The Expert’s Edge: Leveraging AI & Automation to Scale Manufacturing & Cost Intelligence

Lily Thomas:

Many of you do know me. I’ve worked with a number of you. My name’s Lily. I’m a Principal Consultant at aPriori. I’ve been with aPriori for going on six years, primarily helping our customers in the sourcing function negotiate with suppliers and streamline RFQ processes. But I’ve also used aPriori as a learning tool, like John Anthony was saying, to bone up on design engineering, and I help design engineers as well. Before that, I worked in supply chain and commodity management for GE for seven years before joining aPriori.

As you’ve heard today, the customers that we work with who have a cost engineering function or other experts within their organization, senior engineers, manufacturing engineers like Chad talked about, are a huge asset with massive experience. But their time is finite. We experience the same thing at aPriori. We are experts to our customers. I play a cost engineer for some of my customers, and my time is finite. I can’t be on a phone call with every CNH commodity manager for an hour explaining all the assumptions behind the aPriori estimate. Dakota said the same thing. She can’t be explaining the estimate and having two or three phone calls for every analysis that she’s working on. Our resources are finite. Chad said the same thing. They can’t review every single design. John Anthony said the same thing. They can’t review every single design. What organizations want is to scale the insights of those experts because no one’s handing out money to go hire a bunch more experts, and they’re hard to find anyway. So we need to figure out how to scale the ones that we have. We need to put their knowledge in the hands of the people who need it, when they need it in their process, in a format that they can understand. Then they can be more effective and get more results for their organization.

I’m going to highlight not only the content that we heard today, but also articles and studies from the Hackett Group, which is a management consulting firm focusing on digital transformation and AI, and also CIMdata, which is a consulting group focusing on PLM. Ivalua worked with the Hackett Group on a survey that surveyed procurement. Fifty-five percent were in the manufacturing industry, and the rest were services. But what I’m going to highlight I’ve also heard anecdotally from our customers, so I’m confident that it’s going to resonate and be a challenge that you’re seeing. In procurement, there is pressure to do more with less. Procurement’s workload is predicted to increase. They need to do more negotiations. They need to launch more products on time, but with actually a small decrease in headcount. Of course, companies are probably expecting AI to help fill that gap, and we’ll talk about that.

In my experience in supply chain and commodity management, fewer people means less time to spend on the strategic stuff, like getting the best price, developing suppliers, and building relationships, and more time spent on tactical stuff, like paying any price just to get the part in the door. The experts in the organization have insights that procurement people need in order to be more effective with fewer people. But they need those insights at certain points in their process, in a format that they can understand. Based on our experience across customers, we see that these experts are trying to support around 100 design engineers to one cost engineer. We’ve seen it as high as 200 design engineers. The same group of cost engineers is also trying to support maybe 50 commodity managers at the same time. The only reason it’s fewer commodity managers is that, for a given P&L or a given product, there’s usually more designers than there are commodity managers. So we have a few experts trying to support a lot of people, which is exactly what we heard from Chad, Dakota, and John in the presentations earlier.

But the problem is not just the numbers. The problem is also that the experience gap is growing between the people they’re trying to support because, as someone referenced in a presentation earlier, in the manufacturing industry, a lot of people are retiring. That experience is walking out the door. This is from the U.S. Census from 2020. The article is about why manufacturing wages are going down. They said, “Well, it’s because experienced people are retiring, and then the people that they’re hiring are recent grads who command a lower salary.” So that’s causing wages to be lower and the experience gap to be big. The new engineers that the senior engineers are trying to support are becoming less experienced, and the commodity managers that the cost engineers are trying to support are also becoming less experienced.

This is what the Hackett Group found. We worked with the Hackett Group on an article, and they found that for sourcing, supplier negotiation is still the number one savings lever. It is so much cheaper to negotiate with an incumbent supplier than to go and qualify a new supplier. So negotiation with the incumbent is the number one savings lever. But the third finding is that skills and expertise are what the Hackett Group found to be the number one barrier to effective category management. So you don’t just have to take my word for it that the experience gap is growing. They found the same thing.

CIMdata, which specializes in PLM, also found that when it comes to design reviews, companies are still relying on tribal knowledge, so the senior engineers, and spreadsheets. One of our cost modelers was talking to a customer, and the customer has a spreadsheet that only one person knows how to run to generate estimates for certain types of forgings. I’ll never forget when I was in a workshop with design engineers. I mentioned aPriori’s capabilities. I’ve learned enough that I can still support design engineers. We were looking at slots on a machined part that aPriori identified had a high length-to-diameter ratio and would be difficult to manufacture. The junior engineer said, “Oh, yes. Actually, the supplier emailed me with a picture of the chatter in the slots and asked me to redesign the slots.” The senior engineer was also on the call, and he very nicely explained, “Yes, those slots will be difficult to make. Here’s why.” He caught it right away. He knew what aPriori knew, but the junior engineer didn’t. The senior engineer doesn’t have the bandwidth to review all of their drawings. That’s what you’ve heard a lot about today.

