Spencer Penn: 0:00

You might talk to someone who's like a chief procurement officer and they feel that they've implemented some tool or technology. But when you actually talk to folks in their team they'll do all the work over Excel and email. And only at the very end of the process when they've chosen the supplier, they've made the award, they've selected the price then they populate the ERP. And so the first time you really touch a digital system is when the battle is already won or lost.

Tom Raftery: 0:32

Procurement is often described as digitised, but in many organisations, the real sourcing decision still happens somewhere far messier across Excel, email, and institutional memory. Good morning, good afternoon, or good evening, wherever you are in the world. Welcome to episode 130 of Resilient Supply Chain Stories and Strategies that Keep Business moving. I'm your host, Tom Raftery. Today I am talking with Spencer Penn, the CEO and co-founder of Light Source, a direct procurement AI platform. Spencer has worked inside high pressure hardware and autonomous systems environments, so he brings a very practical view of what actually happens when engineering suppliers finance and procurement all collide. We talk about why ERP often arrives too late, how AI agents could shift procurement from reactive admin to proactive intelligence and why the real opportunity is not replacing procurement teams. It's helping them control decisions that can create or destroy hundreds of millions in value. And Spencer has a lovely analogy. He says Procurement is like running water. Nobody notices it when it works. However, everyone realises when it stops. Let's get into it. Spencer, welcome to the podcast. Would you like to introduce yourself?

Spencer Penn: 1:57

Sure Tom, thank you for having me. It's great to be here with you and the audience. My name's Spencer. I'm the CEO and co-founder of LightSource. We are a direct procurement AI platform headquartered in beautiful, sunny San Francisco, California. And by way of background, I am an ex Tesla person, and so I spent most of my pre LightSource career there and helped launch a programme called the Model 3. If I have one claim to fame, that's it.

Tom Raftery: 2:26

Okay. And having worked, you worked in Waymo as well, if I remember correctly. So having worked in Tesla and Waymo and now procurement software, what problem pulled you into this space?

Spencer Penn: 2:37

Yeah, absolutely. So I did work at Waymo. I was the product manager there for simulation, which couldn't be more different from direct materials procurement because we were literally working on virtual goods, I guess you could call it, which is a highly realistic video game that the car can drive around in so engineering teams can test new functionality before putting it on the road. And when I first moved to California, Tesla was a boutique car business. We were making a thousand vehicles per week. My friends and family told me it was a big career mistake to work at Tesla, that electric cars would never be someone's main vehicle, that they were a tech toy or an iPad with wheels on it. But I was very excited to be part of what I thought was a business that could have a meaningful impact on the environment. And I have a lot of reasons to believe that. When I was there, I helped lead a programme. As I mentioned, the Model 3 was our first mass market electric car. And we struggled with a huge amount. We were in a very fast paced NPI cycle. Elon had a two part or a dual mandate. One was to get the car out to market in half the time of a traditional OE three years instead of six. And then the second was we were gonna deliver the car for $35,000, which is just a huge amount of car for the money. Imagine replacing a, an empty gas tank with a an entire floorboard full of batteries. Those things shouldn't cost the same amount. There was a lot of challenges in running the programme. I think that Tesla's famously full of controversy and excitement. I would say it was the most fun I never want to have again. But the big pain point that really spoke to me was the availability of high quality technology to facilitate collaboration between engineering, procurement and suppliers. And I felt like there were sort of two systems of record we had in the business. We had the PLM, which is the engineering team system of record, and we had the ERP, which is the finance and payment system of record. And in between there was $30 billion of supplier RFX negotiation communication of drawings and iteration that was happening on spreadsheets and emails and just, just, it just drove me nuts. And so at a certain moment, after looking for software that we could buy, and in the market finding nothing that good, I thought, what would it look like for us to go out on our own and really build something that's, by procurement for procurement that's easy to use and fast to deploy. And that's the adventure we've been on over the last I guess it's four or five years as we've grown the business. And we work with a range of Fortune 500 and Global 2000 businesses, specifically around helping them deliver new products to market faster. And its structurally improved pricing.

Tom Raftery: 5:22

Okay. And for people who are outside of procurement, what's, broken in sourcing today that most leaders still don't see?

