30 Hours to 90 Seconds: Blue Yonder’s Semantic Layer for Trusted Enterprise AI

Dave Mariani:
Hi everyone and welcome to another episode of our Data-Driven Podcast. Today we have some real practitioners doing groundbreaking work in the area of semantic layers. I want to welcome Brad Lindsey, Head of Enterprise Data at Blue Yonder. Welcome to the podcast, Brad.

Brad Lindsey:
Thanks Dave, happy to be here.

Dave Mariani:
And Jeremy Arendt, Senior Director of Analytics Engineering at Blue Yonder. Jeremy, good to see you.

Jeremy Arendt:
Yeah, thanks Dave. Happy to be here.

Dave Mariani:
Before we get into the details of the work these two gentlemen are doing around semantics for AI, let’s first talk a little bit about each of them. Brad, I want you to kick it off. Tell us about yourself and how you got to be Head of Enterprise Data at Blue Yonder.

Brad Lindsey:
Yeah, thanks again, Dave, and thanks for having us on the podcast today. We’ve been living and breathing semantic layers for the past 18-plus months, so it’s nice to step away and have a conversation about it.

As you said, I’m the Head of Enterprise Data at Blue Yonder. My team focuses on the data foundation that supports analytics, governance, and AI across the company. Blue Yonder is a supply chain software company. We work with retailers, manufacturers, and logistics companies to help them plan and run their supply chains more effectively.

I’ve been with Blue Yonder since 2023. I’ve been working in data since 2002, and I’ve seen a ridiculous amount of change. We’re talking about semantic layers today, but when I started, we were working in SQL and Microsoft Access. Then MicroStrategy came out and suddenly we could build interactive dashboards.

So I’ve definitely seen an insane amount of change over almost 25 years. I’ve worked across multiple industries — financial services, marketing analytics, healthcare, property management, and now supply chain. But I’ve never been more excited than I am today with what’s possible.

Dave Mariani:
Yeah, I share your enthusiasm. I’m so excited about what we can do now in analytics with AI. The semantic layer is really a critical piece of infrastructure to make all that happen.

Jeremy, how about you? How did you get to where you are today?

Jeremy Arendt:
I’ve been at Blue Yonder about 18 months now and have been doing data and analytics for about 10 years. Like many people in data, I had a varied path before landing here, including stints in education and even playing music for a living.

Before Blue Yonder, I worked at JLL for about seven years focusing on large industrial and tech clients. At Blue Yonder, I focus primarily on engineering and infrastructure.

I started my career doing dashboards and analysis, working very closely in that area. But we’d always get to a point where things would start to break down when we wanted scale, reusability, or a single source of truth. As we wanted new ways of interacting with data, those frustrations kept pushing me more into infrastructure and engineering.

That eventually led to my first exposure to AtScale through a demo and proof of concept at JLL. I’ve been working with it pretty much every day for the past 18 months at Blue Yonder.

Dave Mariani:
That’s awesome. Let’s start from the beginning. From what I understand about Blue Yonder, you made a real tough choice about 18 months ago to shift from building dashboards for the business to building data infrastructure. I think a lot of people and companies are contemplating that same kind of move.

Brad, tell us about how you made that choice and what drove that transition.

Brad Lindsey:
I don’t think there was one dramatic moment. It was more the accumulation of signals telling us the old model was breaking down.

When I joined Blue Yonder, we actually had a successful analytics program. Every team was hungry for data. We had dashboards everywhere throughout the business being used thousands of times each month.

But underneath it all, every dashboard was becoming its own version of the truth. Every request became a rebuild. Teams spent more time debating definitions than discussing the actual business problems.

The project that really crystallized this for us was a company-wide health scorecard initiative. Pretty straightforward concept — key metrics, targets, trends, RAG statuses, the kinds of things companies put into executive scorecards.

Everyone agreed it was important. But the friction came when we tried to standardize definitions.

We’d ask a simple question like, “How do you define this metric?” We’d get a clear answer, then look somewhere else in the company and realize the metric was being calculated completely differently.

That’s when it became clear the problem wasn’t reporting. It was the absence of a shared data foundation.

The shift wasn’t just from dashboards to infrastructure. It was from producing outputs to building a system the company could actually operate on consistently.

And to be clear, dashboards weren’t the problem. People still love dashboards. People still love Excel. We use them every day.

