#268 Kiren Sekar: The Future of Physical Operations Is AI-Powered (Here's Why)

0:00We serve tens of thousands of customers across operations, including many of the most complex, largest operations, the power of the economy. And what's been really amazing is how bringing this data and visibility and then now adding AI has created impact. Things that were not possible even a year ago are now becoming possible today. So we really look at what are the things that kind of sort of barely work? And maybe it's too expensive. Maybe it's not reliable or accurate enough. But you can kind of see how it works. And if something looks like that with AI, chances are within six or 12 months, it's actually going to work really well.

0:38Ultimately, I think what's interesting is all of that's changing. What the customers care about has actually been very consistent. How do I make my operations safer? How do I make them more efficient? How do I make them more sustainable? Build the future of multi-agent software with agency. That's A-G-N-T-C-Y. The agency is an open source collective building the internet of agents. It's a collaborative layer where agents can discover, connect, and work across frameworks. For developers, this means standardized agent discovery tools, seamless protocols for interagent communication, and modular components to compose and scale multi-agent workflows.

1:24Join Crew AI, Langchain, Llama Index, Browserbase, Cisco, and dozens more. The agency is dropping code, specs, and services. No strings attached. Build with other engineers who care about high-quality multi-agent software. Visit agency.org and add your support. That's A-G-N-T-C-Y dot O-R-G. So why don't you start by introducing yourself? First, thanks for having me, Craig. Great to be here. My name is Kieran Saker. I lead our product engineering and design teams at Samsara. And I'm happy to introduce myself as well as the company. First, a bit about Samsara and what we do, and then I'll share context of how I came into this role.

2:20um samsara builds technology for the world of physical operations and uh that is the supply chains the construction companies the food and beverage producers the um you know the agricultural operations that grow our food uh the logistics networks like all of these different physical operations that really kind of power the economy that you and i depend on every day and I've been with Samsara since founding and actually I worked with our co-founders at a previous company, an IT networking company called Meraki and Meraki was founded in the early to mid 2000s. It was founded out of our founder's research at MIT and we were building wi-fi networks.

3:09This is the early days of wi-fi and we were really focused on how do you make it really really easy for businesses and larger organizations to deploy Wi-Fi. And so we were working with schools and hospitals and retailers and airports and everyone to help them get everyone connected and to do that in a way that was secure and manageable and all that fun stuff. And what we found was that everyone was putting in networks, everyone was getting connected, including these operations industries. But what was fascinating was in a school or retail environment or hospital, people were connecting iPads and smartphones.

3:58There were a lot of companies building really cutting edge and elegant, easy to use software for these businesses that were running on these networks. And in these operations worlds, they were getting connected, but then they were connecting in, you know, they look like, you know, 1980s mainframes, right? You see literal green screens that people are typing text into. And the systems, the operations were still pretty analog. There's a lot of pen and paper, a lot of big maps on the walls with post-it notes, a lot of making phone calls and tribal knowledge. And it really kind of struck us, these were big, complicated industries that are obviously super important to everything that we as consumers do.

4:46But they didn't have great visibility and great technology. So fast forward, Meraki ended up getting acquired by Cisco Systems for a little over a billion dollars in 2012. Myself and our co-founders, we spent a few years getting the business kind of situated within Cisco and on a good path. We transitioned the business to the next generation of leaders. The Meraki business and products are still doing great today. We see them everywhere, which is awesome. But we were ready to go build a new company. And we're really thinking about what are the opportunities to have really, really big impact. And we kept going back to these operations companies and saying this is such a big important part of our world and we know that if we bring visibility and data and software to a place that doesn't have it things just get better and we saw a couple of you know interesting um uh technology trends that we thought were going to make it possible to really digitize these these world of operations one was because of of cell phones, sensors and cameras that are put into our phones, we're getting really cheap and really inexpensive.