Spencer Penn: 5:30

So maybe I can take a step back for folks who are not in procurement, there are broadly two areas. There is direct and indirect. Direct is anything that the company buys to put into its product. Indirect is anything the company buys to consume itself. So at Tesla, car parts would be categorised as direct. They're going into the vehicle, they're being sold to customers. Tesla also bought soaps and office chairs and fire extinguishers and toilet paper and flights. All sorts of stuff not to be sold to customers. That's indirect. And in the indirect side, there is an amazing amount of solutions available. If you are a company, you want to spend your money you've got the P2P providers. You've got a lot of the sort of, purchase cards. Like you can think about Ramp and Brex. You have expense management tools like Concur. Amazon Business, you just have these all in one type marketplace vendors. But when I look at the direct materials side, folks are just left without tooling. And I think some of the challenges, even though it is the largest area of spend in the business, it is some of the more strategic and some of the more complex. And it varies by company. So every company is buying laptops, but not every company is buying caustic soda flake. And then those same companies aren't buying injection moulded 5% glass-filled, parts. Those same companies are not buying tungsten carbide for their rocket engines. And so there's a pretty broad array of things that people source in the direct material space. And the second element is that it's very often engineered. You have engineers who develop these specifications. They're never sleeping, they're always iterating. And sourcing teams are always playing catch up to try to stay close to what's the latest coming outta engineering. And they are the surface to collaborate with suppliers. The suppliers have ideas. And so what I found is that it's like the engineering develops a spec. They send it to procurement. Procurement sends it to suppliers. The suppliers have questions. They send it back to engineering, they send it back to procurement, they get quotes back. Prices are changing. Finance gets involved to understand what's going on on the bill of materials cost. Manufacturing gets involved as the stakeholders get involved. In a lot of organisations, it's 500 to 2000 emails just to figure out, who's the right supplier at what price, and is that price fair? And we try to bring that into a single pane of glass. And now you might ask, well, a lot of companies have ERP systems like, SAP or Oracle. Don't those have the ability to manage this process? And the answer is not really. Most of the time, what we see almost universally across the Fortune 500 is you might talk to someone who's like a chief procurement officer and they feel that they've implemented some tool or technology. But when you actually talk to folks in their team they'll do all the work over Excel and email. And only at the very end of the process when they've chosen the supplier, they've made the award, they've selected the price then they populate the ERP. And so the first time you really touch a digital system is when the battle is already won or lost. And that doesn't work when you're trying to make your speed to market a competitive advantage, not a competitive disadvantage.

Tom Raftery: 8:42

And why does this matter more now than it did 2, 3, 5 years ago?

Spencer Penn: 8:48

It's a great question. I think it's always mattered, but I do feel that there is a changing landscape for manufacturers that didn't exist before. I think of them as the new wave of manufacturing. So you can think about Hims and Hers is a pharma company. They're a, compound pharmacy, direct to consumer. That's new wave. Pfizer, amazing business. But that's the, existing incumbent You think about Anduril, this is a new wave, aerospace and defence company. The old wave was Raytheon, Lockheed the new wave is Tesla and Rivian, Lucid Motors. The older wave is GM. The new wave is Harbinger for Trucking. Old wave is Mack truck. And so I'm seeing across a lot of industries, there's this next generation of hardware businesses that are challenging the norms and are not accepting the. Hey, it just takes time. Or, Hey, this is just how the industry works. And those are the ones that are winning because I think very often we're not in just local, competition anymore. We're in global competition. And the modern generation of manufacturers are winning on a new metric, which is speed. And they understand this and they move really quickly. And it's very easy to quantify the cost of turbulence or obsolescence or, the cost of retooling and waste. Yes, that happens when you're innovating, but it's much more important and much more difficult to quantify the advantage that you get from getting to market three years sooner. Being able to iterate with customers faster, being able to incorporate the latest technology. And, there are a lot of companies that don't really exist in the same way that they did before. If I think about Xerox Park, they invented the future and let other people build it. When I think about Blackberry, they lost touch with what the consumer wanted. If you think about Kodak, they invented digital, but then defended film. And so there are a lot of companies today that you look at and think these companies are never gonna go away. These are pillars of the economy. And the next thing you know, you'll see companies like Tesla come up as the most valuable automaker in the world, and maybe very soon the most valuable robotics humanoid robotics company in the world. We don't live in the world that we were born into, and the next generation of manufacturers are using LightSource to get that competitive edge. So that's the, what I'm seeing in the market. it's a good question. Why is, why is it different than it was before? I don't know the pace of, change, the availability of, technology. I do feel like AI is helping people to develop more things faster, and your limitation becomes your willingness to move at speed more than the human, resources that you need. It's a good question. I actually don't have an answer for it, but I what I, I know what I'm seeing with my own eyes, but what do you think, Tom? I mean, are you seeing something similar to this?

