Episode Transcript
[00:00:03] Speaker A: Welcome to Aina Insights, where prominent leaders and influencers shaping the industrial and industrial technology sector discuss topics that are critical for executives, boards and investors. INA Insights is brought to you by INA AI, a firm focused on working with industrial companies to make them unrivaled segment of ONE leaders. To learn more about INA AI, please visit our website at www.aina.AI.
[00:00:40] Speaker B: Today we are joined by a true innovator at the intersection of AI and industrial operations, Pablo Palfox, co founder and CEO of Happy Robot. Happy Robot is transforming the freight industry with its enterprise AI platform that deploys intelligent agents to automate communications across voice, text, email and more.
These AI workers handle everything from outbound calls to inbound inquiries, seamlessly executing tasks through customizable workflows built directly onto the platform.
Before launching Happy Robot, Pablo pursued his PhD in Computer Vision and Deep learning at the Technical University of Munich.
He holds both a Bachelor's and Master's in Robotics and Engineering from the Universidad Politecnica de Madrid and has also contributed to cutting edge research as an intern at meta's Facebook Reality Lab.
So with that impressive background, Pablo, we are excited to have you here and thanks for joining us.
[00:01:44] Speaker C: Thank you so much, guys. Thank you so much for hosting us. Very excited.
[00:01:49] Speaker B: Glad to have you. So, Pablo, let's start with your story. You've spent years in AI research.
What first drew you to this space and how did that early interest evolve into a career?
[00:02:02] Speaker C: So I think it's always been my brother and co founder Javi, who's been pushing me to go into computer science. Back in university in Spain, I'm originally in Spanish. I was doing my bachelor's in robotics, but as a whole, mechanical engineering, if you will. And my mom wanted me to go into the energy sector. Then that was when my brother said, no, no, no, you're going to go into computer science, you're going to learn AI.
It wasn't even called AI back in 2012.
That's when I started my undergrad.
That's basically a little bit of how I, I got that push in this case from Javib, our co founder and COO of Happy Robot, who always was pushing me there. But then ultimately, what drove me deeper into specifically deep learning, back when I was starting to do my master's and then the PhD in Munich was really how fascinating that was.
I was fascinated by the fact that you could use, in this case, a camera. I was very deep into computer vision. You can use a camera to, to analyze the setup and your scene and your environment to Have a robot navigate the environment.
In my case, I was actually working with drones moving and flying around a landing platform and landing autonomously. So that was my undergrad bachelor's thesis. So that was fascinating to me. Right. And then I dove deeper into it.
[00:03:29] Speaker B: And that was the start of a career. Which now led you to happy robots.
And then happy robots, it's into supply chain and freight, which aren't the most obvious sectors for say, first time AI founders. So tell us a little bit about that story. What led you and your co founder to focus happy robots on freight and supply chain?
[00:03:52] Speaker C: The first thing that people actually tell us is why happy robot? Like why the name? Are you building robots? And funnily enough, the name actually comes from when we were building a computer vision platform.
It was a completely different product, it was a completely different company, if you will. Although the name remained. But right after I dropped out of the. Officially dropped out of the PhD, I always think that I completed it. My professor wanted me to do more papers. We're in good terms now though. But when I dropped out of the PhD back in 2022, what we thought we needed to do is leverage our expertise, and mine was in computer vision. So we started diving deeper into let's build something for computer vision. Probably not the right move because you should always have the customer in mind, not the solution. Right? So the solution in search of a problem is a real thing and you should be really mindful of that when starting a company. But in our case, we just started doing what we knew best, building technology. My cto, Luis and I were best buddies from college since our first day in 2012 that we met. And so we started building these things.
We get into Y Combinator, probably one of the most prestigious accelerators of startups. So we got into YC in 2023 and right after that we came in with this idea for computer vision. Right after that YC batch, three months of deep diving into your idea and continuing to build it, we realized that we were building the wrong thing.
