Data-Driven Edge: Using Advanced AI and Normalized Data to Transform Aftermarket Operations

Jon Owens: Our distinct advantage is that we have a practice built around normalizing data. I like to refer to it as a data factory. Data comes to us in all different forms, in all different ways and variations. Unless and until you make sense of that data, your AI is going to stumble. Your AI is never going to really return the kind of results you're hoping to achieve.

We've been collecting industry data for over 20 years, long before the concept of big data was even important; collecting, normalizing, understanding, and interrogating data. We have a significant head start on this because it's something we've simply been doing longer than anyone else.

On the panel, it was said that "data comes in dirty." What does mean, and what does the cleanup process look like?

Owens: A good example is data that our catalog itself produces. Our catalog is utilized in many different systems out in the marketplace. The way those systems ping our catalog to find what they need to put on a ticket and subsequently sell the product means we're counting on that system to provide us with the data.

All of it comes in, and then we normalize the data. We're ultimately drilling down and sifting through all the data to get the right part for the vehicle that didn't come back as a return. Once we've identified that a water pump was sold for a 2016 Toyota Camry with a 6-cylinder engine, because our catalog was involved in identifying that part, we know it got sold.

When we talk about normalizing data, we're focusing on the behavior of the part and the behavior of the vehicle—not geographical data or that sort of thing. We're collecting data from platforms that use our catalog, and we know how they use our catalog. You may call the same part number something different. You might call it 123 or 456 or 789, but because we know how to interpret how you use our catalog, we're able to recognize that it's all the same part.

How do you achieve that level of accuracy when pulling data from so many systems?

Owens: When systems are using our catalog, we have something called the MCL—our customer setup in the catalog. It tells us what Epicor calls it and what you call it. We know that mapping because you're using our catalog. So when we get your data, we simply use that mapping to take what you call it and map it to our catalog part number. We're mapping everybody's data to our catalog part number because we know how you map it. We use that map to drive everything to our catalog part number, which is really the manufacturer part number—manufacturer to manufacturer. Understanding how to do that, we have the tool that understands that mapping.

What data quality challenges do you face, and are there datasets you can't currently use?

Owens: I'll give you an example. When we talk about repair order data—what was the job done on that vehicle—every service shop management system handles it differently. Every service writer is putting in codes, whether it's LOF (lube, oil, filter) or W2. If you have a system that requires you to pick a job, that's great. But if you don't, you get a lot of free-form text—shorthand, like a doctor's notes. It's really easy to figure out that it's an oil change, a fuel pump, or an air filter replacement. But we don't care how often someone's getting an oil change or replacing windshield wipers. We're trying to get to the more complex repairs. The more complex the repairs, the more complex the writing is. That's our biggest challenge right now: being able to decipher what job was done on that car. Once we understand the job, we can say, "Here are the parts that were used to solve that issue with that vehicle."

How has your approach to AI evolved since 2023?

Owens: We thought we weren't ready for prime time quite yet because we were applying an AI tool that wasn't our own AI, and we weren't getting the results we expected. We collectively took a step back. We decided we don't need an outside third-party AI service. Epicor has its own AI, and we're going to apply that to our data and run through this process again.

Since we pulled back, we're discovering we have access to more data than just our own. We're going to source data outside our four walls and merge that with the data we have; let our AI filter through it, and mine for gold, if you will. There are certain things we need to find from that data, and our AI tool will be directed to find those things for us.

Will the AI be primarily on your side, or integrated into the software for end users?

Owens: It's both. They really feed each other. We're building AI into the product, but we're also using AI to identify upfront: Is this even data we care about? We've already got examples where the AI is teaching itself at a very rapid pace, moving faster than I thought it would, particularly in our inventory modeling tool, where we've gone all-in on AI across our dataset. Before, this AI tool would have been leaps and bounds slower, but now it's teaching itself, learning the data, and producing results much faster than when we started. It exponentially gets more and more efficient and more accurate.

What improvements are you targeting with the Vision ERP redesign, particularly the UX and UI?

Suellyn Sprague: It's two things: tech stack and workflow. Vision today is a client-server application. We're moving to a browser-based web UI—that's a big lift right there. The UI looks okay today, but we're working very closely with a couple of our very large customers, working with their counter people who use it every day. We're not just sitting in a lab coding away. They're literally working with us every single day, helping us tweak the workflow, cut down keystrokes—all the things you'd want from a modern UI.

Instead of having to go four places to do something, we're finding ways to fit that into one place. We're making dashboards that are configurable—as configurable as we can make them for users. People say, "Well, it's got such deep functionality in it," and that's just not enough anymore. You can't stand on that forever. The graphical interface we thought was great 10, 12 years ago has got to be configurable. Users have to be able to sit down and say, "On this dashboard, I want this information, and I don't care about that." Our UI today isn't configurable to that degree.

We're also starting to embed analytics—related parts at the counter, those types of things. With the UI we have today, it just doesn't lend itself to that. We need a modern tech stack to embed the AI and analytics that we already own.

How does this redesign tie into your broader customer experience strategy of radically simplifying the interface?

Sprague: That's a big part of it. It's about letting users define what their experience should be. We have some customers who like having 120 icons on their desktop, and others who function with three. Giving that to the user—letting their POS, accounts receivable, be as busy or as simple as they want and allowing them to change that on the fly—that's what we're really focused on across Epicor. We have something called the Epicor Design System. All our go-forward products are starting to look and feel the same way, so you can recognize it's an Epicor product. We've done that in manufacturing and distribution, and we're doing it in automotive today. Even with the catalog, it's the same look and feel as the catalog parts lane product you see here.

What innovations from 2025 are you particularly proud of, and what should the industry expect in 2026?

Owens: Our Selenium acquisition was really a gold star for us, but probably the biggest innovation was bringing the new catalog embedded in Vision to our ERP customers. Vision is still a client-server application, so it still has a client-based UI. We're taking a web-based catalog and embedding it seamlessly into the application. It wasn't easy to provide that experience and the ability for countermen to run that catalog and the new catalog next to each other in the same session—not running two different instances of Vision, but having them run side by side in a single instance on a single license.

When I first said I wanted them to run both on a single license in a single instance so they could adopt at their own pace, some questioned it strategically. But I think it makes all the sense in the world. We're never going to get adoption from counter people if they can't go at their own pace. It's proving itself out.

For 2026, I'm going to fall on Parts Lane. Really getting the Parts Lane product out into the marketplace is what will happen in 2026. To what degree it will happen, I don't want to guesstimate, but we're going to land it. It's going to be put into use. It's going to work tremendously well.

We're bringing a product that's going to make it a more efficient and pleasurable experience for the service writer. Service writers need tools that are easy, configurable, and put information at their fingertips so they can make a very intelligent recommendation. It's not the owner of the shop telling the service writer to sell something. The data is telling us that when we sell this part, we should also be selling this part.

One thing we're not talking about a lot is Acadia, which is a learning management system. I envision Acadia being synced up with our Parts Lane tool and catalog. We can hit users with training: "We noticed this. You may want to take this brief course right here that will help you be much better at it the next time you face it." So, embedding training right into the workflow.

Sprague: Our real focus for Vision, in addition to the new UI, is to build the solution around all the other products we have. We're keeping the development team focused on what they're good at—aftermarket parts distribution—while building out the solution with Acadia, our payments product, and all these different products. So we're really leveraging the power of Epicor.