But now those dashboards sit on top of governed, reusable semantic models instead of isolated logic rebuilt over and over again.

Dave Mariani:
We talk a lot about technology, but when I was talking to Jeremy, he brought up the people process behind defining metrics and maintaining them. That’s really a people and process problem, not just technology.

Jeremy, can you talk about how you approached that?

Jeremy Arendt:
One of the biggest challenges driving all this was that every organization wants self-service analytics.

But historically, self-service looked like this:

Someone wants access to data, so they sit with us for 45 minutes while we explain all the context, rules, exceptions, and business logic.

That’s not really self-service.

As we moved toward a semantic foundation, we realized we could have those conversations once. We could encode calculations and context into a single governed layer and make it reusable.

That enables the kind of self-service we’ve wanted for years.

Getting those definitions right is both a people problem and an engineering problem. There’s information spread across dashboards, Excel workbooks, Teams messages, and other systems.

So we start by asking:

What information already exists?
What tools can we use to extract it?
What can AI help us identify?

We build our best guess first, then bring it to the business owners and validate it together.

One thing we’ve focused on throughout this journey is that we never show up empty-handed. We always do our homework first.

Dave Mariani:
And Jeremy, I think you also assign business ownership to metrics, right?

Jeremy Arendt:
Yep. Every key metric needs to be defined and owned by someone within the business.

Dave Mariani:
I know that sounds obvious, but that’s actually not obvious in practice across the industry. So listen up, listeners, because that’s something these guys have done fantastically well.

Let’s go back for a second. It sounded like you had a 360-view initiative that maybe drove some of this activity. Brad, what did it take to get support for this pivot? Was that the real driver?

Brad Lindsey:
There was more to it than that.

Blue Yonder is already a very data-forward organization. There’s strong executive interest in analytics and AI across the company.

But the real shift was moving from seeing data as an analytics capability to seeing it as foundational infrastructure for how the business operates.

Executives don’t fund infrastructure for infrastructure’s sake. They fund risk reduction, scalability, and operational efficiency.

So we intentionally did not position this as a BI modernization effort.

We positioned it as a control and scale problem.

Without standardized definitions and governed logic, you create operational risk:

Financial reporting becomes harder to trust.
Auditability becomes harder to prove.
AI scales inconsistency instead of accuracy.

Once AI entered the conversation, it became even more obvious.

If you don’t have standardized governed data with context, AI just gives you wrong answers faster.

Once leadership understood this wasn’t about prettier dashboards but about creating a trusted operational foundation for analytics and AI, the initiative stopped feeling optional.

There were reservations, absolutely. Anytime you move from decentralized logic to standardization, there’s concern about flexibility or speed.

But what we’re already seeing is that once business logic becomes shared and reusable, teams spend less time debating numbers and more time discussing the business itself.

That’s the direction we want the organization moving toward.

Dave Mariani:
You said something really important there. You didn’t position it as a BI modernization project. You positioned it as a governed data foundation project.

For years semantic layers were mostly tied to BI modernization, but AI changed that completely.

Jeremy, you described a use case where someone spent 30 hours doing a financial analysis, but then an LLM produced a similar answer in 90 seconds using the semantic layer. Tell us about that.

Jeremy Arendt:
That moment honestly caught me off guard.

We do internal demos where we connect Power BI, Excel, and an LLM to the same semantic model to show consistency across tools.

Separately, I was talking with one of our lead engineers about a client analysis that required pulling together product schedules, costs, and data from multiple people.

I asked him to send me the original request email.

I copied and pasted it into our LLM connected to the governed semantic model.

It generated an analysis document that was about 85% complete with no additional prompting.

Everyone was speechless.

That was the moment where we realized this changes everything.

The technology is moving so fast that it continues to surprise even us.

Dave Mariani:
That sounds like a real aha moment. Brad, did that help make the case internally?

Brad Lindsey:
Absolutely.

Whenever you can tangibly show something and move beyond words into action, it changes everything.

One thing we learned very quickly is that most people don’t care about data architecture. Honestly, they shouldn’t have to.

Nobody wakes up asking for a semantic model.

What people want is reliable answers in the tools they already use.

So when we talked about this initiative, we stopped talking about architecture and started talking about outcomes.

The language that resonated was really simple:

You get the same answers everywhere.

Same metric. Same definition. Same logic.

Whether you’re using dashboards, Excel, or AI.