Tom Raftery: 11:45

Yeah. I do. I mean, I, I think you hit on it there at the end, particularly the emergence of AI. I mean, AI has been there for decades, but it was, if I remember correctly, it was the 31st of October, the 1st of November, 2022 when OpenAI launched ChatGPT to the world. So it's been out three and a bit years at this point, and so much has changed as a result. Not just of that, but of everything that's gone along with that. It's kind of had this halo effect and it's brought along a load of other AIs along with it. And I think that has had significant impact on lots of different organisations and technologies. And I think that's, certainly kickstarted and, sped up the world if, to put it, as broadly as possible. Everyone says AI agents now and I'm, I'm sure you're hearing that as well in, in plain terms, what do you see the difference between an assistant that helps and a system that actually owns part of the work?

Spencer Penn: 12:50

Yeah, it's a, it's a really good question. Let me start by talking a little bit about the difference between LLMs, Gen AI and agents, and I'll build that distinction first. A LLM, we all know it's generative AI. What does generative AI mean? It means that the user inputs a prompt and then the output is a generated output. So you can input text as a user, and then what you get in response is an essay or a piece of music or a piece of an image. It could be code. And so that's the output. Input and then generated output. With an agentic system the input is the user input, and the output is action or sets of actions. Now, LLMs are the underlying technology that has really been the big step change for both of these areas. Then if we pinch and zoom within the agentic universe, there's, I would call it agency 'cause that's actually the origin of the word agency. There's a whole spectrum anywhere from the traditional RPA, which I would say is low agency. If you think about something like UIPath or Zapier of course they've changed their products in the last couple of years, but in their original instantiation, it was a sort of a Rube Goldberg machine. If this, then that, and you'd have a person design that, okay, if I receive an email to sales@lightsource.AI, route that to HubSpot, then HubSpot assign the account executive, then email the account executive, send a note in Slack. And so in a way that's an agent. It's autonomous, but it doesn't really have very much decision making authority. On the far other side of the spectrum, what you start to see is with agentic systems a robustness and ability to think, but putting it in more technical terms, what it really is, is that the agent has a set of goal states that is trying to get to, and it's self-referencing and checking either in a single agent or a multi-agent system. Have I achieved the goal state, yes or no? What can I do next? What do I plan my next set of actions? Go do it, check again. Have I achieved the goal? Okay, I've not, let's keep going. And it's an iterative process and also because LLMs are more robust to the kinds of data that they're looking at.'cause again, like these language models are, are really, really powerful and flexible. It can do things like look at an API that it's not seen before. Query the documentation, understand what's the different pathways and calls that the agent can make, and then continue to explore the map, as it were. And so there's been a turning point and I think it's been much more recent than even people would, suggest. I think the first time somebody asked me about whether we were setting up an MCP server, which is a specific API for agents was last summer. So it was only nine months ago. And at that point I looked at them, was like, do we really need to do that? Is anybody and none of our customers were asking me if they could connect to our MCP server, so it felt like maybe cart before the horse a little bit. But I would say that since it felt like it actually happened over the holidays, which is kind of weird I felt like we left in December we came back in January from the holidays and some of these agentic systems, had gotten just good enough that people started to feel like there was a, a second sort of step change in the AI progression path. So the first step change being not just the unveiling of ChatGPT, I think it was really ChatGPT 3.0 was the one where it went from being like a toy to a kind of neat utility. Like my first use cases of ChatGPT were showing my dad how I could make poetry with the voice of Yoda from Star Wars. And that was cool, but I wasn't thinking that this would solve real business problems yet. But I can see the path. And then more recently with the agentic systems, it went from feeling like, hey, this is a bit of a pie in the sky to, this is an all out sprint to incorporate these things into our business because it seems like they can really do work for people in a relatively robust way. And I don't think that there is one architecture that has become dominant. I don't think there's one methodology that's become dominant but likely the way the world will sort out is that there are broadly, are systems of record and systems of action. Systems of record are places where you store data. These are the sources of truth in the business. So you might think of your ERP, you might think of your pricing repository like LightSource. You might think about your PLM as being systems of record. And then you have systems of action, which are things that move those numbers around and do things externally or internally in the business. And those can either be built within the tools, they can build in companies own agentic harnesses, or it could be a mixture of the two. Like we have a lot of customers now where they have an agent system that their IT team is building internally, but it's almost more like a general purpose agentic system, which then goes and calls sub-agents within LightSource to do the more specific tasks. Folks have really woken up to what agents can do for their business. It's both exciting. And for some people it's really scary. I've always felt that, people who live in the past miss the future. And I think it's really important to remember we're never living in the world that we were born into. And even for me, I feel young, but even for me, I have to really stay close to the technology because man, it is changing so quickly.

