Latest / Autonomy Insiders / The Hidden Human Layer Making Autonomous Vehicles Scalable | Nick Allen, TaskUs

Transcript

The Hidden Human Layer Making Autonomous Vehicles Scalable | Nick Allen, TaskUs · Autonomy Insiders

• Daniel Abreu Marques: Welcome to Autonomy Insiders, the show where global industry leaders in autonomous driving unpack their real-world insights. Nick Allen: so the human of the loop side sits from a cost perspective on the OpEx piece, and it it directly is this like impacting the unit economics. where we come into play is really about the operational efficiency as well as the utilization and pushing that to its boundaries of excellence. a current healthy commercial ratio is probably sitting between one is to twenty to one is to fifty. And that that's really gonna depend on the maturity of the stack as well as the complexity of the city or the market that they're driving in. Daniel Abreu Marques: Welcome back to Autonomy Insiders. I'm your host, Daniel, and most of this audience can name every virtual driver stack. Far fewer can name the companies running the human layer underneath those fleets. And today we go behind that curtain. My guest today is Nick Allen, who leads Autonomous Vehicle Mobility and Logistics Solution at TaskUs And TaskUs is a publicly traded digital services firm with more than 65,000 people across about 13 countries in AV. They do the work that rarely makes the press release, data annotation, remote assistance, fleet response, rider support, and the safety frameworks that let a fleet scale without breaking. So Nick, so happy to have you on the show today. Nick Allen: Thanks for having me, Daniel. Glad to be here as well. Daniel Abreu Marques: So to start, what does Task Us actually do for an autonomous vehicle company? Maybe walk me through as if I run a robotaxi fleet and we just met. Nick Allen: Yeah, of course. So, Daniel, if if you're running a robotaxi fleet, your engineering team is is brilliant at building the brain of the car. But what we build is the central nervous system. So we're connecting your brain to the messy real world. our team started off by doing data labeling and annotation work for, you know, in HD mapping and working through some of the geofence exploration and expansions that were taking place. And then that led naturally into remote assistance. so our teams are are not only doing the remote assistance, but also the rider support and the emergency response groups, field operations, including safety drivers, validation testing, structure testing, and then also roadside assistance and helping out with the fleets when a vehicle is not able to get unstuck and being able to respond to it in each individual market as well. Daniel Abreu Marques: And TaskUs is also known as a custom experience or also outsourcing company, but how did AV become a real line of business rather than a side project? And how big is it inside the company today? Nick Allen: Yeah, so it it started off nine years ago. Our first AV client had come to us and asked, you know, understanding our background with content moderation pretty heavily, you know, early on in our our client process. they had asked us to start doing some annotation work for em And one of our C suite members, Jared Johnson, had looked at this and said, you know, this is what the future is going to be. And we need to be able to jump on this early, become subject matter experts and really put the time and effort into it. And that is expanded piece by piece. over the last nine years to where now we run an end to end ecosystem across the entire board. Daniel Abreu Marques: Yeah, the whole AV thesis more or less is removing the human also to cut costs sometimes. what you're selling is more human back into the loop. So can you make the case for us? So why does an AV company actually come out ahead by adding your people rather than building towards zero? Nick Allen: Yeah, so I think building towards absolute zero takes a lot of time, but also takes a lot of testing. And that's really where we come into it because the real world has an infinite amount of long tails. even if your driver stack is 99.999% perfect, there's always that point zero zero one in a city like San Francisco or London, which means a car stopping dead in traffic every single day, blocking an intersection or freezing next to a construction site is gonna need some type of an intervention to take place. So if you try to build the stack to handle every single bizarre human behavior or weird weather event that is coming into play, it really takes a lot of RD costs, but also it takes a partner along that way from a human of the loop aspect to be able to assist in training the model, assisting your engineering group with the complexities and providing that real time data back to them so that they can be able to continue to evolve what they've built from a foundational model. Daniel Abreu Marques: And maybe staying on the cost theme. So one important metric is also cost per mile and for an AV operator, so where does the human support cost actually sit in that number? And how much of it can you realistically take out? Nick Allen: Yeah, so the human of the loop side sits from a cost perspective on the OpEx piece, and it it directly is this like impacting the unit economics. So if you think about like early stage fleets, you know, they're running a human cost at about