6:02So you can start to put them on everything. And two, thanks again to smartphones, starting to get wireless connectivity everywhere. So you can bring that data from the physical world into the cloud, into software. And then three, cloud computing was really becoming mature. So you could bring all the data together and find patterns and find meaning in it and we said hey there's there's something here let's go build a company really focused on building great technology for the world of operations and that was the genesis to uh to samsara to samsara fast forwarded a day uh company's about 10 years old um we serve tens of thousands of customers across operations including many of the most complex uh largest operations the power the economy and what's been really amazing is how bringing this this data and visibility and then now adding ai has created impact um so just to make it concrete one of our customers dhl um you know they operate uh uh delivery vehicles all over the place uh they are one of our our large customers and they're really focused on on safety and they found that with this uh through this technology, they were able to cut their accident-related costs by 49%.

7:22So effectively, one out of two dollars that were going towards accidents were saved. That's a big bottom line saving, but it actually means their workers are safer, right? And they're coming home to their families. And in the process, they actually, it's a very high turnover industry. They found that their job vacancies and their turnover got cut in half. So it actually made the experience better than for their frontline workers. So that's the type of impact that we started Samsara for. And it's really awesome to see that now happening at scale. And we're still going. Yeah. Not too long. Let me pause.

8:02Let me ask, first of all, what does Samsara mean? Yeah. So Samsara is a Sanskrit word about the eternal cycle of birth. and then life and then rebirth. And I think it was somewhat inspired by this being the second company that we were doing together and kind of going back from having a small startup, building it to success, and then transitioning and starting again. And then, you know, now we really think about it as, well, we have no plans for this company to be sold. We wanna be doing this for as long as possible. how are we kind of continually reinventing what we can do for customers? And so it's kind of taken on a new life, a new mean for us over time.

8:53Yeah. And you guys are public now? We are. Yeah. We went public in late 2021. I see. Has that changed how you operate at all? You know, I think it has actually had small changes that have generally been positive. You know, there's an overhead, right? So the quarterly reporting and the overhead that comes with it, it's a small kind of tax that a small number of us have to pay. Most employees, for most of our engineers, building products, folks serving customers, that doesn't impact what they're doing. But I think that it's been helpful in terms of actually attracting talent. There are a lot of really great engineers and product builders who want to know that what they're doing is going to have an impact on the world and that it's not going to shut down in a week or a month or a year.

9:58And so having that kind of certainty has been great for bringing in great, great builders. It's been helpful for customers who are looking for the same thing. It's amazing. A lot of our customers are, you know, 50 to 100 plus year old businesses. You don't see that very often in technology, but I'll go visit our customers and, you know, you'll see their photo galleries in their lobby and they'll show, you know, their first trucks that were like modified Model Ts, right? Like literally 100 year olds, sometimes you'll see before that black and white photos of, you know, like literally like people on horses and then early railroads and then their first, you know, automated systems.

10:40And then fast forward to today where you see, you know, these really advanced robots and state of the air technology. But they also want to know that their partners are with them for the long term. So I think at the margin, it's been helpful, but it hasn't made an impact on kind of the day-to-day most of our team. Yeah. And so are you guys primarily a networking company or are you a layer above that? No. You know, our previous company was focused on networking. Here, it's really about a system to give customers visibility into their operations and then insights into what can be improved and then tools to help them act on it.

11:23So practically speaking, most of what our customers interact with day in and day out is a software application. They log into our dashboard and they see a big map with all of their vehicles and their heavy equipment and their tools and assets and their frontline workers out in the field. And the system is just sucking in all of this data about what's going on. And then we surface insights to customers so they can see, hey, out of all of my drivers, for example, most of our customers have big vehicle fleets. these are the ones who are our safest and they're reducing risk for the business. You should go maybe recognize them or give them a bonus.

12:07Here are some folks who maybe have some bad habits. And here's specifically what you can coach them on to reduce their risk. And here's a button you can press and it'll actually push out training to them to help them maybe remember to not use their mobile phone while they're driving or operating heavy equipment, et cetera. And all of that's kind of powered by AI. And at the base of the system, we actually do make hardware devices as well that customers put into their vehicles, their assets, that's kind of sucking all that data up and into the cloud. Right. So that's what I was going towards. Are you doing the networking as well as the analysis?

