to testing and iteration. Instead of writing detailed specifications and handing them to an engineering team to wait weeks for implementation, they worked within a self-contained rapid feedback loop. They described requirements in natural language, evaluated results, refined their approach and built again.

That loop is what turned the project into an internal example of AI-native engineering (AINE), proof that deep business and data expertise can be just as important as software development experience when building AI-enabled solutions.

What other organizations can take from this

What Strata proves is possible matters more broadly than the platform itself.

Many organizations believe they need perfectly prepared data before they can begin realizing value from AI. WWT took a different approach, accepting that enterprise data would always contain complexity and inconsistency. The focus shifted to creating a governed way to connect that data to AI and provide the context needed to interpret it correctly.

That decision paid off quickly, in part because it didn't require starting from scratch. Existing investments in Oracle and Snowflake already provided a strong foundation. The work that remained involved building the connective layer that let AI systems interact with those assets responsibly and effectively.

That same resourcefulness shaped how the platform got built. WWT didn't need to staff a dedicated engineering team to get Strata off the ground. Two people with deep knowledge of the data, paired with the right AI coding tools, produced a governed, production-grade platform in a couple of months. For organizations weighing whether they have the engineering capacity to modernize their own data stack, WWT's experience suggests that the capacity gap may be smaller than it looks.

Our experience also points to a more useful question than "Is my data ready for AI?" Instead, try asking, "Is there a governed path from my data to the AI tools my people want to use?"

In many cases, that path may be the difference between experimenting with AI and creating measurable business value.