We also talked a lot about reducing rework because teams were spending huge amounts of time recreating logic and reconciling numbers.

Once AI entered the picture, it became obvious that AI only works at scale if the underlying definitions and logic are consistent.

At that point, people stopped seeing this as infrastructure for the data team and started seeing it as operational infrastructure for the business.

Now the term “semantic layer” comes up constantly across the company.

Dave Mariani:
I can’t tell you how happy that makes me feel.

What people often miss is that semantic layers fundamentally change what AI can do.

Without a semantic layer, tools like Tableau or Power BI still require humans to manually explore data.

But when you put an LLM on top of a governed semantic layer, the AI becomes the analyst.

It can ask rapid-fire questions, generate queries, synthesize results, and deliver answers.

That’s not a faster dashboard. It’s a completely different way of interacting with analytics.

Brad Lindsey:
Exactly.

Before, people asked questions and our answer was:

“Here’s a dashboard.”

But that’s not an answer. That’s giving someone more work.

Now people can ask questions and get answers instantly.

Dave Mariani:
Jeremy, let’s talk about speed. How did you avoid becoming a bottleneck?

Jeremy Arendt:
My team is probably sick of hearing me talk about speed because we talk about it every day.

We never stop building.

We didn’t start with some giant monolithic plan. We knew we needed:

A semantic layer.
AI integrations.
A few key business domains.

And then we got moving.

A huge part of our success comes down to empowering smart people with autonomy.

People on our team have a lot of trust and ownership. We focus heavily on explaining the “why” so everyone understands the vision.

We also use every resource available:

Existing dashboards.
Excel spreadsheets.
Conversations with stakeholders.
AI tools.

We honor business users’ time by doing our homework before engaging them.

And we deploy iteratively. We don’t wait until everything is perfect.

We’re learning constantly, and failure is treated as useful data.

That’s a very different mindset than traditional BI.

Traditional BI says:

“Give me a request, let’s build the request, then move on.”

We’re now building things people don’t even know they need yet.

Dave Mariani:
I love it. It sounds like you’re great partners to work with.

One thing I’ve seen before semantic layers is that the business had total freedom to do whatever they wanted.

When you bring governance into the picture, it introduces constraints.

Have you seen resistance from the business?

Brad Lindsey:
A little, yes.

But ultimately, you can’t enable true self-service without governance.

That means restricting access to raw data and establishing controls.

Otherwise, you simply propagate inconsistency.

One thing that helped is that our organization now rolls up under security. We sit side-by-side with the security organization.

That reinforces the message that trust and governance matter.

People can still do what they want — but within governed guardrails.

Dave Mariani:
You definitely have air cover there, especially as headless agents become more common.

There are so many great nuggets here for listeners.

Let’s finish with a forward-looking question.

Brad, where does Blue Yonder go from here with this semantic-first foundation?

Brad Lindsey:
We don’t see the semantic layer as the finish line. We see it as foundational infrastructure.

The next step is embedding governed, reusable business logic directly into how the business operates.

Whether someone interacts with AI, builds a report, analyzes financials, or triggers a workflow, they should all operate from the same standard governed definitions and semantic context.

As AI adoption increases, this becomes even more important because AI is probabilistic by nature, but businesses still require deterministic controls around metrics and governance.

We really see the semantic-first foundation evolving into an operational control plane for enterprise data.

Dave Mariani:
I love that.

Jeremy, last word to you. What excites you most about the future?

Jeremy Arendt:
The semantic layer solves a lot of longstanding data problems, but it also creates exciting new opportunities.

Two major areas we’re focused on now:

First is user experience.

How do we create a delightful, intuitive experience where users can discover metrics, semantic models, dashboards, and AI agents easily?

Because if all this amazing infrastructure is difficult to navigate, adoption becomes hard.

Second is realizing that the descriptions we write for human users are different than the descriptions AI systems need.

We’re starting to think about role-based context and how to provide richer semantic context specifically for AI users.

There’s still a lot of foundational work ahead, but we’re also thinking constantly about what comes next.

Dave Mariani:
I love it.

You guys have been great. This is very actionable advice, and you’ve reached a high level of maturity very quickly.

Thanks so much for sharing your journey. This is really valuable for everyone trying to get their data ready for AI.

And to all the listeners out there, thanks for listening, and stay data-driven.

Brad Lindsey:
Thanks Dave. Bye.