Physna (Columbus, OH) is a company dedicated to helping manufacturers leverage their entire part strategy. If that sounds a little fuzzy, think of this: Physna can eradicate duplicated parts, it can be the tool you need to standardize your parts (even across different assemblies or products), it can be the doorway to DFM (design for manufacturing), and it can bring the power of 3D representation to every team who can use the information—increasing speed to bid, even .
As Physna CEO Paul Powers points out, the savings can pile up very quickly, well into six figures for some manufacturers. Those are just the direct savings; there are more downstream from these adjustments. Powers introduces us to the company (it has an interesting origin!) and the strength of the software and processes. While Fifth Wave mostly concentrates on machine-based intelligence, or machine tool networking, this is an exceptional way to streamline your part making and so it is included in our coverage.
Fifth Wave Manufacturing: Hello friends. We’re here today with Paul Powers, who is the CEO of Physna. And Physna is all about the data behind part-making and taking a lot of the friction out of making a part. First, welcome, Paul. I appreciate the opportunity to speak with you. I would like to hear from you, please set us up with where people are, what their process is and what you intend to change about that to help them.
Paul Powers: What we do at Physna—short for Physical DNA, which is why it’s pronounced “fizna” and not “fiesna”—is take information about three-dimensional models, like CAD models or you name it, and we turn that into normalized code. We allow you to do the same thing with parts that you can do with text. To put it simply, we act as the bridge between the physical and the digital for AI machine learning. But we’re also used to do things that are very straightforward, like querying a part. The way that people use Physna in the manufacturing context is all the way from the design phase, where they’ll use Physna to essentially act like an autofill.
They start to design something and we tell them, “Hey, it’s been designed before. You don’t need to redo that.” All the way up to identification in the field and identifying alternative parts, alternative suppliers and determining manufacturability. Let’s say I see something, I’m not sure what it is. I can ask Physna, even with an image, what this part is, do I have it in inventory maybe under a different part number? Do companies around me sell that part, even if they call it something totally different? If both of those would know, well, who can manufacture this, right? And how would they manufacture? Who would be a good option to contact to build this thing?
There are other use cases as well, but most things kind of center around that. The only thing I would mention that’s important is that Physna can take large data sets both not only from companies themselves, but also their supplier networks. It can analyze that and figure out their opportunity to standardize and reuse parts they have.
One of the main things we find when customers adopt Physna is that their overall costs go down dramatically. We have multiple examples of our customers saving hundreds of millions of dollars a year or two after implementing Physna because of just the massive reduction in inventory that they can execute. It’s not just about saving money and the acquisition of the parts, but also maintaining upkeep, maintaining operational readiness if you’re in the military for example, or if it’s you’re running a factory making sure that your machines are up and that it’s quick to repair.
FWM: That’s wonderful. In fact, it touches on the big trend of design for manufacturing. Let’s say you’re looking for a part, and you have four different ones that essentially do the same thing. The system could say that if you designed a new one that took advantage of these four capabilities, you could save a whole lot of money across different parts or different products.
PP: Exactly. Unfortunately, before Physna, there wasn’t really a way to do that with 3D data. And if you try training off 2D data, like pictures and drawings alone, you don’t get as much information out of them, because you’re missing literally an entire other dimension. They are not as data or information rich. Every megabyte of data that we have at Physna produces more information than what you can actually extract from a 2D image. We can analyze 2D images and tell you exactly what that part is. And some people get lost when I say that. Most people who use Physna don’t even know that we’re using 3D. We use it for training to accelerate AI and machine learning’s ability to understand the world around us.
If your input is a part number, like just text, or if it’s an image because we’ve trained on 3D, which is a much better proxy to reality than part numbers or pictures alone are, we can understand those part numbers and those pictures much better. To your point, not only can you understand how to go about manufacturing something in a cost saving way, we have people who are on the receiving end of projects, and they’ll use Physna to do things like quoting automation. Somebody will send them a part and say, how much would it be to manufacture this? And look at the data about past designs they’ve done, past work they’ve done, that will analyze all the different features present on those models and tell them this is how much you should charge for it. For custom manufacturers, this is a really big deal. Normally speed to bid is one of the biggest determinations for who will succeed.
FWM: I don’t know if I’ve ever had an interview that centered in so quickly on who exactly my audience is, but you did. That is exactly what my readers think about every day. We have customers who want us to design this thing, and many of them have design departments and innovation departments, and I guess those would point to a good landing pad for Physna.