Tom Raftery: 18:32

Yeah, Yeah, yeah, yeah. And where should procurement leaders be genuinely excited, and where should they be sceptical?

Spencer Penn: 18:39

Yeah, it's a really good question. I think procurement leaders should be generally very excited. Like full stop period. I don't think that procurement is a function, at least in the direct side. I really don't believe that procurement is a function that is going to be automated. I think it's a function that's gonna be augmented. And what I mean by that is that there are two brands of companies that I'm starting to see emerge in the AI space, at least AI native businesses like ours. There are companies like Decagon or Sierra where their goal is to do human replacement. And Decagon and Sierra, they make autonomous call centre agents. So you call in, you talk to somebody, but it, it's no longer a person in the Philippines. It's actually an AI agent that, ideally follows the very exact new script. If you notice, when you call a human at a, call centre, they don't have a lot of flexibility or creativity. They're actually being constrained to specific questions and answers and pathways. And so it's actually a really good application for agents because you don't want it to be, you don't want to give them that much flexibility. You just want to give them the ability to understand what the user's saying and then, route the request and give solutions accordingly. When I think about procurement though, the scale of the dollars is so enormous that the human cost is actually relatively small. So when I think about, the cost of a call centre and the cost of an Uber driver, for example, the majority of the cost is the, person doing the activity. That's why self-driving ride share is such a good market. I had this friend at Waymo who, was thinking about doing a startup. This is a little bit of a tangent, Tom, so hopefully we can keep this thread and I'll, I'll get back to it, I promise. But he had this idea for a self-driving, like an autonomous intercontinental shipping container vessel. These big container vessels or oil tankers. And then he had a whole philosophy, it's, well, you're in the water, you're navigating the ocean, so it's actually easier because you don't have traffic, you don't have pedestrians. And then you can have someone go to the boat at the last mile and bring it into the dock. And I looked at it and I thought, yeah, I believe it's technically feasible, but that doesn't mean it's a good idea. And he's well, why not? I'm like, well, how much is the value of the cargo on that boat? Maybe a hundred million dollars. And I say, how much does it cost these three, eight, folks that you have running the ship? They're making a hundred thousand dollars a year, $200,000 a year if you're being really, really generous. The pie chart of the dollars of value, dollars under management, as it were, versus the human cost is like extremely lopsided. And when I think about procurement, it's very much the same thing. And so when I think, okay, should I take my procurement team headcount and reduce it from 30 to 25? Okay, maybe you can win or lose, a few hundred K. But individual procurement managers are sometimes overseeing decisions that will create or destroy hundreds of millions of dollars of value. And so my philosophy is if we have imperfect technology, which is the case today and probably will be for a long time we should focus on augmenting that function. By the way, I don't think that's universal. I think some functions are likely to be augmented, some are likely to be automated. And I think it comes down to the, the scope of the function and the portion of the human cost that makes up the overall cost of the function. Now in indirect, I think this story's very different, and I break up indirect into two kind of buckets. One is strategic, indirect, and then there is tactical or tail indirect. Strategic indirect actually to me, I, think of these as identical to direct. If somebody is sourcing logistics or packaging, it's still SKUs, it's still managed in a TMS or an ERP. You can think about it very strategically. They're tracking market pricing, they're tracking benchmarks. It is essential arteries in the business. Whether it literally is in the product or not is not the more important distinction. It's the way in which it gets sourced and what's the dollar value and what kinds of people are managing it. But when I look at the tailor, tactical, indirect, you know, just basic requisitions for, ad hoc equipment under 5,000, $10,000, I don't see a reason why you wouldn't try to make that experience completely touchless.'cause it won't really move the needle. Now the question of how should procurement leaders really be thinking about implementing technology changes over time. But I think there's broadly three approaches I would suggest that I think are no regrets first moves for people. It's very practical. One is start by getting AI tools into your team's hand. If your IT team isn't enabling you with this already, I'd be surprised, but every person on our team, at least I want them to have an instance of ChatGPT or Claude from Anthropic so they can do their work better and faster. I want everyone on our team to be tech literate in all their little work streams that they're doing, whether it's helping them draught a memo or write an email or interpret data because you can just throw it into the machine and ask for results. So that's the first thing. Get your people just using AI on their own. The second thing is have an AI strategy. And so think about what is our AI roadmap from a tools and technology perspective. And I think that coordinating with your IT team is really important on this because the best companies, they're mediating potential right hand, not talking to the left hand, where the IT team is thinking about an extensible agent framework for all business functions, not just procurement. You should be really plugged into that and talk to your counterparts and understand it. And then the third is to become a venture capitalist. And this is a bit of a weird thing to say, but when you're looking at what is different between venture capital versus private equity or public market investing, venture capital makes many more bets. And they know that many of those bets will fail. Right. They invest in 50 companies and they hope that they get one Facebook out of the fund or one Snapchat. And then the rest they know might not work. Yet some of the best venture capital funds have the highest returns of any financial product, anywhere at any asset class. And so it's not about trying to find, a few optimisations that can get you 20% improvement, 30% improvement. When you look at technologies that are really on the cutting edge, like AI, you start to see opportunities to create 10 x, a hundred x, a thousand x kinds of outcomes for your team and for your business, and especially procurement, which is, again, biggest flow of dollars out of the company. I think the cost of running a POC, the cost of running a pilot if it's 50 or a hundred thousand dollars with a new shiny tech vendor is very low. And I think that companies should be a little bit more open to the idea of experimenting with new solutions. And by the way, that doesn't have to be at the conflict of your existing tools. If somebody is implementing SAP S4 HANA, that doesn't mean that you can't also run a POC with a, next generation AI, native ERP provider, as long as there is a pathway where if you learn something, you can push your vendor to be better or you can actually switch over. I would just do it. It's like saying to a venture capitalist, Hey, you've made one investment in one software company. Don't make any more bets for four years. That would be silly. And so those are the three very practical pieces of advice that I would give to chief procurement officers at this moment. One is to give your individual employees AI tools that they can use on their own just get it into their hands. The second is to get close to your, IT team and understand their AI roadmap. And then the third is make venture capital investment style bets on POCs that might be able to, have a game changing impact on your organisation.

