Blue Yonder had a thousand dashboards and no idea which one was telling the truth.
Ask three teams to define the same metric, and you’d get three different numbers. Each one was defensible, but none of them were trustworthy because they couldn’t be reconciled with the others.
Then the company rebuilt its data foundation around a governed semantic layer. Same question. Same underlying data. One answer instead of a thousand.
A Thousand Dashboards, No Single Truth
Blue Yonder’s BI team wasn’t struggling with low usage. Dashboards were getting hit thousands of times a month, more traffic than Brad Lindsey, Head of Enterprise Data, had seen at any company he’d worked at before. Trust was the problem.
“We’d ask what sounds like a simple question. ‘How do you define this metric?’ We’d get a clear answer, but then we’d look somewhere else in the company, and we’d realize that the metric was being calculated completely differently in other places.”
— Brad Lindsey, Head of Enterprise Data, Blue Yonder
The problem was the lack of a shared foundation underneath the reporting.
Each dashboard had its own bespoke dataset built from scratch, with nothing in common with the dashboard next to it. The same metric lived in four places at once, in Databricks, Power Query, and an Excel macro, and nobody could say with confidence which copy was correct. Tracing where a number actually came from was a manual hunt through someone’s spreadsheet history.
Even within Blue Yonder’s robust analytics program, self-service wasn’t truly self-service. Business users still relied on the analytics team for every request. The small team had become the bottleneck for the business it was supposed to be serving.
The Reframe: Not BI Modernization, but Operational Risk
Blue Yonder’s leadership made a deliberate call here, refusing to position this as a BI modernization project.
“We positioned it as a control and scale problem… because really the reality was that without standardized definitions and governed logic, we’re creating operational risk.”
— Brad Lindsey, Head of Enterprise Data, Blue Yonder
Executives don’t fund infrastructure for its own sake. They fund risk reduction. Once leadership understood that inconsistent data wouldn’t just fail to scale, it would actively compound once AI got layered on top (“you’re going to get wrong answers faster with AI if it’s not sitting on top of this foundation,” as Lindsey put it), the initiative stopped being optional.
The team’s new charter came down to four goals:
- Centralize the logic in one place
- Make governance unified and auditable
- Get the ecosystem AI-ready
- Deliver self-service without the bottleneck.
One Layer, Four Fixes
Blue Yonder moved away from a traditional medallion architecture toward what Jeremy Arendt, Sr. Director of Analytics Engineering, calls a “semantic staging architecture.”
- Raw data lands and transforms into dimensional staging objects that feed directly into the semantic layer, rather than having business logic buried in a gold layer where nobody can find it.
- Governance stopped being a side process. Every key metric now has a named business owner inside the semantic layer. Definitions get proposed, brought to the business for sign-off, and versioned, which means auditors and finance now have a trail they can actually rely on instead of tribal knowledge.
- Self-service finally became real self-service. With metric logic, aggregation, and query performance managed inside the AtScale semantic layer, business users don’t need to know SQL or data modeling to work with governed data. That shows up now in Power BI dashboards, tabular reporting, and direct AI tool integration, all of which pull from the same source.
- Blue Yonder connected its semantic layer to internal AI tools via AtScale’s MCP Server, enabling AI agents to use the exact same governed metric logic that dashboards and Excel use. Lindsey now calls it “an operational control plane for all of our enterprise data.”
Three Moves That Bought Them Speed
Most infrastructure overhauls this size stretch into multi-year slogs. Blue Yonder’s didn’t, for three reasons.
First, they upskilled rather than hire. Rather than recruiting externally, the team identified BI developers with a data engineering aptitude and moved them directly onto the analytics engineering team.
Second, they used existing dashboards as the requirements document. Instead of running months of discovery workshops, the team mined current reports and used AI to help extract metrics and lineage. This became the starting point for new semantic models.
Finally, they shipped iteratively: ideate, build a proof of concept, validate, deliver, repeat.
“We never show up with nothing to any call. We always do our homework, build what we can with the information we have, and go from there.”
— Jeremy Arendt, Sr. Director, Analytics Engineering, Blue Yonder
The 30 Hours
This is the moment where the semantic layer work pays for itself.
A lead engineer’s team was asked to run a multi-day financial deep dive on products, spend, and a deprecation schedule for a client. It was the kind of request that normally takes a week. Jeremy decided to run a test. He took the original request and ran it through Blue Yonder’s AtScale MCP-powered AI workflow with no extra prompting.
“We essentially collapsed 30 hours of work over a four-day period between multiple people… into a one-and-a-half-minute wait time out of an AI tool. The outcomes, the analysis, and the financial figures were all very similar.”
— Jeremy Arendt, Sr. Director, Analytics Engineering, Blue Yonder
The output wasn’t perfect, but it got the team 85-95% of the way there. Because it ran on governed semantic queries, every number could be traced back and validated.
Eighteen months into the project, a team of seven analytics engineers, seven data engineers, eight BI developers, and one data governance lead had deployed 10 governed semantic models and 600 semantic objects, replacing what had been over a thousand disconnected dashboards and 800 tables. Business users and BI developers can now query the same production models, whether they’re in Excel, Power BI, or an AI tool.
Word travels fast when the demo actually works. Every internal showcase of the semantic layer connected to AI has generated two more demo requests by word of mouth alone. Jeremy says it’s the first time in his career he’s seen this level of executive pull behind a data infrastructure investment.
The pitch that sold it internally was simple: you get the same answer everywhere.
What’s Next Isn’t a Finish Line
Blue Yonder is building an asset-focused data portal that treats semantic models, metrics, reports, and AI agents as discoverable assets in one place, instead of a long list of dashboards nobody can find. They’re building AI-specific semantic context, because the descriptions written for human users aren’t the descriptions AI tooling actually needs. They’re using AI coding tools to speed up the manual work of building and maintaining the models themselves. And they’re pushing governance deeper into the layer itself, so metric ownership and sign-off history become a built-in trust mechanism rather than a side process someone has to remember to run.
“This isn’t about laying AI on top of BI dashboards. It’s about what becomes possible when our infrastructure is semantic first.”
— Jeremy Arendt, Sr. Director, Analytics Engineering, Blue Yonder
The 90 Seconds
Metric logic that used to live in a dozen disconnected places now lives in one. Whether you use Excel, Power BI, or an AI agent, you’re computing from the same trusted definition.
Thirty hours became 90 seconds because the foundation underneath it finally told the truth, the same way, every time, to everyone who asked.
Ready to build a semantic foundation your AI agents and BI tools can both trust? Talk to AtScale about deploying a governed semantic layer for your enterprise.