a one to one ratio in terms of one human to one vehicle, or you know, up to like one to five, where we come into play is really about the operational efficiency as well as the utilization and pushing that to its boundaries of excellence. So when you're building an internal team, you're paying for those empty chairs during those off peak hours, where you have a lot of overhead coming into play. And then by leveraging our footprint, our automated routing, our cross train teams, that's where we're able to assist in building you towards a one to twenty, one to thirty, or one to fifty in a higher ratio perspective. we like to strip out a lot of the overhead from a facilities management piece. But we handle everything with your recruitment, your specialized training, as well as shrinking that cost per mile up to thirty to fifty percent. Daniel Abreu Marques: And when in a client's journey is the smartest moment to bring TaskUs in, so what do the best operators set up early that others maybe only figure out the hard way after a stalled launch? Nick Allen: Yeah, I I like to think the best time to bring us in is as early as possible. That allows us with a smaller fleet to get to know the vehicle, to get to know the company, and as well as to get to know your stack, both on the hardware and the software side. But we have taken over full fleets, upwards of a couple hundred at a time and have just been able to interject ourselves with with full efficiency and execution. where I look at it from a stance of what are the best operators doing early. Than others, it's really making and understanding what needs to be done for the future, not what needs to be done right in front of you. Because if you're planning on, the short term, then that long term creeps up pretty quickly. And being able to understand with the experiences that we've gone through, what will be coming down the road and planning for that future and planning for if failure takes place, how how is everything going to react in the system? I think that's where the best operators are currently looking at is understanding how. And being prepared for a failure versus thinking that it'll never happen Daniel Abreu Marques: And maybe now let let's open also a bit the toolbox. So from what I can see and also what you told before, you cover four or five distinct services. So maybe we go through everyone because we want to understand each one and which one is the real anchor. So maybe let's start with data annotation and sensor labeling. I think this is the oldest part of your work. Is this still a growth area for you? Nick Allen: Absolutely. by the nature of the work, it's constantly changing though. So it's no longer about 2D bounding boxes around pedestrians or obstacles or items that are on the road. today it's really about AI reinforcement, complex 3D semantic segmentation across multimodal sensors, including your LIDAR, your radar, thermal imaging, anything across the board. and then also scenario reconstruction and looking at what took place and what could have been better through that process. we're labeling highly specific edge cases that the car encountered yesterday so that the simulator can then train the model on it tonight and be better for tomorrow. It's far less about volume and now much more about high fidelity expert level labeling that's coming into play. Daniel Abreu Marques: And remote assistance and teleoperation is I would say where the real time work lives. So can you tell us a bit how your model actually works on that point? Nick Allen: Yeah, I would say the easiest way to explain it is that our setup functions, very similar to an air traffic control or tower. we're monitoring the entire fleet. We're directly integrated into our clients' systems and working on their applications. So what we're focusing on really is how we can impact the entire fleet and ensure that there's a high utilization across the board. if a vehicle's AI or its onboard system encounters Any situation where the confidence drops below a certain threshold, it pings our command center to at that point jump in, communicate with it, but always allow it to remain in control. we're working through the system to teach it if there's a temporary item that maybe it didn't recognize from the day before, or if there's a new environment interaction that it needs to be able to understand how it how it can interact with it in the future, or the best way to get around it in that moment. Daniel Abreu Marques: And I think also remote operations has somehow a perception problem right now. So there's a a loud narrative that these cars are quietly being driven by people in a room somewhere in the world. so what does the public and even some insiders maybe get wrong about that? Nick Allen: Yeah, I would I would say that there's there needs to be a clear distinction around it. So remote operations, literally driving like with a joystick or with it with a controller. that's where the human is in control of the steering wheel, the throttle, the brakes, everything. it's incredibly rare. It is highly dangerous at speed and it's fundamentally unscalable due to physics and it's just latency concerns. But remote operations primarily lives with your delivery robotics and things that are, you know, on the sidewalk needing to be piloted at that point. Remote assistance, though, is where the human is not driving the