Tom Raftery: 26:40

Okay. And where could AI add the most value first? Would it be something like supplier discovery, quote comparison, scenario analysis, negotiation prep? I don't know, something else?

Spencer Penn: 26:53

It's a good question. And I do think about this all the time. The answer is, it, can do all of these things. Where I would start is where the biggest pain point is. You know, I think that we all have a hundred priorities that we can think about on a given day, but there's only two or three things that will truly move the needle. I would focus on those things first. It is tempting to go and look at things like tail spend and say like, oh, can I automate this away? But the CFO's never gonna really be super excited by that. But if you can show the CFO a way to better manage the outcomes on cogs or direct materials that will be visible to shareholders and the board. If you can get your product to market faster that will be the number one priority of the CEO and the board. If you can unblock your engineering team on their delivery of the product that they're sprinting to get done that will make your, VPs of engineering or your CTO extremely happy. And I, I always think about focusing not on, the, tail, but the core it people, it's funny. It's so human nature Tom, I mean, you've probably experienced this where it's man, I've gotta do my taxes. And it's this is such a big project, but you know what I should do first? let me respond to an email. Let me wash my car. That's not the, the number one thing you really have to do, but it's really easy to look at the hard problem and then focus on the rounding error problem because it's, it's much more approachable. You can nibble away at it. And so I, I would encourage people to focus on the core problems in the business that can truly make the difference. If you're paying $5 more or less for your pencil sharpener, I, I really just don't believe that that's gonna be the thing that makes the difference. I think it's gonna be how you manage your direct materials, your core operating spend. I think for me, that's the main thing, at least for the customers that we work with, right? If you're talking about a company like a Salesforce they're a tech, business. Of course, they don't have any direct material spend. So the, the equation is totally different. So I'm, I'm imagining the audience and the people that I'm talking to are manufacturers people making physical products in the real world. If they're not, then the equation is a little bit different. Then, in fact, like I would say, if your human capital costs a big part of it, then AI for HR is the place where I would focus. How can you be most effective with your people, if that's the biggest thing in your balance sheet.

Tom Raftery: 29:18

Yeah, no, of course. And what changes first when teams move away from spreadsheet and email?