PP: Absolutely. We find that typically for those custom fabricators, where we come into play is one, quoting automation; two, finding ways to manufacture things at a lower cost; and three, preventing yourself from having issues down the road. Let’s say that you’re designing an engine, or you’re manufacturing a component of an engine. What if I can tell you there are many functionally identical alternatives you can use, you can buy?
It’s best when people make these decisions sooner, not later when there’s already a problem. Maybe they chose a part, and it is no longer in production, or it’s ordered from overseas and there is a problem with compliance, or it is from a manufacturer that shut down. We will tell you the companies that sell it. If the number isn’t many and there are few alternatives, then we quickly identify if there is a part that will be safer and cheaper for you.
These are examples of how we are used by our customers. To your point, that is exactly who we work with.
FWM: How many SKUs do you handle, or do research on? I imagine a lot.
PP: We’ve processed over a hundred million parts here. I mean we have a database that has millions of parts. For people to access on demand, we build out custom databases for people from suppliers all the time. Our customers very often have their own databases, and those can be massive as well. We don’t like doing training using our customers’ parts in the way that feeds Physna back information. Any training that you would do on your own parts stays with you. That’s important for many IP reasons.
Google is one of our investors, Google Ventures, and they invested around 2022 after we showed them what was possible. There was an article that came out from Google DeepMind that said it’s not possible to generate true functional, manufacturing-ready and dimensionally 100% accurate 3D models. With generative AI, you can’t do that because the internal geometry isn’t there. The thought was, there aren’t enough 3D models in the world to train generative AI in the way that you would really want to.
We contacted them told them it wasn’t true; they were just looking at 3D wrong. If you ingest 3D and just say, here are the different file formats, all the different kernels, go ahead and generate something out of this then they’re right. AI has no idea what it’s looking at. Even if it can see 3D, it doesn’t know how those parts relate to each other. It would be like trying to clone someone or trying to create something in a lab, all based on pictures versus actual DNA. You could do it with DNA, in theory. But with pictures you can’t; it’s just too little data. When you break models down, when you break 3D data down, when you break parts down mathematically, so that every feature, every characteristic is represented in this complex matrix of code, it’s deterministic. It’s always the same no matter how it’s generated. The geometry is the geometry.
As far as the material properties, it doesn’t matter if they come in a different format. What that allows you to do is train at an exponentially faster rate. We showed them that you could create something like generative AI for 3D with 8,000 3D models. We also showed them a public dataset that had way more than that. We just took 8,000 models, showed them a very small data set. That’s where people struggled. They didn’t really understand that if you make it as easy for software and computers to understand physical things as it is for them to understand text, it opens a new world of possibilities.
FWM: You mention generative AI. I would think that with a dataset like that, there would also be an advantage of having an agent, and use agentic AI?
PP: First, a lot of people get confused about Physna because we’re very close to AI. But what Physna does at the beginning isn’t truly AI. And that’s very important, because any time you use AI, you introduce the risk of hallucination. If your AI makes a mistake, if you’re using AI to generate your data, that data is going to be inherently flawed.
When we generate data about parts we do it deterministically, meaning it’s, it’s an equation, it’s an algorithm. It’s the same every time. And that’s actually the really hard part about our business. From there, you can leverage that in AI, which is how you get to things like generative or analytical AI. Yes, in theory we could leverage all of the data that we’ve gathered for generative AI.
What we found is more important and more impactful for these companies is because of how information-rich their 3D models are, and their parts are, you don’t need to train from the entire world to create a generative AI tool. In fact, there’s a downside to it. If I’m learning from a class of models, and I’m trying to generate a part for an aerospace company, I don’t want the influence of other types of data are here because it adds risk. The risk is something being a flaw because it came from different data in the first place. It would be better to train from that company’s own data, if they have a broad enough dataset. If not, you can always augment it with a little bit of external data, but try to make it focused on that area, that industry. When they generate something, it’s much more likely to be appropriate. If you’re in aerospace and you want a brake pad to be generated, it’s very different than if you’re an automotive company and want a brake pad generated. There are different standards, different requirements, and it’s important to know that ahead of time.
FWM: Yes. On the other hand, how many times have your customers, or you said, that’s not what we expected, but it might just work. Are there times where things are brought in from influences you don’t immediately recognize, and yet it’s a beautiful solution to what’s at hand?