car. The car may be stopped, and then the human is looking at the scene and it's giving the car a hint or permission. That is allowing the ADS or the ODD to can stay in complete control and notifying that yes, this is the right action to take, or yes, that is the obstacle that you were expecting it to do. And then the car is proceeding based on its stack from there. and then remote management, fleet operations and response. This is really just managing the vehicle state. So understanding it's charging, if it has any damage to it, understanding if it needs to have a roadside technician delivered and to be able to address any of those concerns. But 99% of what we are doing in real time is remote assistance. the car is always doing the driving. We are just acting as the navigator, giving it permission to break the standard rules and understand what it needs to be able to do from a learning model Daniel Abreu Marques: Super interesting. And maybe we can also put some some number behind that for the audience. So if you think about remote assistance also versus actual remote driving, even though remote driving is not usually done by the AV companies, but what latency is genuinely acceptable as a between those two and maybe also what changes in how you build your operation once you cross that threshold? Nick Allen: Yeah, if if we look at remote driving, really your latency needs to be sub, I would say like 50 milliseconds, because that is as real time as it needs to be. And ensuring that there's no packet drops, no loss that's coming into play, because you're in essence steering a 4,000 pound vehicle or a few thousand pound vehicle, which has high safety implications. So you want that latency to be as low as possible if that ever has to come into play. On the remote assistance side. Because the vehicle is interacting and giving tasks, you know, you're looking at 100 to 150 milliseconds is really where you want to be. Because then you need to put yourself in the driver situation and the rider, not driver. So you wanna you want to look at it from the rider situation and understand, you know, how much time is passing between the vehicle moving and not moving. being able to ensure that there is the assistance coming into play on a on a real time basis, but always following the safety protocols that are all also in place. Daniel Abreu Marques: And staying on remote assistance, you also mentioned already some kind of ratios. and also it's often discussed that what kind of vehicle to operator ratio some companies are. what is your opinion on an a healthy ratio today Nick Allen: I I would say a current healthy commercial ratio is probably sitting between one is to twenty to one is to fifty. And that that's really gonna depend on the maturity of the stack as well as the complexity of the city or the market that they're driving in. Daniel Abreu Marques: And if we think about the ODD, so the ODD can also change, for example, in winter or summer. So does the ratio then also change if you think about different ODDs? Nick Allen: absolutely. honestly that's where it has to become incredibly dynamic because if you look at it in terms of a real-time aspect, in a calm, sunny suburban environment on a Tuesday, the operator can easily oversee or you know, the vehicles will be at a much larger ratio in terms of what you would be experiencing in a sudden winter storm, heavy rain and flooding, or localized traffic conditions. that's where the flexibility needs to be able to come into play to go from a one to twenty, one to thirty, possibly down to a one to five, and just making sure that within a few hours of a weather event coming in, you have that ability to really break the teams into their groups and ensure that they have that reaction timing to be able to assist in a much larger format. Daniel Abreu Marques: when I think about a number of one to thirty, one to forty, I think at the beginning it's hard to imagine how you can can do that. So is this also kind of a progression from an employee perspective? So if you are a new employee, you maybe start with one to five and then only if you gain experience and you can increase the ratio per employee or how does that work? Nick Allen: Yeah, so I I think that's also one thing to clarify is that when a task or a vehicle is communicating with an employee, they're monitoring that vehicle directly. So they're not monitor like they're not operating or assisting with, you know, 50 vehicles at a time. That's what the telematics and the software stack are monitoring from that piece. So we have eyes on the vehicles, but whenever there is a task or an interaction needed, then it's a dedicated individual focused on that one vehicle at that one time. but as our employees are becoming more educated or more experienced within the process, that's where they have, you know, certification levels coming into play to be able to handle more complex environments, work their way up into an emergency response team to be able to handle multiple conversations at the same time, multiple different environment interactions or scenarios that may be coming at them as well. Daniel Abreu Marques: Super interesting. And most of the time these are also in kind of remote operation centers located. And do you build and run these facilities then yourself or are they customer owned? Can you walk us through that? Nick Allen: Yeah, I I would