I think you’re already convinced that there is an experience gap. There’s a limited number of experts in the organization, and one-on-one time with an expert to support the high number of people in the organization who could benefit from their insights is not feasible. Our customers, and you’ve seen some of this today, are solving this with AI and automation. I’m going to give examples of what I mean by AI and automation because they’re very general terms.

Let’s talk about automation first. Let’s say you’re someone who’s been using spreadsheet models. Those of you who aren’t aPriori customers yet or recently started using aPriori might have been using spreadsheet models to generate cost estimates. Investing in aPriori, where you’re getting automatic geometry extraction from the CAD model, is a big step up. If you’ve got tolerances or PMI on the CAD model, we can extract that too. You can get DFM feedback and cost in minutes. We kind of take that automation for granted at aPriori because it’s our bread and butter. But it is a huge step up if you’ve been relying on spreadsheets or purely tribal knowledge to generate that. That’s the first level of automation that customers are using to get more DFM insights. Like John Anthony was saying, they are using it so that their more junior engineers can get that feedback directly from the 3D model.

Then we have bulk costing. You can upload many models at a time from a spreadsheet, and it will generate the initial cost estimate. Then we have matrix costing. You can propagate that estimate out to different regions and make data-based regional decisions, which Evan touched on. Finally, we have a level three of automation where things are kicked off and initial estimates are generated automatically from a trigger in the PLM system, like CNH is using, or from a web app like John Anthony referred to. We have customers where a sourcing professional can go and upload part numbers to a web app. It will retrieve the models. Then the sourcing person can either tell it the regions and the volumes that they want the analysis done in, and aPriori will initiate the analysis. The person doesn’t log into aPriori at all. Or I have some customers who are retrieving the volume data, the current pricing data, and the region data from the ERP system, retrieving the model as well, and then generating the initial analysis.

I’m saying initial analysis because, in many cases, we do still have experts refining the analysis. They’re using automation to do the grunt work and get an initial estimate. They walk in and they’ve got a roll-up in aPriori Professional that they can go and edit. Or at least for some commodities, they may know, “I want to look at these for some basic sheet metal parts or injection-molded parts. I know I can just send those downstream.” All of these three levels of automation are driven by aPriori Professional.

I think this is key. When experts leverage automation to give themselves more time, one of the things they can spend that time on, and Chris was asking Dakota what she’d want to spend her time on, is maintaining and enhancing the digital factories, or all of the assumptions that aPriori is using to generate the should-costs. One of my colleagues likes to say, “You need to memorialize feedback that you’re getting from the field.” From the people who are taking the data to the suppliers, you need to memorialize their feedback into the digital factory. I think it’s really important that the experts in the organization invest their time there. Set up the infrastructure. Make sure the output is useful and relevant.

We did hear about customizing the design-for-manufacturability feedback. Chad is spending time building in new checks that are relevant to their organization, and then the engineers can self-service with the results. We heard John Anthony talk about using a web app, and then the designer gets an email with green, yellow, red. We heard John talk about designers getting an email. In the video, we’re showing data being written back to PLM, so the system that the designer works in has the data available there in a way they can understand. It’s got not just cost, but other critical information. Then there is a link to aPriori Design where they can go and dig in further if they need to.

This is the web app again. John Anthony was showing the web app. We have a number of customers using a web interface like this. The previous slide showed them triggering from a PLM system. They can also have someone upload part numbers. John Anthony talked about that. We have sourcing professionals uploading part numbers and requesting analyses that way. This screenshot shows that they specify the process group and material. But I have a customer where the material is on the model. The process group is from the model based on a commodity tag that’s metadata in the model. The volumes, regions, and current costs are pulled from the ERP system. All of that information goes into aPriori to generate the analysis. Then a report is sent downstream to the design engineer or sourcing. They have the ability to request a deep dive with an expert if they need to. In that way, we can free up some of the expert’s bandwidth so they’re not spending as much time generating estimates. They are refining estimates and refining digital factories, assumptions, and DFM checks.

I talked about the automation levels and how that can help increase the bandwidth of the experts. Then I referenced a couple of times that they’ll send the data downstream. They’ll send the data to design engineers or sourcing professionals. I mentioned that people need it when they need it, in a format that they can understand. For the “when they need it” part, automation can help with that because we hear from sourcing, “I don’t have time to wait for the cost estimate. Getting the should-cost back takes too long.” We need to try to automate and find ways that the experts can churn through estimates quicker and get high-quality data back to the end users faster. Automation can help with that. Especially if we can figure out which commodities need refinement from an expert and which can just go straight through to the end users.

But then the end users still need it in a format that they can understand. That’s where AI comes in. You guys have heard about aiSource and aiDesign, so I’m going to give an example of that. What I’ve experienced today, I’ve seen it in my customer base when I’m supporting cost engineers, and I’ve experienced it when I’m playing the role of cost engineer with a customer. Usually today, when an expert does an analysis, there is a minimum of one meeting with the person who requested the analysis to explain all of the assumptions. They want to know: Are the labor rates fully burdened? How does aPriori calculate the overhead rates? Did you include this process? How did it pick the routing? What’s the cycle time? How does it get that cycle time? Then usually they also want to refine it. “Well, I know the supplier, and I know they’re always complaining about high labor costs, so let’s increase the labor.” “This anodizing process is probably outsourced, so can you add some outsourcing charges on that?”