PP: I should be clear that we’ve been talking about generative because that’s what led Google to invest; we proved that you could use it for that. We don’t actually focus on generative here. We do have a couple of customers that are having us do some work in generative for them. Other than doing a project in the past showing that what you can do with Physna, it’s not something that we have yet actively delivered for anybody. What is always shocking, though, is on the analytical side, and how much opportunity there is for cost savings. It’s mind blowing. I can’t tell you how many companies have told us that they didn’t have a problem, and then a year later turned around and said, “Hey, you saved us $800 million this year.” The problem is that massive.
In the commercial sector, broadly speaking, you can expect probably around 40% of your procurement budget to be wasted. And that’s what we normally see—about 40% is waste. I’ve seen it north of 80% or 90%. I’ve seen it lower, in the twenties or thirties, but I’ve never seen a case where there was not a problem. I think most people assume that their waste is in the single digits. I can’t think of a case where we’ve seen it that low. The reality is that every company we’ve dealt with that has more than a handful of parts has an issue of standardization and duplication interchangeability that just comes with the territory. That’s just because that’s how it’s been forever.
Beyond that, if you don’t have enough information on sourcing—if you don’t know how many places you can order it from—you’re paying different prices which means you’re overpaying and you have multiple points of failure. That results in downtime. But even if you don’t include the cost of downtime if you think about it, like 30% of your parts have duplicates, and on average there’s about 30 or 40 duplicates per part. You’re overspending on those parts by many, many fold at the end of the day. Even if you have only a small percentage of your parts have duplicates, if the problem’s bad enough it ends up accounting for about 40% of your procurement budget on average. That’s what shocks me all the time, just how bad the problems can be sometimes. And most companies don’t have any idea that that’s going on.
FWM: So they’re not looking at a solution that’s coming from another industry, but they will still have that reaction of, oh my gosh, look what we can do when it’s something that is simple, easy to integrate and meets all of the criteria, and yet you only buy one type of those things and run that across products. Maybe they’ll use it in who knows how many manufacturing processes. So that is where your savings come from. I’m sure they would still have a very surprised look on their face after saving that much.
PP: Absolutely. It’s shocking. It’s not as disruptive as you would think, either. I would say most of the time what we find about their solution is that they already have the part in inventory, they just didn’t know it. It saves them the hassle and the money of going out and buying the part very often. I don’t have an exact percentage on that, but it’s extremely common to see that they already have the part in stock.
FWM: Wow. That happens not just in manufacturing. It happens in software and in many business functions. You and Physna introduce pipelining things, streamlining things, and use that to create solutions.
PP: Absolutely. The idea is to bring the physical world and manufacturing space up to the 21st century. Without that data being normalized, it’s barely in the 20th century. We are still reliant on what people call parts, and we don’t understand how they relate physically except what our eyes pick up. CAD tools are very limited in this regard. We have introduced a way for multiple industries, not just manufacturing, to reap the benefits from information that is the real world. If you are going to train software to do a good job at analyzing or composing a book, you want to train using books, and you’ll do the same thing with images.
The closest proxy that we have for physical items are the 3D models of those items, or 3D scans of those items. These models and scans capture as much about the items as possible. When you can normalize that and make it just as easy for software to understand something physical in your hand, suddenly capabilities that don’t seem very intuitive or seem like science fiction become easy to implement.
People are coming up with a lot of other applications with Physna. It’s exciting. They’re using our APIs to do it. Other software companies are finding ways to implement solutions in areas like healthcare, for example, or defense.
FWM: You mentioned the CAD files on all these things. They can be large things to manipulate. I’m assuming it’s still kind of vector driven.
PP: It depends on what tool you’re using, but yes, I understand what you’re getting at. There is a steep learning curve to CAD. We’re a company of very nerdy people who love talking about how the technology works; we’re always saying things about 3D and the analytics. What most people really care about isn’t how it’s done, it’s what it does. What I tell people is that you don’t need to know how to do anything with 3D to use Physna. Over 90% of the people who use Physna don’t have a CAD license and don’t even know that anything 3D is happening.
They’re using an image or they’re using the results which might be 3D models, but they’re not opening them up and trying to manipulate them. They’re just saying, okay, tell me more about this. Tell me how these parts relate. They’re just thinking of it in terms of parts. Even in the cases where you do want to see how things are similar or different, Physna has a 3D viewer. It’s intuitive, and it will immediately highlight the area that’s relevant to you. You don’t have to understand how to manipulate a part or any controls—there are very few controls. You just push and move your finger and say, this is what I want to look at. We intended to make things straightforward because we realize that most people don’t want to learn how to deal with 3D. Even the people who open that comparison viewer to see it in 3D, it’s rare for them to use any kind of 3D controls. They’re just looking at it as if it’s an image.