say that on a more than often basis, the request is to work with our customers and their sites. And that's because if you think about the time, the effort, and the safety that's been put into these systems, they want to make sure that whoever is operating them has the experience and the foundation to do so. So we will build our way into our customer locations and be able to operate from their facilities. But we do have fully functional sites that are also supporting. different pieces of the AV operational support model. So we have TaskUs run sites as well. And that's why the flexible hybrid model comes into play where we can do a little bit of both from our client locations as well as from our own sites themselves. Daniel Abreu Marques: you already mentioned also the the long tail of edge cases. can you give us a concrete example of an edge case where your team resolved that the autonomy stack could not and also what happened next? Nick Allen: Yeah. think about when when a vehicle is driving down the road and they're coming up to an intersection, and within that intersection, there's a fire hydrant that is currently just spewing water out the side. At that point, the vehicle's path says it should be going forward, but that there is a hazardous condition or something that may be blocking. You know, that's what's going to be pinging over to our team to understand what is currently in place. You know, the first thing it may be asking is, is it safe to proceed forward? As our team says no, it's going to want to understand why. So that if it encounters it again, how can it interact without having to have another intervention in the future? So at that point, our team is walking it through step by step from a task basis just to be able to say, you know, there's a current hazard in place. It is temporary. You know, it's expected to not last more than a few hours. So if another vehicle in that fleet comes to that inner same intersection a few hours later and doesn't see it, then it's understanding that it's okay to proceed forward. But if it's still there, then our team is continuing to monitor and understand it how it needs to divert the fleet around and if it's gonna be a longer term solution. Daniel Abreu Marques: And maybe now let's switch to fleet response and also emergency support, what you also mentioned. so when something goes wrong on the road, what does your team actually do in the first sixty seconds and who has also the authority to act? Nick Allen: Yeah, so I'll work backwards. I think from an authority to act standpoint, all of our operators have the full authority to act and initiate the passenger as well as first responders, depending on the severity of the situation. So within the first 15 seconds, our team is looking at and doing a full situational triage and looking at what took place, what's currently happening, as well as checking in on the rider if there's one in the vehicle to ensure that they're safe and that they're okay. within the next 30 seconds after that, they're establishing full control. They're looking and seeing if the vehicle is currently stable, if it needs to be pulled over into a safe area, initiating that remote pullover at that point to a safe shoulder if the autonomous stack allows. And then at the same time, initiating outbound calls to either local emergency services as well as our field operations teams. And then for like the last 15 seconds of that first minute. That's where we're working through and understanding how we're going to be communicating and setting up the rider for the safest response, as well as being the most comfortable and just taking care of them in that moment and giving them a play by play of what they should expect in the next few minutes and how we will communicate and take care of all efforts with first responders when their next vehicle is going to be showing up so that we can get them on their way and then handling any and all needs that they may have in that exact. Daniel Abreu Marques: It's like a really structured process, so really good to hear. And you also do training and proficiency, so learning as a service more or less. So how much of your value is now also teaching the clients' own teams rather than doing the work for them? Nick Allen: I would say it's it's a massive value add that comes into play. because I always like to say that, the engineers who created and built these models are extremely smart people. When you operationalize that, it doesn't always translate well. And that's where our learning as a service process can come into play. we can assist with building out SOPs. We can also work through and be a part of any regulatory changes. I think as you know, in Europe as well as in the states. There's constant rules changing, either state by state or country by country. And that's where our teams sit in the middle and make sure that any changes that are being notified through the public or through our clients directly are being updated, communicated to both sides of the group. And then there is an assessment and recertification being done just to validate that everybody not only received the information, but is also processing it for future execution. Daniel Abreu Marques: So we now talked about a lot of services that you offer and what would also be interesting to know is how do you price all of that? Do you price it per seat or per per managed