12:46Or are you relying on someone, an existing network or some other company that's doing the networking and then you're doing the analysis. But it sounds as though if you're doing devices in the field that you're doing both sides. Is that right? Yeah. So from the customer's perspective, they're just buying a system. They're getting the hardware, the connectivity, the software, the networking and the connectivity. It's provided primarily primarily by the cellular providers, right? So we partner with 23 plus cellular carriers, all the folks that would connect our mobile phones as well. But because we can partner with so many of them and the devices can switch between them, there's very few dead spots and it kind of always works.

13:38And then we have, you know, think about it as like a giant family plan with millions of devices. That's what we manage on the back end. but from the customer's perspective they plug in and it just works yeah uh and and then all of that data from iot devices or uh handheld devices that uh people are using in the field or from vehicles all of that gets uh transmitted to the software layer in the cloud that that where the ai sits is out, right? Exactly. And there's, we have some, you know, on the topic of AI, we use AI in a lot of different ways. We actually have some AI running on the devices. So think about a vehicle driving on the road.

14:30These vehicles might drive for eight hours a day, five days a week. You know, you think about it, that is so many hours of footage and most of the time, nothing interesting is happening. So we actually have, we have dash cameras, for example, a little bit different than a dash camera that you might put in your car. It's actually running AI and you can think about it like a self-driving car, right? If you drive in a Waymo, it's got all of these cameras and sensors, And then it's got really expensive high-end hardware in the trunk that's processing all of that data. And it's basically understanding the world around it.

15:12And it's running all these models and figuring out how to drive the vehicle and how to handle all the corner cases. What we're doing is actually we have a much, much more lightweight version of that type of technology. And we've kind of shrunk it down to something that you could run on a dash camera in, you know, literally stick on any vehicle at a very approachable price point. And it's kind of running similar algorithms and it's understanding, hey, what do we think should be happening and then what's actually happening? And then when there is a deviation from that, it can actually, you know, alert the driver.

15:52So if there's a hazard coming up or maybe that detects that they're getting drowsy or they're following too close, it can give them a reminder. And then also if there's something that's kind of interesting that happens, it can then upload that video to the cloud. And if you try to upload all the video all the time, it would be insane amounts of data, hugely expensive. But that model allows the AI to run kind of at the edge. And then, you know, even with that model, we're still taking, you know, I think it's something like 14 trillion different data points from all of these exceptions, because those exceptions happen all the time at scale.

16:29And then we bring those up into our cloud. And then we have additional AI that can sift through that and help folks understand what's most relevant to me as an operator. yeah um and can you walk us uh through a a real world example of of of how this works uh and how uh the a the ai how ai is is uh comes into play at different points in the in the the workflow or you know from someone driving for example to the analysis to the alerts pushed out. I mean, for example, you said, I can't remember just whether one of the parcel delivery services had cut its, was it accident rate? Yeah, accident costs.

17:25Right. Can you explain that example to us? Sure. So at a high level, and we have lots and lots of examples like that, so it's not specific to DHL. But at a high level, what happens is for a customer that's focused on safety, they would deploy our dash cameras. And the dash cameras are running AI to basically look at a couple of different things, right? One is they're looking at risk factors for the driver. Is the driver doing something that's putting themselves at risk? And it'll actually give the driver an alert and give them an opportunity to correct that behavior. So that could be following too closely, driving too fast, being distracted, falling asleep, et cetera.

18:19So that's one aspect. Another aspect is how do you actually take that and then build a coaching and rewards program using that data? And one of the things that we found is that if all of it does is tell the driver when they're doing something wrong, there's actually worse results. And, you know, people don't tend to like it. Right. And these folks don't want to upset a fleet of a thousand drivers. Right. So actually saying, hey, how do you turn this not into just the thing that's going to tell you what you did wrong, but that's going to recognize good driving has been really powerful. And so DHL, who I mentioned, one of the things that they did is they actually put together a safe driver rewards program.

19:08And this is something that we've now built into our product and we help our customers put together. You know, but they will say, all right, you know, every month, you know, for each division, let's look at who are our safest drivers or who is above a certain score. And then we reward them. And in some cases, that's providing a cash bonus. In some cases, it's a Starbucks gift card. Sometimes it's just, you know, a letter from the from the from the leader or maybe the preferred parking spot or something like that. And then they start, it turns into a game, right? People want to compete with each other to see who can be at the top.