So imagine as three guys that just have moved to San Francisco to build their amazing company and turns out it's the wrong thing to build because we were not really getting a lot of traction.
So we throw that away. We actually had some revenue, some traction. We had like 70k of ARR or something like that that probably made us think that we could build something there. But it was kind of a red herring, if you will. We shouldn't have pursued that. So what we did is we throw that away back in November 2023. We're starting from scratch.
So picture us, Javi and Luis and myself in a room for two months straight, just looking at a whiteboard, like an empty whiteboard or blackboard. And we're just not even sure why we're there. What happened? We had an idea and we say, okay, let's go back to the basics.
What has changed in the past few years and months? And obviously LLMs were getting out there and that was when GPT 3 and 3.5 were actually starting to become really useful.
You can actually do real useful things with LLMs. So we started diving deeper into that.
My co founder Luis always has loved voice and he wanted to build stuff in voice, text to speech, like the so called text to speech generating fake voices or virtual voices. Right. But that sounded human.
So we combined the LLM piece with the voice piece, a little bit again of a solution in search of a problem again. But thankfully our co founder Javi was like, okay, what do you guys have put together here? We built a very simple demo of a voice agent, basically an AI that was able to speak on the phone. So we put something together and Javi's like, okay, this is really cool guys.
We have built something amazing. Now let's actually see where we apply this. And turns out that Javi was the CFO of the largest olive oil company distributor in the world, D olio, which is the company behind brands like Bertoli, Bertoli Olive Oil, you probably see that in Walmart or Kroger. So he knew a little bit about specifically freight brokerage, which is the first niche that we went after. And he was actually working with freight brokers himself to deliver throughout the us. He was already CFO of Diolia North America. So that experience gave him the, the knowledge and the understanding that there was something broken, if you will, in terms of all the communication that went keeping and that kept freight moving.
So he actually had to hire, this is a fan fact, he actually had to hire interns to call drivers directly, bypassing his freight broker. And that was just for a period of time. Obviously they were not really happy about that, but they're in good terms now. But that's to the extent that he had to get resources internally to reach out to drivers and see if they're going to arrive on time. There's been so many attempts to digitize the industry, specifically domestic freight, but holistically supply chain turns out that at the end of the day you just need to meet people where they're at.
You just need to meet the Players where they're at, be it an app. An app is great, but it's not the only medium. You might also have to send an email. You can also. Or you might actually also have to reach someone via phone and that's the only channel that they have in that player. Maybe like you're reaching out to a driver and they only have a phone and they don't want to have more apps, so they don't want to install your tracking app anymore. So things like that, that's what drove us into the space.
[00:09:05] Speaker B: It's interesting, you're meeting the user where they are and if I were to kind of read back from a happy robot standpoint, you started with voice agents, which is kind of unusual among AI platforms. Everyone's doing text based because it's more mature. It's there what led to the voice and what's its unique value in your context to unlock the customer?
[00:09:30] Speaker C: I have to say it was not fully intended that way. We just decided we would build something calling voice first and then we actually learned to.
We quickly realized that that could be our unique differentiator because we were seeing a lot of people tell us voice is impossible. To build something realistic, you're going to just like fail.
Don't do that. It's years away from actually working. It has to be very realistic. Drivers will never talk to an AI. Shippers will never talk to an AI. Receivers never talk to an AI.
We actually did that and we started improving it. Obviously at the beginning it was very bad. It was like, oh like five seconds of latency, you would say something hello. Five seconds later, hi, how can I help? It's obviously not super realistic, but as tech got better and we started building our own infrastructure to support lower latency and more volume of calls, we saw that that could be actually really powerful. And it turns out that voice is really the hardest piece. And we have customers today that tell us, hey, these email use case, I think we're just going to build it in house perfectly fine. We can actually even provide our developer tool for you to build on top of it. You don't have to. You can build your own framework for building AI agents via email. But it truly is more simple now when you have a value proposition which is, hey, this platform is for you to build your multi channel AI worker.