mile or a mix of it? Nick Allen: A a mix of it because if we look at it from like early conversations, it's gonna be typically about like an hourly model and being able to understand how you know we're gonna build ourselves in as well as the size of the team, the focus, the scope of work. But really, what we want to build into is more of a per mile or per task that removes a lot of the additional overhead. It puts the accountability on us from a utilization standpoint and also really helps reduce on that overall cost to our clients in the in the long run. Daniel Abreu Marques: let's talk about now what is live today. So I know much of this probably sits under an NDA. So tell me only what you can and frame the rest maybe by type. But without naming names you cannot name what kind of AV companies are you live with today? So are these robotaxi companies, delivery bots trucking drones? So where where's the volume? Nick Allen: Yeah, so we we're across the the entire spectrum of what you just talked about. So we're assisting and supporting robotaxi clients, delivery robotics, as well as trucking and freight. we are speaking with different drone groups as well, but we haven't really built ourselves into there yet. And where the volume sits is kind of in the same order that I put it. We started ourselves off in RoboTaxi. So that's where a majority of our volume is currently at. And then delivery drones came in next. So I would say that the next largest portion is gonna be with them. we are building up our AV freightwork, helping out from a fleet operations as well as a remote assistance perspective. Daniel Abreu Marques: And what changes in your work in the moment a client crosses also from a testing into a paid commercial service? Nick Allen: I would say everything changes because you shift from, potentially closed course RD to a strict service level operations on public roads. when that testing is coming into play, there's an ability, Daniel, for your engineering team to to reach out and say, Hey, I I'm looking to test A, B, and C and also wait a certain period of time to see how the vehicle is going to react. But in a commercial model, the time is of the essence because the rider who is sitting there is understanding that the vehicle is expected to take certain actions and the longer it sits, the longer it feels really. I mean, sit in silence for a minute and it feels like ten minutes at times. So that's really where the split second decisions need to come into play once that commercialization is in place. Daniel Abreu Marques: Makes sense and maybe now let's get to the bigger picture. You sit as we we just heard underneath many fleet at once. so maybe you see also pattern that no single operator sees. So from your perspective, what is the real bottleneck to scaling AVs right now? Nick Allen: I would say access to the technology as well as the framework that is currently regulating it. I was just over in Europe and realized that what the European countries are going through right now, where they have different rules and regulations compared to each other, is very similar to what the US has here, where the states are kind of providing what regulations and policies are in place versus a a federal framework for everybody to work off of the same rules. if if those are able to align, it'll make crossing borders a lot easier as well as the ability for companies to interact or take action the exact same way, no matter what country or state that they're currently in. And I think access really to the technology to allow, you know, future customers to be able to get into the vehicle, understand it, and that adoption and loyalty level to be built. is really something else because this technology is rolling out piece by piece and without it being available to everyone, then there's always that inability for every single person to really be able to take their first ride and understand how safe it is and really what what is to be expected for the future. Daniel Abreu Marques: Yeah. now let's also talk about the people that doing this work that you described earlier, because this is in the end the engine of your whole model. So more than sixty five thousand people in about thirteen countries. So where does the AV work actually get done? Is it more like concentrated in a few hubs or follow the sun across regions? And also why maybe there? Nick Allen: Yeah, so out of out of our total employee base for the company, we have about 2,700 people that are just dedicated to AV operational support for our current clients. that is rapidly growing month over month, but where that currently sits depends on the services that are being provided as well. Because if we look at field operations, obviously that has to sit within its market. So wherever the fleet is operating, our fleet operations, our roadside teams need to be present. From a remote assistance standpoint, we're able to do that out of operational hubs. you know, within the US, we have two or three different facilities that we are doing this from to be able to support our clients across the base. And then when you talk about some of the labeling and annotation work, we're able to do a follow the sun model and be able to build those into our centers of excellence where we have the experience as well as really the operational efficiencies that are being able to support them on a nonstop