19:49And so the AI provides that information as well. And then it also helps understand how can people improve. So, you know, Jim has one instance of mobile phone usage, right? You remind them they stop it, it never happens again. Great. Joe actually has a repeat pattern where this is becoming a bad habit. Hey, here's, we can actually push down some proactive training and see their behavior improve. So it's doing all of that. And then, you know, the capabilities are just advancing. You know, it seems like every 30 days right now, there's more that we're adding to the product. And I'll give you a great example, one that I think is really exciting.

20:38Harsh braking is something that the system detects. So you slam on the brakes based on the type of vehicle, the environment, etc. It'll flag that. In the past, harsh braking was classified as a behavior you want to avoid. It often means that something else was going on. Now what happens is the AI can actually understand the video and say, oh, the person was driving down the road, they were attentive, they weren't speeding, they weren't following too close, and then someone unexpectedly just ran out in the middle of the street and they slammed on the brakes to avoid hitting a pedestrian. That's a great example of attentive defensive driving.

21:26We're actually going to raise their safety score. And so that's the kind of thing that these newer models are capable of doing. And so, you know, if we think about our technology, it started with just gathering data, right? And then it turned into, hey, how do we detect these events that we know are of interest? And now it's, hey, actually understand what's happening and look at all these patterns, look at exceptions, and then figure out the meaning and the insights. And so that's been really exciting to have that progression as we've got more and more data and then as the technology has been progressing.

21:59Yeah. And in that case, is it the AI that understands what's gone on or does it flag a human to come in and look at the video? Yeah. So in the past, people actually had humans looking at these videos to figure out what was what was happening. And now the AI can actually do it. The AI can actually look at a lot more context than would be practical for a human to look at. And it can say, hey, that person slammed on the brakes to avoid a pedestrian who jumped out unexpectedly. And they have been in the top 2 % of safety scores for the past two months. And they've been driving a lot, et cetera, et cetera, et cetera.

22:44This is someone who you want to go and recognize. And it just wouldn't be practical for a human to kind of sift through that all the time. But now it kind of tees up, this is what you really care about. Yeah. And most of the, what's the range of devices that you're tracking? I mean, you said that a lot of the customers have fleets, but what other sorts of networks or devices are feeding data into this? Yeah, so we got started with fleets, kind of first kind of looking at location and vehicles and then adding the video for safety. And we've just expanded over time. So now we track a lot of the assets that are out in the field that are part of customers' operations, like heavy construction equipment or dumpsters or tanks or all these different types of equipment, cranes.

23:38And actually, we recently launched a product that customers can use to track even small tools and assets. And it's really, really cool. It actually works kind of like an AirTag that you would use to track your keys. except its industrial strength and it can connect to all of the other different subsara connected vehicles and now customers are using that to track all different kinds of things that are valuable or critical to them so you know we've been talking a lot about safety the customers use the technology to save fuel and find fuel savings and fuel waste right we've got a customer in there, a large grocery delivery chain up in Canada, they're selling, they're saving about, you know, 46 ,000 gallons of diesel.

24:28That was in the first four months after. Right. That's like, you know, planting almost 8 ,000 trees and letting them grow for 10 years. We have customers who are using it for things like, how are you utilizing your assets effectively and saying, hey, actually, rather than going and buying more tools or buying more trailers, we find that there's patterns where certain ones are underutilized. We can go move them to where they can be put into effect. Or maintenance is another kind of classic use case. So all of these different areas, safety, efficiency, sustainability, that's kind of what's really critical to these operations industries.

25:04Yeah. And how large can these deployments be? I mean, is it infinitely scalable? Do you, what's the largest network of sensors or devices that you're managing? Yeah, you know, we actually serve some of the largest operators in the world now. DHL would be a great example. But, you know, our largest customers have tens of thousands of assets connected in the system, tens of thousands of drivers or frontline workers. But what's interesting about the operations industries is actually a lot of the businesses are mid-sized. And so we started with mid-sized businesses, then kind of worked our way up to the largest and most complicated over time.