You don't have to go elsewhere and put things together. You have Voice, one of the best voice orchestrator platforms in the market. Have email, you have texting. We're introducing WhatsApp soon. We have some pilots Already going on. So at the end of the day, you're trying to create a platform to build the AI workforce for supply chain and more holistically, which is what we're now realizing that we're doing is we're serving physical operations, we're serving the real economy, if you will. We can talk about the ICP and who is our ICP later.
[00:11:35] Speaker B: But yeah, so the real economy, the physical economy that you talk about in our world, we think about that as industrials, the titanium economy, industrials.
And what we've seen is they're often seen as slow to adopt new technology.
As you dwell deeper into the freight and supply chain and the logistics space, have you seen this perception match up with your experience in terms of how they are adopting the product, sales and where they are in this journey?
[00:12:07] Speaker C: I think after AI, AI meaning the search of LLMs or specifically LLMs, now we just refer to AI as anything.
After AI came to be something real that people could actually touch and like interact with on ChatGPT or whatever.
I would say the sales cycle has advanced, has reduced. Like the length of a sales cycle has reduced dramatically. Now people actually know that they could use AI to automate their customer support line or to do a campaign for 10,000 calls in parallel to reach out to all of their suppliers to see if their parts are going to arrive on time. So now people have their wheels turning and saying, like, oh wait, wait, like I know this is possible, I just need to find who can do it at an enterprise level, who can be a good AI partner for us. So I would say, like that concept, that framework that you presented before, like logistics or physical operations or even industrials have been slow to adapt technology.
I think it's rapidly changing now. We're seeing that with customers, how they already even come to us with ideas like, hey, I know this can be done, just do it with us. We want to partner to implement this. Obviously there's still your enterprise security reviews and your CISO is still paranoid of AI making up some stuff here and there. But overall business is driving decisions and they're like telling it, or the security department is like, hey, let's figure out the problems about integrating with this startup, which at the end of the day that's still what we are. But let's figure out how to work with these guys because they can truly transform the business. So when that's there, the conversation is relatively easier.
[00:14:11] Speaker B: Right, right, right. So with this change in behavior on the customer side with AI as they adopt it as the sales cycle as you said, is getting shorter from a customer experience standpoint. Like, what kind of time or cost efficiencies are your customers realizing as they adopt Happy Robot?
[00:14:33] Speaker C: So it's gonna vary throughout the spectrum of industrials or physical operations. Like, we're serving today over 60 customers, over 100, if we include ongoing pilots and explorations, if you will. We're serving companies that are like over 50 billion in revenue. Some of them. Like, we. We actually were at Samsara, the Samsara conference last week, and we had the. We shared some of the customers we're working with. Some of these customers have a crazy number of use cases that we could be helping them with.
So what we are doing today is going after the highest ROI for them, right? Like, what is your highest volume use case? What is your most repetitive workflow where humans actually don't thrive? Or the uses that a human would never want to do? Like, we have customers where we're automating the outreach for carrier sales negotiation or for setting an appointment with Home Depot to deliver, or maybe reaching out to potential customers that you would want to work with, but you don't have the bandwidth to reach out to them and show them your products or your kind of like an SDR motion. Maybe they were at your website. Like, we have some customers on the LTL space. LTL brokerage space. They're reaching out to smaller shippers to say, like, hey, you shipped with us like a month ago. Like, do you want to ship again with us? Oh, I forgot about that. Yeah, send me over some voucher. I'll ship with you guys. Again, I forgot. So we're enabling these many different use cases. So we're seeing obviously different ROIs, but as a whole, what we're seeing is their employees are able to focus on what matters, on their relationship that they want to build with their own customers and partners. They don't have to do the repetitive work.
They're sometimes even getting better margins when, for example, negotiating rates for loads. We have customers using us for sales negotiation. Specifically on the brokerage piece, the body's negotiating better rates just because it sticks to the guardrails that it's given. We have customers doing payment collections, reaching out to their own customers and telling them, please, can you pay me?