basis. Daniel Abreu Marques: And this work is also hard for human. So watching near misses, handling emergencies or repetitive labelling. So you also mentioned resiliency, screening and wellness. So what does that look like in practice? Nick Allen: so what that really looks like in practice is we have licensed therapists that we've built into this from a wellness and resiliency standpoint to be able to support right from the beginning. So we're doing cognitive screening at the beginning to understand, you know, where everyone's mindset is currently at and what they're able to support from that perspective. And then as operations are taking place, we're also watching for fatigue management and looking to understand where we need to be able to have team members step away for a moment, have a break, be able to reset themselves. that also includes the ability, Daniel, to do pre-shift check-ins and mid-shift check-ins when they're coming back from lunch to really be able to understand where their mindset is currently at, how their morning has been so far, if there's anything distracting them, if there was an argument with you know at home that may have, taken their attention away. We really want to have their full attention and their full focus, on shift. That also goes for physical wellness. So if we see that somebody is not feeling well or they look a little bit ill. We would rather them notify us prior to going on to into production or on to shift because it's a safety protocol that we want to ensure that again, there is nothing that's gonna be taking their attention away. There's not gonna be any delay in their reactions, and we would rather be able to make sure that they are feeling well versus putting them in to a situation that may cause a safety implication as they're coming into place. at any time also. If there is a need to have an interaction with our resiliency team, they have that ability, no matter what time of day or what part of the shift that they're currently in, or even if they're off shift just at home, they have a twenty four hour service that they can call our teams and have open dialogue in terms of any support or wellness that they may need in that moment. Daniel Abreu Marques: you mentioned that you rather have them saying okay I have some trouble at home I'm not at 100% so this can also happen suddenly these are things that just occur so do you then always have kind of a buffer for such incidents that if someone is not feeling good you have someone right away that can jump in or how does it work? Nick Allen: Yeah, we do. we have a a buffer thats built in. It's typically about, ten to fifteen percent above. But we always have additional team members that are there just on our side to be able to support so that if anybody is sent home not feeling well or if somebody calls out for PTO, then at that point we're not dropping SLAs or utilization. We have the additional team that's there present being able to support the entire fleet and our clients as well. Daniel Abreu Marques: And this is also something that you can do centrally and not all your customers need to do by themselves because they would also need to plan in buffers and everything like that. So it's also kind of a real good value proposition for your customer. And I think also the AV operators care a lot about consistency across regions, probably. So how do you keep quality and safety identical? Nick Allen: Yeah, so we we are solving this through unified central operational governance. we have teams that are built in from a global ownership model to be able to support across each individual group, as well as from a safety perspective, a training perspective, or a compliance aspect. And we're building these out through our standard operating procedures that are then rolled out, executed across all of our teams. no matter what state or which region they're supporting from. And then ensuring that all of this is built into a learning management system as well that can be referenced at any time as well as looked back on in case that there were recent changes that they need to have some additional upskilling or communication on. And at that point, also our quality teams. Our quality teams are trained at the exact same level and same efforts as our operations teams to ensure that there's nothing that they're looking at differently. Than what our operators are doing on a daily basis. And then finally, I would say that localized training. So as we're going into a new market, everybody is going to be trained on the local laws and regulations as well. in order to ensure that as you know, we are running these different fleets and different ownerships across regions, that we have the right approach, the right safety and the right communication levels being done for the local teams as well as the quality teams that are modern training them as well Daniel Abreu Marques: And maybe staying also on new markets. we can use this as a bridge to a topic that I love to dig into because Europe is finally moving on AVs, and you also have started to show up here. So maybe let's go deep on the European plan from TaskUS So Europe is emerging later than the US and China on AV. how real is the opportunity for you here today versus 2027 or 2028? And what specifically convinced you to lead in now? Nick Allen: Yeah, I mean the the opportunity in Europe is happening right now. we're