A couple of weeks ago, I spent an hour on the phone with one of our customers. They had a supplier breakdown. I had aPriori Professional open. It was an injection-molded part. They were just going through: “Supplier labor cost is this. What’s aPriori’s labor cost?” “Supplier direct overhead cost is this. What’s aPriori’s cost?” “Supplier machine is this. What’s aPriori’s machine?” I was just clicking, getting all the information for them. An hour later, we finally figured out that the supplier was using a bigger press and they had one and a half operators, whereas this customer’s digital factory was set up to have one operator to three machines. They didn’t need me to be sitting on the call pointing and clicking in aPriori Professional. That’s what a large language model can help with.

That’s really one of the value propositions of aiSource. Dakota doesn’t have to be on a call for an hour. Chad doesn’t have to be on a call for an hour with every person who’s trying to use aPriori. They can self-service by interacting with a large language model that has access to the aPriori scenario and is trained on aPriori’s assumptions. They can ask the LLM, “Are aPriori’s labor rates fully burdened? How does aPriori come up with the overhead rate?” They can also paste in a supplier cost breakdown. The LLM will do the adding and mapping of the aPriori data to the breakdown, flag where the gaps are, and then say, “Given where the gaps are, here’s what you should go ask the supplier, and here’s the tone and approach that you should take with them so you won’t piss them off.”

Right now, you can paste that into aiSource. Barton is working on a file upload so you can just import a PDF from the supplier. Then finally, in February, they will be able to ask aPriori to rerun the estimate with different assumptions. So they can not only query it to ask, “Are the labor rates fully burdened?” They could say, “Can you rerun the estimate with a labor rate that’s 20 percent higher, a different margin, a different material cost, and a different sheet size?” We can really free up a lot of the expert’s time by enabling people to self-service on, I would say, 50 to 60 percent of the stuff that they need.

Then they can use the collaboration functionality that John Anthony alluded to. They could tag the expert and say, “Hey, aiSource told me it doesn’t know the answer to this, so Dakota, can you tell me the answer?” There is a limit. It will tell you, “Check with your administrator,” or “Check with your expert. I’m not sure.”

Here’s an example where the user can put in feedback that they’re getting from the supplier. They can say, “Hey, the supplier’s using a four-by-ten sheet. What did aPriori assume?” They don’t have to know where to find the sheet size in aPriori. It’s just going to serve up what aPriori assumed as the sheet size. It’s going to tell them a little bit about what the impact of the supplier using a different sheet size is. I spend a lot of time with customers helping them vet supplier feedback. “The supplier said this sheet size.” “The supplier said this material rate.” “The supplier said this machine.” Okay, well, is that fair for both you and the supplier? Is that something we have to live with, or is that something we can negotiate?

aiSource can also give them tips. “Yeah, that seems reasonable. You should rerun the aPriori estimate to account for that.” Or, “Here’s how you could push back on the sheet size. Five-by-ten is a standard sheet. aPriori assumed it, and it’s standard available. You should see if they can get that sheet size instead. You should ask about their nesting assumption. You should ask what they’re doing with the scrap.”

What does this look like when enterprises are actually doing this? Well, you heard from CNH already. They are using the option of having the PLM trigger the analysis, and you heard how the design engineers are getting emails. We also have an example from Rivian. They use the web app. Before using aPriori, they were reliant on supplier quotes to get an understanding of their cost, and that was a problem for their R1 model. Then they invested in aPriori because they needed to get cost during the design phase to be able to drive cost out. They use the web app. They can upload a group of part numbers. They specify the process group and volume, and it goes and generates the estimate. The cost engineers come in. They’ve got a roll-up in aPriori Professional, a group of parts, and they can refine it and send it downstream as needed. Rizwan, their cost engineering manager, said, “We never would have gotten through this volume of work without aPriori.” For them, the automation really made them able to increase their throughput and generate the cost estimates.

Using automation and AI, we can increase throughput. We can handle more requests from more people. We can turn them around faster if we’re using automation. And we can handle more one-on-one requests when it’s warranted. If we can increase bandwidth for the experts and use AI to reduce the number of meetings they’re having to be in, they can join the meetings that really warrant their expertise. Because AI and automation can’t fully replace an expert. But if we can allow them to focus on the more critical stuff, maintaining and enhancing digital factories, supporting complex negotiations or design reviews, and reducing the number of meetings they’re spending on easy questions, while reducing the amount of time they’re spending on the grunt work of generating initial estimates, that’s where we get value.

In summary, sourcing professionals and design engineers will be more effective when they can leverage the insights from the experts. There are not enough experts to support them one on one. Our customers are using automation and AI to close the gap, and we are trying to support them right where they are.