In this episode of Great Question: A Manufacturing Podcast, Smart Industry's Sarah Mattalian and Nancy Majure, the head of product for Adaptive ERP at QAD, explore how manufacturers can scale AI without adding headcount by improving real-time visibility, data quality, and workflow integration.

Below is an excerpt from the podcast:

Sarah Mattalian: Hello, everyone, and welcome to another episode of Great Question: A Manufacturing Podcast, this one by Smart Industry. My name is Sarah Mattalian, and I'm a staff writer at SI. And I'm joined for this episode by Nancy Majure, the head of product for Adaptive ERP at QAD, a global enterprise software company that specializes in cloud-based applications for manufacturers across industries. As manufacturers across verticals continue to adopt AI agents in company functions, many struggle with scaling their AI-driven operations. While there are a handful of reasons why manufacturers struggle to scale and successfully implement AI, issues mainly arise in data collection and scaling AI agents for factory floor deployment. However, some manufacturers are scaling AI without adding headcount. They are able to scale based on frontline results, not pilot data. Nancy Majure is here to tell us what headcount neutral scaling looks like in a live manufacturing deployment, the architectural decisions that make AI scale without adding operational overhead, and where manufacturers get this wrong when they try to scale AI without the right ERP foundation. But first, a little about Nancy. Nancy has spent more than three decades in manufacturing technology, starting her career in the early 90s and working across the industry since. Today, she leads product for QAD's adaptive ERP, where her focus is on how AI can live inside the systems manufacturers already trust rather than alongside them. She's also written recently about the risks of shadow AI and bolt-on tools in the ERP, which we'll get into today.

NM: Hi, Sarah. Thank you so much for having me today.

SM: Yeah, thank you so much for being here. We are at the QAD Champions of Manufacturing Conference in Chicago today.

NM: It's a big event. We have quite a few customers and prospects as well as analysts and of course our partners. So this is a big event for us. This is our annual, let's show the proof of what we've said that we would do.

SM: Yeah, for sure. Can you just start off by telling me a little bit about your career and how you started in this industry?

NM: Yeah, you know, it's an interesting story. I started out in the early 90s. Actually, my education was a very good liberal arts education. And so I graduated with a degree in English and two minors, one in psychology and sociology. I was very quickly brought to reality the fact that I was starving to death. So I started working at a brand new greenfield manufacturing plant. And at that time people had to do data entry. And so I came in just right at the bottom doing data entry, and it was one of those indications of how business was at that time. There was no automatic recording of data. And even that early in my career, I started seeing the potential, the potential for if we had this data amalgamated, the brain trust that would sit around the table at the production meetings And they were able to evaluate trade-offs and make good decisions, but it was all tribal knowledge in their head. And so I've been very fortunate to work for QAD now for 22 years. And manufacturing is my passion. You talk to my customers, you talk to my colleagues and to our partners. And this is one of the most exciting times I can think of in my career. to be in manufacturing.

SM: Well, thank you so much for being here today to kind of talk about that, especially with data and your experience doing that. And as AI is being scaled in manufacturing, that's super relevant. So I kind of wanted to start off by asking how you define headcount neutral scaling and kind of describe to me what that looks like in manufacturing settings.

NM: That's a really good question. A lot of conversations these days are about headcount and we hear a lot about the lack of skilled labor, for example, or the difficulty in retaining people. But this concept of headcount neutral scaling, it essentially means that a manufacturing plant can take on more volume, more complexity, perhaps even add on a new customer program without having to go back to that old paradigm If I'm going to produce more, I have to add more headcount. And so what's the reason why that's finally possible? This headcount neutral scaling and I think it boils down to is that it's finally possible to have real time visibility. So if you think about it, when your frontline workers can see line performance issues or potential downtime problems or quality issues as they happen, and we're not talking about in a report that comes three days later, we're talking about right in the moment, they're not only capable of resolving that problem before it becomes a greater problem, but then with the tools that we have available, the AI embedded in their workflow can begin to learn about how those decisions are made and become better and better and more proactive. And so I want to make sure that we're not emphasizing that it's about doing more with fewer people by just putting greater pressure on them. Instead, it's about giving those people you already have the proper information, the proper tools, and that powerful dissemination of what's happening to make recommendations so that the plant can run closer to fully optimized capacity.

SM: Thank you for explaining that. Can we get into a little bit more about what this scaling requires from manufacturers architecturally and what decisions go into scaling AI specifically without operational overhead?