If you have a human doing that, it's something I would not want to do myself. Who wants to call a customer and tell them the bad news that, hey, you haven't paid. Can you please pay me? So an AI can just do that very formally, very respectfully, Very efficiently and give the customer ROIs as crazy as 100 eggs ROI. This is stupid, but we've seen like 100 eggs ROI in a case study we did on specifically on payment collections.
So it's all around, it's all across, but ultimately it's a matter of efficiency. But also, not only that, it's also creating new opportunities, which this is kind of the big mindset shift. It's not about automating what's already there. It's like what can you now do with an AI agent that can communicate in parallel with multiple suppliers, with multiple customers? Serve all of your after hours customer support lines.
The ideas are just endless. So it's fascinating.
[00:17:52] Speaker B: That's an explosive growth, meaning you're getting a large number of use cases, large number of customers, both small and large.
So how do you scale a platform for handling all these varied use cases while maintaining the performance and SLAs and expectations that folks have?
[00:18:14] Speaker C: Transparently, we are focusing more on the enterprise and we're very thankful that we're able to work with some of the largest players out there in supply chain logistics, physical operations. We're even working with some of the largest ocean carriers which move all the containers in the world almost. So we're able to tap into this enterprise layer of supply chain and physical operations.
That is what really helps us drive this growth because we do have a very wide glove support where customers today are relying on us to connect to their own systems, to build the agents versus them having to train people internally to learn how to do things like prompt engineering. That's the thing. Today you have to tell your AI what you want to do. Like I see AI as a, or LLMs as a very basic building block. It's like basic intelligence. But you still need to tell that AI what to do. Same way that you hire someone that is new to an industry and you need to tell them like this is what happens in these use cases. You're going to have to call a supplier and give them the po. Oh, what is a po? Oh, okay. So there's some level of learning that goes into it. What comes with us is that we've already done multiple of these use cases. So we come to the customer and we have templates.
So we bring templates to the customer, we bring templatized use cases to the customer, we bring our expertise, our engineers. We have this so called forward deployed Engineer. It's a very weird name, but it's pretty much reflecting the reality of today's deployments of AI. These FDEs or forward deployed engineers, they're really key in implementing these AIs because they hold both the technology knowledge and they also have implemented similar use cases in our industry. So then it's a very easy conversation with the customer when you tell them what to do versus them having to explain from scratch what they do as a company.
[00:20:22] Speaker B: So there's been a lot of learnings as you kind of start started with machine vision to happy robots and then figuring out how to scale it.
So for first time founders who are looking to build a new company for we're using AI in industrials, what is the one piece of advice that you could provide to these founders?
[00:20:48] Speaker C: I've been asked this question before and I think I always typically say conferences.
Being at a conference or visiting the customer in person is what's giving us a huge edge.
And we saw that early. If we hadn't started going to conferences where potential customers of ours were, we wouldn't have learned as fast as we did. So that fast iteration, going to those conferences, meeting the customer at their office, visiting them, sending someone like myself going there, sitting next to them, them looking at their screens, at their daily operation and understanding their problem.
That is really, especially when you're doing B2B enterprise, like enterprise sales. Right. Obviously that's not applicable. If you're building like a B2C company, like a consumer product, that doesn't really apply, not in this fashion necessarily, but like sitting next to your customer and hearing them, understanding what's a pain for them, that's huge. Right. So we try to stay as close to through that as we can. Obviously as we scale there's growing pains and now we're like trying to. We cannot be at a customer for like an entire year, but if we could, I would just have one engineer per customer for forever.
And that's really when you learn a lot about their operations and what you can bring back to your product. And really what to your point before now, like what sustains as well our growth is having a.