seeing it day after day. Different companies are emerging, different countries are being announced. While it may be true that the US and China led with the initial commercialization of robotaxi deployments, I would say that Europe has quietly been laying a far more standardized framework that's coming into play. the progression of European automated driving regulations as well as the UK's Automated Vehicle Act, what we're seeing is a massive shift from localized RD testing to really cross-border commercial plans being rolled out on a regular basis. We're leaning in because our clients, as well as a few of our prospects, have requested that we come in and we look and understand how we can support them early on, as well as if we have the ability to go into new markets ahead of time to ensure that what is going to be required is present. that it is built early versus having to be reactive as the the structure is launching within those new cities. Daniel Abreu Marques: And what services do you intend to offer in Europe first? So do you want to to offer the full stack or more a narrow entry point? Nick Allen: Yeah, so we're we're open to offering the full stack, but what we're seeing is a lot of requests that are coming through from more of a safety driving and validation work, structure testing, high speed to be included. that is a lot of the requests that are being made right now. And that's because when a new market is opening, obviously you want to ensure that the ODD is communicating and reacting to the city, the community, as well as the different, local weather patterns that may be in place. So a lot of our focus has really been on those validation efforts and our safety driving teams coming into play for the initial rollouts. But at this time we have open dialogue with offering the full stack of our services. Daniel Abreu Marques: And maybe let's also get a bit more concrete. So you're building a site in Manchester right now. So how many people are you bringing on there and also what exactly will that operation do? Nick Allen: Yeah, so at as of right now, we're looking at about forty-five, but in reality by the end of the year that could nearly double. our teams are focused on the the validation work within the market, you know, ensuring that the stack is responding as it's expected to, and then also building in some of our roadside efforts to be able to support the fleet from a management standpoint, ready vehicle readiness and charging and cleanliness side, and then being able to expand that out. across the UK as well as upwards of five to seven other countries in Europe. Daniel Abreu Marques: Yeah, and as we as we talked about and as you mentioned right now, Europe is not one market. So you have Germany, France, the Nordics, the UK. They have different rules, they have different languages. And we also mentioned that you ha starting in UK right now. But how do you actually want to to enter Europe afterwards? Is it then more country by country? Do you want to have some blocks or how do you plan to do that? Nick Allen: would say that you have to respect the localization requirements while also capitalizing on the regulatory commonalities. So where there is an ability to have similar rules and regulations in place, it makes it a lot easier to come in from a block perspective. But with us, you know, in the requests that are being made, we're going into this country by country at the moment. that's also being done because of our flexibility and the ability to adapt as needed versus having a very, rigorous like only cookie cutter model. We we come at this with different options, how we can jump in from a localization aspect, as well as working with local regulators well ahead of commercial rollout, just to be able to understand what needs to be there, such as if the remote remote assistance teams have to be built in country or if they're able to be centralized at a command center. Daniel Abreu Marques: Yeah, so maybe let's dig into that deeper because as you mentioned, some European markets lean towards requiring the remote operator to physically sit in the same country as the vehicle. So how do you think about that and how much does it also shape your build out in your Europe strategy? Nick Allen: it's shaping it in the sense of we may have to build individual sites within those countries that are going to require the operators to be able to be within the country. But in the long run, I do believe that there will be an option to eventually roll this out into a central location. But all that we're doing ahead of time is really that's you know, going back to the previous answer, looking at what we need to do with local regulators, how they are currently requiring it. As well as the safety measures and expectations that are in place. It's just requiring us to be one to two months ahead versus just going in and building piece by piece. But our teams are really hitting the ground running by being able to understand the specific countries that are going to be demanding these aspects and ensuring that we are meeting those qualifications ahead of time. Daniel Abreu Marques: And I think data is also another hard part in Europe. So GDPR data residency restrictions on where annotation and video review can happen. So does it force you to also build operations inside Europe more rather than serving it from other regions as well? Nick