FWM: Let’s say I’m in the aerospace industry, looking for a part, a brake part. As we know, weight is of the essence. And a lot of times people save weight by putting holes in parts or assemblies. Often it also means you have gained strength, too. Will your system take those things into consideration?
PP: It can, for sure. It will know what happens if those changes have been made. It will know that the holes were drilled. It can learn from other parts in the past. It can notice that the strength went up and the weight went down under certain changes. The nice thing about Physna is that you can easily connect to the insights that we have about these models and use that in any tool you want to use, like ChatGPT or an AI agent. Also, we work closely with Palantir, AWS, Google and many others.
If your company has an AI agent, you can ask Physna to tell the agent everything it knows. The agent looks at the Physna data, which is deeper than the data that we ingest. We get the basic stuff that you already have. You might want Physna to help you learn from supplier models, and you do not have a 3D database. That’s fine, you don’t need one. But we’ll share everything that we’ve learned from the parts that we have access to in your system. Then Physna can start to make predictions and give advice. For example, for your next design, you should drill holes here, here, and here. It does not require much coding, if any. You are just connecting two different systems and one system is getting data from the other.
FWM: Let’s say I’m a fabricator and Physna sounds interesting. How do I go about exploring whether I am a good candidate for this or not?
PP: The best way to do that is just to reach out to us. You can request a demo or a trial, or both. I would recommend a quick demo call. It’s short, it’s fast and it’s live. Someone is answering your questions, and you can keep it as short as you want. But the demos are the best because if you tell them what you’re interested in, they can go deep and answer the questions and show you exactly how we can use that. You know, you can use Physna for your use case. A trial’s fine too; the only thing is you must get approved for a trial. In any case, you can reach us at www.physna.com or reach out to one of us via LinkedIn.
FWM: Let’s say I find out that my company is a good candidate for this. Do I need to install anything?
PP: We try to make it flexible. The most common implementation is Saas (Software as a Service). You do not need to do any major installations of anything. If you want to upload data to it, that’s very straightforward. You can do that through the interface, or we can integrate it into some other system. If you want that to be synced up at the same time, that’s fine. The other thing that you can do is use our APIs. Everything we have at Physna has APIs. Even if you’re not very technical, we can work with your team or even help you out directly help you integrate a particular functionality into whatever system you’re using.
FWM: Let’s go back to our engine, and I need to create the head cap. Where would I start that process?
PP: Do you already have the part, or do you have a drawing, or something else?
FWM: I don’t have the part. I do have a drawing. I also have a big array of materials for the part that I could draw upon. Tell me what would happen on your end.
PP: If you’re just missing the part, you would push one button which would be “match.” Let’s say that on that engine that you’re designing, where the cap is missing, if we’ve seen that section that has a missing cap anywhere else, any data we can access on your behalf, we would immediately go find it.
You could be halfway done designing your engine, and it’s half an engine, let’s say. I know that’s not how it’s done, but let’s just say it was right. We could find other things out there that represent parts for half an engine. There is no guesswork in it. You can see a color overlay of what was found vs. what you have, showing what the parts have in common, and what’s missing. You can decide whether it’s what you wanted, or close, or not close enough. Maybe it’s close, and you have to make a small modification to make it fit your needs, rather than having to start from scratch.
FWM: What is it that the person gets delivered to them?
PP: It would be basically a search result. The system looks at the database and finds things that are similar to the missing cap. Only the search results that make sense are delivered. You can control that too by searching only for that cap or maybe for a small section of the engine. If there are valid results out there, you will see those, almost like a Google result, but with images. If you click on them, you’ll see comparison of the part vs. your design. You’ll see if you have a fit or if you need to modify it.
FWM: You could source that part, or you could redesign that part, then?
PP: Sure, you can source that part. You have the data on how you could redesign it, but you’d probably be using another tool besides Physna. Physna is not a design tool. It helps people who design, but you’re not in Physna. It keeps the learning curve small and allows the designers to design in their preferred software. We are developing functionality that bypasses even the simple process I described. It’s a couple of clicks. You can tell the system that you want a cap that covers this part. It will be designed for you; you won’t have to do anything.
FWM: Oh, I see. Wow. That’s kind of the holy grail.
PP: That’s the goal. And that’s why we are really focused on the type of data that we’re working with and the methodology that we have. We make it possible to generate that specific section of data, and to have dimensional accuracy. It needs to be correct. When it’s generated, it can’t just be almost right because a tenth of an inch or a tenth of a millimeter can be enough to cause big problems. You cannot take the same risk that you can take in letting ChatGPT do your homework <laugh>. The approach we take is getting so much attention.