We have two products, we have platform, which is kind of like our building blocks, like where our FDEs can build new AI agents very quickly for different channels, email, voice, text. And then we have another product which is called Bridge, which is meant as that control tower. Bridge as in the space control tower of a spaceship almost. So that control tower for the operators, for the customer really to understand what's happening at a high level. So we're investing heavily on that aspect as well. How can we provide visibility and better observability to the customer so they know at any point in time what the agent is doing.
[00:23:07] Speaker B: That's a fascinating story and journey about happy robots and thanks for sharing those insights. I see heavy emphasis on AI and AI agents.
I'm going to switch the conversation towards your thoughts on future of AI agents or future with AI agents, whichever way we want to go. So there's an ongoing debate whether AI will argument a human worker or replace a human worker. And as happy robot engages with customers both in the frontline field, how do you see AI agents fitting into these future workflows in the industries that you are operating in, such as logistics, supply chain and others that you will get into?
[00:23:50] Speaker C: I think this is best explained with an example.
Last, last October, last Halloween, we had a customer send us a picture of one of their reps that has, that had dressed up as a little robot face with like some, some long hair.
That's because the agent that we deployed to this customer was called Kate.
They had dressed up as Kate.
This was a carrier sales rep in one of the largest freight brokers in the us he had dressed up as the bot that is theoretically taking away their job. That's because this is truly augmenting their day to day. This is making them be way more efficient in their case, cover more freight than they could on like traditional tooling. At the end of the day, humans are just interfaces to data in many roles today that is fine if the level of reasoning that you need is extremely high or like if the level of relationship with the customer has to be really, really high, meaning you have to show face with the customer, be there and talk to them physically. But there's a lot of interactions where humans are just a lot of instances where humans are just interfaces to data and they're just looking at one screen, copying that number, putting it in another screen, sending one email, waiting for that customer to reply or that partner to reply back. That's a huge time sink and that's again something that one can get frustrated about. Now if you have an AI 10x ing you, you're going to be able to do more yourself, which basically making them more, win more.
They're looking better in the eyes of their managers, they're more productive, which ultimately it's really what we're looking for. How can we augment the employees in these companies and enable them with better tooling and better assisted AI? Really obviously there are going to be use cases where maybe a human doesn't even need to do it anymore, like payment collections. Again, who really wants to Send an email to a customer you didn't pay, that's like something. Now they're winning time back from that and they can maybe reach out to the customer to tell them how great the day is going for them and see if they're also doing great.
[00:26:21] Speaker B: That's a great example of how the agent is compounding what a human was doing, an argument. So that's that augmenting story. And in the payments collection is a great example of how we are compressing it. And that, hey, those 10 step workflows that you had, just have the agent negotiate and close the loop out. So instead of going through 10 steps, now we are just doing two steps. Others have been eliminated and no one wanted to do that anyways.
[00:26:46] Speaker C: Exactly, Exactly.
[00:26:49] Speaker B: Awesome. So then from your vantage point, as you look at some of the AI capabilities and deploying them at scale, what are some capabilities that have not yet been fully realized and brought into the workforce or into the market that you see emerging on the horizon?
[00:27:07] Speaker C: We ourselves have realized that what we've built so far has been more of an agentic workflow, but we haven't really built like truly smart AI workers.
And maybe I'm just being too critical with ourselves.
We're very excited about what we're building, but also very critical with our work and we want to always do better. So let me just explain. So what we've built so far is really an AI that can communicate in a very structured manner to some degree. You're not like letting it think about how to do the work better. You're telling it what to do.
[00:27:54] Speaker B: You're telling it the before and after exactly like the.
[00:27:58] Speaker C: You're telling it what when to maybe trigger a phone call or an email and what to do after collecting that information. You're like being very prescriptive about it. For example, if you need to, for example, set up an appointment or maybe even reach out. I was talking with a manufacturing company and they have to reach out to suppliers to see when the party is going to arrive.
A buyer, a procurement person today is just looking at that PO and seeing what is the commit date and when supposedly that part is going to arrive.