NM: Yeah, that's I think that's a problem that has faced a lot of manufacturers I've talked to because we all know there's a lot of hype out there about AI. It seems to be the panacea. You can just throw it on top of anything. It's going to work. It's going to solve all of your problems. But there are very important. architectural decisions that need to go into a manufacturer really being able to adopt and get benefit out of this. And I think one of the most critical things is data. Data absolutely has to be captured at the point of work. So whether it's at the line level or at the shift level or at the moment that an invoice comes in in an e-mail, Getting that data in real time, not trying to reconstruct it later from some spreadsheets, is that foundational architectural decision. So you get that real time data. But the second piece of this is that for scaling of AI to actually be effective, that AI has to be a part of every individual's daily workflow. So if it's sitting adjacently and someone has to remember to go look at a dashboard or look at some analysis, you're not gonna get that efficiency that you would that as I'm just doing my job, my AI companion is popping up and saying, hey, I'm perceiving a problem. There's a disruption here. There's something to deal with. So there's the data. There's the ensuring that AI is actually just part of their job, not something adjacent to it. And then finally, you really want those purpose built role specific agents that are it's different from having a generalist AI that's just trying to do everything. If you think about it like this, an agent that's specifically tuned for aligned leaders decision making. is going to be very different than an agent that's tuned for, let's say, a scheduler or a buyer or a planner. Each of those roles have different objectives, they have different KPIs, and they have different business logic that goes into their decision making. And so if you are able to get these three things right, getting your data in real time, embedding the AI so it's just part of the job. And then third, having those specifically tuned agents that work alongside and are the partners or the assistants of these roles, all of a sudden what you're gonna find is that AI in the concept of scaling it, that overhead is, you're not having to deal with extra systems, you're not having to employ extra headcount to run those. Instead, it becomes just a natural extension of how the plant operates already.

SM: And I'm curious kind of how ERP plays a role in all of this. So why is having the correct ERP foundation important, and what happens if manufacturers try to scale AI without the right ERP foundation?

NM: Wow, there's a lot packed into that. I've been saying for years, we've been saying for years, ERP is not going to go away. ERP is that foundation. It's that system of record. It is that single version of the truth. And it's that place where your production data, your quality specifications, and your financials, they all agree. They're all reconciled together. And so I think it's important to note that AI absolutely will not fix a shaky foundation. In fact, it will amplify it. At machine speed, it will amplify it. And so that ERP is that foundational deterministic system that really puts those guardrails in place to ensure that you're not putting your operations at risk.

So the most common pitfall is essentially I see manufacturers bolting these kind of generalist third party AI systems on top of their ERP instead of building into it. And what that does is that they are, first of all, they're not thinking about the importance of the governance. So ERP by its very nature is intended to be deterministic. So always, I put in A, I'm going to always get out B. There have been decades of work that has gone into building out the business logic and the governance and the controls and the rules around what happens within your business. And so since ERP is deterministic, a lot of generalist AI is going to provide you those probabilistic responses. And so it's going to give you probably the most likely answer as opposed to what would be the right answer. So when that AI is sitting outside of your governed environment, it doesn't take much, a single bad signal, hallucinated potential spike in demand that can trigger a snowball effect that can actually be very damaging to a business. triggering over purchasing, triggering overstock. And so that first pitfall is trying to think that you can get outside of the bond, the boundaries of an ERP deterministic system.

But the second one, and I think it's all too common, especially manufacturers that have been around for a while is fragmentation. And you'll see it so often in companies, especially where they don't have a solid ERP single version of the truth. You have production data in one system, quality in another, maintenance in another, perhaps you have your financials someplace else. And the reality of their current life without AI is that they spend a lot of time and a lot of effort in reconciliation between those systems. They do not always agree. What happens is if you then try to scale AI on top of all of that, it's literally just going to scale the disagreements between the systems.

SM: Wow, that seems like not a situation that manufacturers want to be in ever.

NM: No, definitely. I quite frequently say that if you would not trust an unsupervised intern to execute a particular transaction or to stay within your financial controls, you absolutely should not trust an unvetted third party AI to do it either.

SM: And if I could just ask another follow up, you mentioned governance, which that's a really big term right now. Can you kind of expand on why having governance in AI that you're using is important. And again, kind of what happens if you don't have that?

NM: Oh yeah, it's and especially in regulated industries. So, you know, for example, if we talk about food and beverage, they're really facing this FSMA 204 traceability requirements. And in a 24 hour period, they have to be able to turn around and provide full traceability in the event of a recall. Similar requirements exist in the life sciences world. They have 21 CFR part 11, which is all about the audit trail capabilities. What happens with a bolt on AI that sits outside of your governance model is that it literally breaks that accountability cycle. And that chain of accountability is certainly is something that regulators always expect.