25:53But that actually required us to make the product really easy to use. If you're a company with maybe 100 employees, you don't have experts in technology, you might not even have experts in areas like safety or operational efficiency. You've got people who really know about construction or really know about manufacturing a certain type of tool or what have you. And they're able to get the system up and running and working and kind of see similar types of benefits. So it's been interesting to learn about these industries and how, unlike some other industries that are very concentrated with a small number of businesses, it's actually pretty distributed.

26:38yeah and generally a a company like DHL then are there are you tracking by by region or is there sort of a master dashboard at the corporate level where they can see everything going on across the around the world or or is that left to the AI so somebody at the corporate level can query the AI you know how how or how is you know safety how are we doing in Asia or how are we doing in in North America That's exactly right. So both the AI as well as the dashboards, they're basically tailored for, as a user, what do you have access to? So what are you allowed to see? But then what's relevant to you?

27:38And if a lot of these businesses, they've got a lot of subdivisions, they've got different regions, a lot of them have acquired different businesses over time that have different characteristics. And so if you are a, you know, a regional supervisor, you probably want to know, hey, what's going on with my team of 50 workers, right? And that's the level you're looking at. And how am I coaching them? Who am I giving kudos to, etc. And then at the, you know, top corporate level, you're saying, hey, how are my divisions comparing against one another? How are the different regions comparing? What are the things that the, you know, Northeast division is doing right that i want to replicate through um through the rest of my my teams and actually now with with ai the um the the executives who might not be experts in the specifics of the operations can actually just talk to the system directly through samsara and ask these questions about hey which of my regions has the highest um safety score or fuel efficiency score What are they doing right?

28:49And what would I want to communicate to the other regions to help them? And then the individual safety manager can say, hey, can you help me come up with a coaching program for Sally to help with X, Y, and Z? And so that's been really powerful. We have the data, we have the data structured, but AI just makes it much more accessible. Yeah. So you built this from scratch, right? And how in building it, you know, there's, and then agents, I imagine, are going to play a role. How do you decide as a chief product officer where to focus to start implementing AI? I mean, because it presumably could be anywhere.

29:52I mean, how do you build a roadmap for developing the product? You know, it's actually the same with AI as it was before AI, but just the speed has become a lot faster. And the need to throw out what you thought you knew and hit refresh, that happens a lot faster as well. And so our philosophy is always start with a customer. We are not running physical operations and we're not experts in construction and logistics. We've learned a lot over time, but ultimately our customers are the ones who know their business is the best. And so we spend a lot of time out in the field with customers learning about what problems they're trying to solve, what's most important to them.

30:35We then start hearing patterns. And then we are experts in the technology and come up with ideas for how could technology solve these problems. And we start hearing about the same problem over and over again. And we come up with an idea for how technology could solve it. we then kind of figure out how do we make a prototype? We'll share it with customers, get their feedback. And they're very candid about, hey, this would be amazing. That's actually not that useful. Don't waste your time with that. And they guide us towards making things that are most relevant. And that was, you know, when we were building our very first products and figuring out what data should we keep connecting and what alerts and reports are useful and what can we help them see.

31:22And it's just that approach works just as well now with AI. But I do think that the thing that has changed is that the rate of change is so high that things that were not possible even a year ago are now becoming possible today. So we really look at what are the things that kind of sort of barely work and maybe it's too expensive, maybe it's not reliable or accurate enough, but you can kind of see how it works. And if something looks like that with AI, chances are within six or 12 months, it's actually going to work really well. And so that's how we think about where is it going? And then what are the things that can be really impactful in the near term?

32:10This can improve the quality of life for our customers. It could have another incremental 5 % reduction in accidents, which might mean many people's lives. Also, what are the really big things that might actually take several years to fully pan out, but could be really transformative? And how do we kind of explore those in parallel? well. Yeah. What have you come across any really surprising insights in looking at the data or the analysis of the data that's flowing into these systems? A ton. I think one is, I'll give you a couple of examples, but one is we look at the data and we see, hey, there's still a lot of paper out in the world.