Allen: Yeah, absolutely. from a GDPR data restriction aspect, it it's non-negotiable. It's not something that we're trying to work around or trying to have different views into. We want to make sure that there's no blurred lines there, that everything that is being captured from a high definition, public camera or anybody that is walking the street from or a passenger within the cab knows that all of that data is remaining within the EU borders. Daniel Abreu Marques: And I think from from an cost per mile perspective, European labor is more expensive, maybe also more protected than in some of your existing hubs. do you need to adapt your model there or how do you think about that in economic terms? Nick Allen: I think that's where a model works even better. So while we look at that from like, yes, the the labor is more expensive and sometimes the cities are more complex to operate into. if you're building all of this internally, there's a lot of intricacies as well as a lot of facility overhead that has to come into play. but being able to maintain a 247 compliant operations team is really where we're able to maximize the utilization as well as you know, amplifying the efforts with technology. cross training our talent pools, advanced routing software to ensure that the European operators are highly utilized across multiple non competing tasks, or that the fleets during low frequency hours are, having everything taken care of from a vehicle readiness standpoint. That's where we can come in and shine and really be able to even push our current model of what we've been doing even further. Daniel Abreu Marques: And if we look out five years from now, so what role does TaskUS want to own Nick Allen: Mm-hmm. Daniel Abreu Marques: in the AV ecosystem? So when a fleet scales, so what is the one thing you want them to feel they cannot do without you? Nick Allen: in in five years, I'm looking for us to still continue to be looked at as the indispensable partner for autonomous mobility, whether that's a small fleet or it's a global fleet at that point. When an AV company is designing either its next generation platform or it's looking at a new market to go into, what I'm looking at from five years from now is the question coming up from our clients or from these AV companies is really. How how do we build TaskUs into this model? How do we maintain the exact safety structure and management systems that we've built into place now and just continue to evolve as the technology improves and that we see these fleets really start to expand on a commercial scale Daniel Abreu Marques: And I think you are already moving into robotics and physical AI, so training humanoid systems. So how big is that adjacency and is the AV work maybe also a kind of an on ramp to something much larger than that? Nick Allen: Yeah, it's the adjacency is massive. AV is absolutely a perfect on-ramp into the robotics or the humanoid side because there's a lot of commonalities that are coming into play from the labeling and annotation work, the data collection work that's going to be required, and then currently as well as in the future, the remote assistance that's going to be in place and monitoring on a ratio basis. different pilot operators that are are gonna be monitoring a fleet of humanoids versus a fleet of vehicles themselves. Daniel Abreu Marques: And as the autonomy matures, so the human role clearly changes, rather than vanishes, maybe. how do you see the human in the loop layer evolving and where will the highest value human work sit by let's say 2030? Nick Allen: Yeah, so I I think by 2030, the human role transitions from completely being tactical reaction to a high level of cognitive supervision. Because if you think about the highest value of human work and where that's going to sit, it's really going to be at the intersection of complex ethics, cross industry orchestration, and then crisis management. So if you think about how delivery is being done today, there may be a model in the future to where you have a delivery truck. That is then interacting with a drone or robotics delivery coming off of the truck and having multiple steps in that process. And that's where the team is going to have to be able to be able to switch up how that operation is being done from an AV to an AD And then also from a crisis management standpoint, really being able to deliver on the emergency response and being able to ensure that everything that is being done today can even go further from not only if somebody is rear ending. an autonomous vehicle and the team interacting. But if there is something that is being witnessed by the vehicle, how else are we able to then communicate with local municipalities and emergency response to be able to get out to those areas as well as give them a heads up of what they're walking into or driving into from a scene aspect. Daniel Abreu Marques: That's super interesting. And I think this is a really important part that is sometimes really overlooked. and so happy that you shed some light today on all those aspects of the things you you have to do behind the curtains so that robotaxi fleets or in general autonomous fleets can drive on the roads. So Nick, this was a very insightful conversation and I'm so happy to had you on today. Thanks for your time. Nick Allen: Yeah, thanks, Daniel. Thanks for having me. It's great seeing you.