Spencer Penn: 29:25

It's a good question. It's actually interesting what really changes first is actually the perception of capability. I know that sounds really silly, but a lot of times when engineering, at least at at Tesla I felt this a lot, when engineering would update a part or update a spec, they would send it procurement. And if they got a glimpse of the way that procurement was interacting with suppliers, they would be kind of sceptical. Does it really make sense that you're using spreadsheets and emails for this really critical work? It makes it seem like everything you do is like a big project because you have to go back in time and figure out who sourced it a year ago. Are they still at the business? Did they collect our standard Excel form breakdown? Then if I can find it. Someone asks a simple question, why did the price change? And you have to click cell by cell through this spreadsheet to figure that answer out. And it takes you, you know, you need to get back to people. You have to go do some research. You just don't have the data at your fingertips. And I think the folks in the business can really see that. And I think if you bring a level of, tech excellence to your own organisation, the first thing that changes is the way that other people in the business view the function. So, that even happens before you see results. Now the second thing is, it actually does make your life way easier. And so if you get off of Excel and email, and I don't think Excel and email, by the way, are bad tools. I think they're really great tools for certain things, if you're working as an individual in silos. If you're working as a team in an organisation towards a common goal, that's what makes businesses powerful. Like I think there's a lot of great individual contributors doing things, individual proprietorships, but a lot of the great technologies, great innovations great global impact come from businesses. And what are businesses, they're collections of people that get together and say, we're all going to work towards some common goal. In exchange we're gonna get paid by the corporation and own stock in the corporation, and we're gonna have management. And we're gonna work with other people. And so what makes a business different from an individual is that ability to coordinate among one another. And those kinds of things are made possible through having systems, technology systems good or bad or otherwise. Then the third progression that you start to see is when you use a system like LightSource or anything else, you eventually build up this incredible data repository.'cause the minute that you start to get out of these silos and you start running your RFX supplier collaboration through a single pane of glass, a tech enabled single pane of glass, you'll notice yes, it has made my life easier, it has made my job easier and faster, but also wow, like we've been inadvertently collecting all this incredible rich source of information that we never had access to before and didn't live in any repository. And so all of a sudden you can start to think about things like, who are the other suppliers we've worked with in the past that we could also consider. Are there other suppliers that we didn't include in this event that we should be looking for? Is there supplier discovery from the broader universe that we should include in here? Did that same supplier give us a different price in a different area of the business? I can actually see that. Prices have changed, steel went up. Does that mean that the part price should have gone up? Or did it go up by the right amount? Should it have gone up by less? All of these things are not just about, automating your process, but it's about having novel insights that emanate from the systems. And the tagline around all of this is proactive intelligence. And so going from reactive to proactive, when you start with you know, you might have some data that you have to look at, the user has to come with an idea and go seek out the proof. In an AI enabled, AI native system of record, like LightSource, the equation is flipped. The system presents to the user the insights that they want see. And so that's a different approach in terms of mindset and methodology. And, and with agentic you start to see the world where people wanna wake up and the job has been done for them in its entirety, you know? So it's a, again, a very different way of thinking.

Tom Raftery: 33:26

And have you seen a moment where this has changed a business outcome, not just a a team metric, for example?

Spencer Penn: 33:34

Of course, we have a, I can't say the name of the company, but we have a, a big three automotive customer and they launched a new EV programme. And they were using LightSource for a portion of the programme. And what I wanted, it's funny because when, by the time they got the LightSource licence, they had already sourced half the vehicle. And then the second half of the vehicle was sourced on the platform. And I was peeved about this originally because I wanted a hundred percent of the vehicle on the platform. I wanted to see the full bill of materials. I wanted it to be soup to nuts through LightSource. But the happy, accident of all this was that we got an an AB test. So we were able to look at the portion of the parts that were sourced through LightSource and the portion of the vehicle that was not. And we found two big outcomes that did really transform the business. The first was that the categories that were sourced through LightSource had a sourcing cycles that were 25% shorter, meaning that every RFX event was 25% faster in terms of supplier response, to award, like the full end-to-end process. You compound that across a large number of SKUs and a bunch of long lead time items that can have an acceler, a real acceleration effect on the ability to deliver the NPI programme. And then the second thing was that we saw, I don't wanna say a reduction in cost, but we saw a 37% reduction in cost creep. And so it's sort of like, if you remember Richard Nixon, it's not the crime, it's the coverup. It's very often not that they didn't get a good price upfront from suppliers. It was that once they awarded and they were no longer competing to win the business, the supplier's motivation flipped to how can we get more margin from every incremental design change? And so this is the sort of painful thing, is that if you are, Lockheed Martin, you freeze your design and you never change it, and then you source it and then your prices are fixed. But in the most innovative companies, innovation doesn't stop. And even when you launch the vehicle at Tesla, we would continue to improve and iterate and update the design throughout its lifecycle. And the problem is you get a lot of commercial loss if you don't have your really good systems in place like LightSource. And so for this OEM, what we noticed was that for the portion of parts that were sourced outside of LightSource, the delta between the LOI and SOP price, which is like the initial letter of intent to final production price, was a certain number. I had have to go back and get the number exactly, but for that same set of categories that were sourced through LightSource, that cost creep number, like the increase over time was 37% less, which is, again, huge, huge impact. And when we're talking about the scale of big three auto, it again, it's, it's hundreds of millions of dollars, when they use a tool like a LightSource.