What we could do today is tell an AI, hey, make a phone call, make a phone call to the supplier and tell them that, ask them when the part is going to arrive and then write it back into a spreadsheet or write it back into our SAP instance or our erp.
What we're building towards is the AI should be able to every morning, every Monday morning, every Week, whenever, look at your ERP and be able to proactively make decisions or at least ask the manager, hey, am I good to send this email? I'm going to send this email because this part is going to arrive late. Am I good with that? Oh yes, do that. At some point the manager doesn't even have to say do that. It just monitors all around.
So we're building towards that which goes back again to the point of that rep, that representative, that buyer, that procurement manager now is going to be able to do a lot more on their own. So how do we go from agentic workflows to truly smart and truly smart AI workers? That's kind of our differentiator. We're going from agentic workflows which is really perfectly fine. It's super high roi, something kind of next level which is an AI worker that truly is thinking about, hey, how do I make myself more efficient? It's almost like self fixing itself. It doesn't have to get feedback from a user like hey, you messed up here. Next time you should say that the AI on its own should be able to self correct next time by reasoning about the mistakes it's done. So we're building towards that and I think that's a bit of an unsolved challenge in AI as a whole.
[00:30:22] Speaker B: Really Got it. No, that's super helpful. And as you're describing it, I'm kind of connecting this back to the industrial roi. So right now, as you said, the prescriptive workflows, current tools are great, you can put them in.
How do you think about agentic AI or AI in general, helping industrials think differently as they shift the ROI equation.
As a case in point, like you touched upon it, like hey, today the agent is following a flow of what the beginning is, steps it has to take and it's very tightly controlled. How can you open the door to intelligence or data? So as you said, look into the erp, look into orders, the ordering system and make either decisions or suggestions or maybe suggestions is the first step. So how should industrials think about it? As you, you're on that, you're on the cusp of the journey, you're solved. Step one. Now we are staring at step two. What advice do you have?
[00:31:23] Speaker C: Yes, I would say the biggest piece of advice would be to start simple and see the value, to basically just let us or other vendors get access to your day to day operations and work together with your vendor to see where you could apply.
Call it AI, call it automation, to make that more efficient. There might be things that surprise you as a manager, as, as an executive in terms of. Oh, like hadn't thought about that use case.
Yes, that's a great point. We could be making 10,000 calls in parallel to reach out to our suppliers 10 days ahead and make sure that we don't have any delays.
That's something that until you don't see this working, sometimes we're like not seeing the whole picture, but starting small with a partner and understanding what they can do and where that could be applied.
That is what I recommend. Customers we've had customers or folks we've talked to in the past that have taken a few months to get on our platform and at some point they complain, we should have started earlier.
You could have. You just were a little bit scared to start when someone else started. That someone already is seeing millions in savings and in top line revenue generation. In certain use cases you just took like six months more, which is fine.
We're not all like the same level of tech savviness or not, not even tech savviness, just like, like tech adoption is obviously there's different, different appetites for that, but the longer you take to make the decision, the, the, the more in a disadvantaged position you are, if you will. So I would urge folks to like at least look at what, what parts of their business could benefit from some form of automation. Call it AI, call it smart automation.
[00:33:24] Speaker B: Hey Pablo, thank you so much for your time. That was super helpful and I love the closing state thoughts you have around. You have to just start. The longer you wait, you're missing out. Longer you wait, you're not seeing the use cases that you have seen by doing the adoption and you're missing out on those opportunities. So with that, thanks a lot for your time, appreciate all your insights and good luck to the happy robots and we'll be looking out at Gen 2 of what's in store with your product and company.
[00:33:53] Speaker C: Appreciate it. Thank you Prates. Thank you.
[00:33:55] Speaker B: Thank you. Take care.
[00:34:02] Speaker A: Thanks for listening to Aina Insights. Please visit Aina AI for more podcasts, publications and events on developments shaping the industrial and industrial technology sector.