FWM: Again, you’re spot on.
PP: While I don’t have a lot of direct experience in the manufacturing space, I have a lot of customers who do. I find it fascinating to listen to them talk about their issues, and they’ve really taught us how this product helps them. It’s been fascinating journey.
This product started off as a tool for intellectual property protection—identification of stolen models and things like that. The bar for that was so high and so complicated. When I started taking it to market, I found out that there were some needs and some interest in intellectual property protection. There were many more needs elsewhere. Intellectual property is great, but a first-year company must be economically viable for that to matter. Everybody at our company supports and promotes American manufacturing. The idea that we could help that industry revitalize itself was also appealing at an emotional level as well.
FWM: I agree with you 100%. I do feel like things are really coming back. All the outside reshoring efforts…I don’t know how much more reshoring help we need because we’re proving we can do a great job in manufacturing.
PP: I think so too, and I think people need to be aware that it’s possible to have things manufactured in the U.S. Some of our larger customers, in automotive, or aerospace, or military, they are often surprised at how many parts they can source from the U.S. For a long time, the situation was that you could only get this part from country X, or maybe only one company in the world makes it. There are exceptions, but for the most part, U.S. companies manufacture parts they need.
One thing I love about being able to work with some of these larger enterprises and the military is that we get to showcase that and tell them there is a company down the road that is American owned and operated, and they can make that part for you tomorrow instead of you waiting for it for six weeks from overseas. Now we’ve brought more business back to the U.S. and the military, getting that part faster means that that plane, submarine, tank, whatever it is, is operationally ready sooner. That’s important because if you have all the planes in the world and they can’t fly, it doesn’t mean much.
Modern manufacturing methods allow for that. The newer machines make it increasingly affordable to get into water jets, laser cutting, CNC machines and even 3D printing. The nice thing about those new tools, aside from the fact that they’re coming down in price, is the ease of changing from one job to the next. You don’t have a giant factory for one part, and then it takes three weeks to switch it over to make a different part. It’s you just push a button and it goes, and, and the machines are making something different. That’s what enables what you described—that mass customization or semi-custom parts. With those types of machines, it’s a different economy and it is not like the problems with ordering parts from China.
The reason why we make so much stuff overseas is because it has been cheaper. That can only hold true for so long because wages go up and you still have this one massive issue, which is you have the time, fuel, and expense of transporting all that stuff all the way across the other side of the world. If you can manufacture things nearby, you get faster turnaround times and you’re of course supporting the economy. You will also pay less money along the way, particularly if you need a smaller order, say 1,000 versus 500,000. At some point it makes less sense to go overseas and more sense to stay local.
FWM: I agree with you, and there is no Suez Canal to get blocked. There are probably 80% of fabricators or so, with 20 people or less. Now they’re doing a lot more work than they could have with 20 people in 1995. It is probably 3-, 4-, 5-fold what they were doing. Because of all the automation, what earmarks are you looking for to roll out your technology? Is it one size fits all? Is it a two-man shop to a 500-person shop?
PP: Well, we traditionally served larger companies and military providers because of how we had to architect everything. It was required to do business with the customers that we had to do everything in this highly regulated, single tenant environment space. To spin up an environment for a customer would cost a fortune. We couldn’t charge a small enough fee for a small company to afford us. Now that’s changing, though, because we found another way to set up environments that are multi-tenant. It costs us a fraction of what it would cost to set up one of those big, dedicated environments. There is no real need for that unless you have an IT team that’s making you do that. Now we can work with smaller companies. We have companies that are pretty small—20 people and under—as customers already. I think they’re some of the best customers to work with. And so absolutely we work with them today.
FWM: I think those are all my questions. If you have anything else you would want to make sure that our folks know, please tell us and or we can get back together again another time. Anything else you’d like to add to what we talked about?
PP: No just thank you, since your audience is heavily U.S. manufacturers, and I’d like to thank them all for what they do. I think they’re the backbone of our future economy here. If you’re interested in working with us, we would love to hear from you. It doesn’t matter if you’re a one-man shop or a Fortune 50 company. We support all great American manufacturers. Thank you.
FWM: Thank you, Paul. I appreciate your time today. We will keep in touch because I want to see how this story continues to unfold. And thank you for being our guest today.
PP: Thank you, I appreciate it.
More information: www.physna.com