Tom Raftery: 36:26

Sure. And ahead, what parts of sourcing do you think will become truly autonomous?

Spencer Penn: 36:34

Yeah, it's a great question. I would say the, first thing will be identifying for the user the changes that have happened in the bill of materials. That sounds like a simple problem, but it's actually quite complex. The second thing is preparation of the RFX. I think there will be a little bit of human in the loop for strategy and review, but I think preparing the actual tender, there's no reason why that can't be done fully autonomously. Finding suppliers, creating a long list and a short list, which is a mix of both your internal vendor, approved vendor list, AVL as well as scoping out of the market. Who are the alternatives that you could look at? That can be done autonomously. We have an agent called the supplier activation agent, so I'm talking my book a little bit here, but getting the suppliers to respond, acknowledge the event, answer their questions about what they're looking at, and then help get them through to a quote that is accurate and representative of what they intend. It's a lot of work for the category managers to do. They often find themselves having to kind of shepherd the suppliers along. We use agentic systems to facilitate the suppliers' side as well, which is something very unique about our platform. And then finally, the analysis and award, and then the post award. Something that's often dropped in the ground. The post award work that has to happen, that includes proactive insights from market changes. So steel's gone up, steel's gone down. This might be a good time to re-quote these certain items. Hey, by the way, these are not index adjusted, but there's a risk that the supplier will come back and ask for an index adjustment because when prices go down people let sleeping dogs lie. When prices go up, the suppliers make sure that you're aware of it. So it's kind of this one way ratchet unfortunately. You can solve a lot of those problems with better technology. And then I'll just throw one more on there, Tom, and this is something procurement doesn't have the time to do today, which is quantify and demonstrate their value to the rest of the organisation. I think procurement is the most underrated and underappreciated function in the business where, it's if sales goes out and they sell a hundred million dollars contract, it's like an incredible feat and they're getting bonuses, they're getting 10% as a commission, they're getting high fives. It's like this, big thing. If procurement goes out and they save money or they prevent a crisis, people don't look at them and say, Hey, you've just done this incredible thing for the business. They say, good, you've done your job. Or worse yet, they might look at it and say, why didn't you find, savings last year? You know, If there was money to be had, how was that left on the table for the last two years? And so it's sort of one of these things where it's like running water. Like people don't notice when it's working. They only notice when it's out. And I think the best procurement organisations, I would say 90% of their time is spent doing the work, and then 10% of their time is spent communicating and explaining their value to the stakeholders and rest of the business. But that is not the core of the work itself. And so I'd love to see more systems able to help them, automate and capture that. But to know if a cost reduction is legit, you have to understand what's the market backdrop? Did prices just go down because steel went down or did prices go down because in spite of steel going up, you landed a great negotiation strategy. That is almost as much work as, as the work itself. And so I think I'd love to see that automated over time as well.

Tom Raftery: 39:52

And if listeners were to take one thing away from this conversation, you know, how, how should they think differently about procurement now?

Spencer Penn: 40:00

My first thought would be for procurement professionals, and this actually applies to, probably all professionals, is to start to think of yourself a little bit less as a subject matter expert and a little bit more of an orchestrator of systems. And thinking about how you can automate your own work. I mean, we're not talking five years from now, we're talking the next 12 months. I would be shocked if we don't find that almost every procurement person, professional has an army of effectively AI interns doing work on their behalf. And so learning how to be good at managing, overseeing, providing feedback to those AI agents or AI interns, I think is a new skill that folks are gonna have to learn if they wanna really thrive in the modern era. We see this with the earliest adopters who are engineering teams. A lot of the highest velocity engineering teams today in Silicon Valley are no longer writing code. What they are doing is they are reviewing code written by agentic systems, and it's comical to think about even just a year ago, you know an engineer would be typing in like function bracket equals, dot, dot, like typing code by hand. I think they, they should be able to read it, but the work should be coming from the agentic systems and then the people should be thinking about the strategy, overseeing it, and then doing the things that humans are uniquely able to do, which is one, go out and do things in the physical world where LLMs have not been able to touch yet. They work in screens, they work in software, they don't go in the physical world. Procurement, get out there, go to your suppliers. And then the second thing is comprehend strategic thinking, which you can ask the LLM what to do, but the number of questions that you could pose to an LLM is infinite. And so it's not gonna come up on its own with a, approach to how you should manage your time or your work. So that would be my one way to, to change how you think about the function. And then procurement teams that do this really effectively, I think they're gonna start to see, their stakeholders view them in a completely different light. And I'm really excited for that and it's well deserved and it's oh, long overdue, This has been the sort of mantra of procurement for a long time, which is, hey, we want to get a seat at the table. We want to be a strategic thought partner to the business. And I think some, some organisations have made really good strides there, but it's not universal.

Tom Raftery: 42:26

And a left field. Question for you, Spencer.

Spencer Penn: 42:29

Sure.

Tom Raftery: 42:30

if you could have any person or character, alive or dead, real or fictional as a champion for automating procurement, who would it be and why?

Spencer Penn: 42:44

That's a really interesting question. The first person who comes to mind is, Henry Ford. He was a brilliant, brilliant business thinker. Obviously he is not without his, rough edges and almost anybody now in history, you, you can look at and, and, and, and accurately say they were horrible people in, in certain ways. And I don't agree with everything that Henry Ford did, but I, I think that in his day he was a complete maverick in his ability to innovate the way that something was built. and, the real Ford product was not the Model T, it was not the vehicle, it was the, the manufacturing process and infrastructure and repeatability behind the Model T. I also love the idea of somebody just doing something different. Looking at the world and not saying, what is, but why not? And I do think that he was the first principles thinker to be able to innovate his industry really well. And, and he's built one of the, three iconic, vehicle businesses as a result. And, when I went to go visit some auto customers of ours in Detroit, I went by the Henry Ford Museum and, just felt that feeling all over again. And he was also a great violin collector. So that's another personal interest of mine is musical instruments. And so it's the weirdest thing if you go to the Henry Ford Museum, there's like this display case that's just tucked away and there's four Stradavarius violins and a Guarneri, and you're like, what? And an Amati and you're

Tom Raftery: 44:18

Wow.

Spencer Penn: 44:19

why is the, one of the best collections of Stradavarius violins in the Henry Ford Museum in Detroit? And it's because Henry Ford he had an interest in other things. And I like to believe that that kind of intersectionality between his interests in, other disciplines, whether it's not just manufacturing, but design, art and music may have contributed to his, innovative eye. But I bet if he were to see the procurement function today, I think he would find it really, really tough to stomach. And he would probably go and say, every single manufacturing business in the world needs to be using LightSource, or they need to, file for bankruptcy. know, and I think that's an extreme, an extreme statement, but I do feel like that's, I would love to have him as an advocate.

Tom Raftery: 45:06

Okay, great. We're coming towards the end of the podcast now, Spencer, if people would like to know more about yourself or any of the things we discussed on the podcast today, where would you have me direct them?

Spencer Penn: 45:18

Absolutely they should go check out LightSource.AI. That's our website. That's where they'll find all the latest news, all the latest product demos. A lot of our thought leadership is there. We are not a procurement company. We are a technology business that serves procurement. And we are a thought partner to a lot of our, clients and customers on not just the things that we do today, but the way that they should be approaching AI and agentic systems in their business more broadly, more long term. And we'll always be direct and honest. And, my CTO and our, my co-founder Idan, he came from Google X Google Research, the birthplace of the LLM. I came from Tesla and Waymo. So we're, all here in Silicon Valley, hardcore technologists, as it were, we're really plugged in. The other thing folks can do is I, I'd love to hear from you, so feel free to find me on LinkedIn or reach out to me directly at Spencer@lightsource.AI. That is my real actual email. I won't say that I don't have over a hundred thousand unread messages, but if you send me a note and we have a reason to connect, I'm happy to say hi.

Tom Raftery: 46:23

Fantastic. Spencer, that's been really interesting. Thanks a million for coming on the podcast today.

Spencer Penn: 46:27

it was really an inspiring conversation. You got me thinking about some really new topics, so I really appreciate your having me.