# Selling Trust in Physical AI Product Strategy

**Podcast:** Product Momentum Podcast
**Published:** 2026-09-15

## Transcript

PMs at Samsara spend a ton of time on the road.
You know, we are visiting factories, we are out in the field, we are on the airstrip, we are in the oil refinery, like we are, we're like everywhere all the time.
And I think, you know, it's become like kind of this like, I don't know, kind of in vogue thing to talk about forward deployed engineering and that sort of thing.
Like Samsara's engineers and product managers have been doing that.
Since the beginning, they were going and visiting physical operations companies and figuring out what problems do you need solving?
And it sort of evolved from there.
So an example of something that you've that you've built for the physical world.
Walk us through like the sleepy driver intervention agent.
What did building that teach you about like designing AI for physical safety?
Yeah, I think that one's an interesting one because, you know, we're in the backdrop where everyone feels a little bit.
somewhat suspicious about having a second brain, myself included, like as I kind of do my work day to day and I offload tons of context into Claude and it's getting super smart and knows all the things.
I'm like, well, what am I like?
Where is my alpha?
Where am I adding value?
And I think the, you know, in the physical world, we have to overcome a lot of suspicion about the product, you know, and so.
We sell trust and product is about building trust and it's about helping people get home safely at night.
I mean, this is like, it's somewhat basic stuff.
And this is kind of like my highest level take on the way that product management has evolved over the last, I don't know, what, year and a half?
It's sort of all the things that...
like from Good PM, Bad PM, which is like my Bible for PMing, which was written probably in like the late 90s.
You read that and you're like, yeah, this is the job.
This is still the job.
Like you still need to like identify value, understand the market, like prototype quickly, work with your engineering team, be able to tell the right story, tell the ROI story to the customer.
You still need to do all those things.
You just have a different set of tools.
So what's next for you?
And what's next for like, where you see this designing products for like the designing AI products for the physical world?
Yeah.
You know, I think there's kind of two ways to answer it.
One from a craft perspective, I think, you know, we are all, we've all been kind of climbing the, this, this mountain in terms of like learning how to use the tools and all that stuff.
And I, and I come back to like, you know, the job hasn't changed.
You do need to just kind of be a connoisseur of all the things that are effed up about.
your customers experience and go and fix them.
That's like the, that is the job and building the right thing in the right way at the right time has always been the product person's job.
And it is now more than ever.
On the product side, you know, I think there's a lot that we are kind of discovering on the agentic front and we are going to see these kind of massive digital transformations of physical operations companies.
You know, it's just the mission stays the same and the tools get better and the craft stays the same.
Dan, we just got off with Arpan Potitore.
That was a super unique conversation for us, right?
I'm really excited for our listeners because this is a whole new perspective than we normally provide.
Yeah, you know, it was really cool getting the talk to a product leader whose focus is designing and developing solutions for workers out there doing physical jobs, right?
Like driving trucks.
Super cool to hear that perspective.
Also the perspective that, hey, as product people, we really sell trust, right?
That's the product.
So some really awesome stuff in the conversation.
I learned a lot, yeah, in the 30 minutes.
So let's get after it.
Let's do it.
Joining us on the pod today is Arpan Padaturi.
He's the vice president of product at Simsar, where he's leading product strategy and development of physical aid products that help protect frontline workers.
Before that, he was working at roles in such household names as Shopify, XE, ESPN, and others.
We're super excited to have you here today, Arcon.
Thanks for having me.
Pumped to be here.
So I want to jump right in into this idea of AI for, you know.
Physical products.
Yeah.
Sean and I spend a lot of time, you know, working with companies and talking about, you know, kind of B2B, B2C AI solutions that are more GPT based, more, you know, aimed at somebody sitting at a desk.
Yeah.
So first, can you just kind of tell us what it means to be developing AI solutions for physical in the physical world?
Yeah.
You know, so it's a.
bunch of different things, but we have models that run on the edge and they're very focused models that do like kind of custom detection.
So they find things that frontline workers are doing that are dangerous.
So if you're driving a vehicle and you are speeding or vaping or.
you know, falling asleep, like we have AI that detects that and alerts the driver that they are doing something dangerous.
And we've got those kind of use cases across industries and across different type of frontline workers.
So there's a lot that happens locally on the edge and fires in real time.
And then we have AI that also just helps the kind of.
back office, the like overall organization run more efficiently.
It streamlines a bunch of workflows, kind of does the things that you'd expect it to do.
So we sort of touch every aspect of a physical operations business from, you know, the kind of like people who are like working in the back office, crunching on a computer to like people that are like on an oil refinery doing really hard physical labor tasks.
So pretty fun to touch all of it.
So as a product manager, right, Are there differences in how you do discovery when you're working directly with a frontline person, supporting their physical labor versus somebody sitting at a desk doing an electronic workflow?
Yeah.
I mean, PMs at Sims R.S.
spend a ton of time on the road.
We are visiting factories.
We are out in the field.
We are on the airstrip.
We are in the oil refinery.
We are on the mines.
We're everywhere all the time.
You know, it's become like kind of this like, I don't know, kind of in vogue thing to talk about forward deployed engineering and that sort of thing.
Like, as far as engineers and product managers have been doing that since the beginning, like Sanjit, our CEO, and John, our CTO, basically built the company that way.
They were going and visiting physical operations companies and figuring out what problems do you need solving.
And it sort of evolved from there.
And we've just kind of taken that playbook and...
like rinse, repeat several, several times over again and, you know, have built out this product suite now.
It's funny you mentioned the Ford Deploy AI engineer because every time I see that it's like, but that's how you're supposed to.
How can you talk to users?
Is that not a new thing, right?
That's just going and talking to people.
Why did we come up with a new name for this?
Yeah, it sounds pretty cool.
It does.
You know, talk to users and build the things that they're, you know, having problems with or build solutions to their problems is kind of like the oldest song in the book.
Right, it's like bullet item one of product management.
Yeah.
When I go on site today, I'm going to use that term.
I'm going to give myself a new title today, forward deployed engineer.
That's right.
Really, really like that concept.
So an example of something that you've built for the physical world, walk us through this sleepy driver intervention agent.
What did Billy Knight teach you about designing AI for physical safety?
Yeah, I think that one's an interesting one because we're in the backdrop where everyone feels a little bit, I would say, somewhat suspicious about having a second brain, myself included, as I kind of do my work day to day and I offload tons of context into...
clawed and it's getting super smart and knows all the things I'm like, well, what am I like?
Where is my alpha?
Where am I adding value?
And I think the, you know, in the physical world, we have to overcome a lot of suspicion about the product, you know, and so we sell trust and the product is about building trust and it's about helping people get home safely at night.
And so.
You know, for us, what we have done, that use case is very helpful because it focuses on a use case that a human otherwise wouldn't do.
Usually drowsy driving happens like early, very early in the morning or late at night when no one's working in the back office.
And so historically, you would have an alert fire when, you know, if Sean is yawning or his head is nodding or his eyes are a little bit fluttering, we would have an alert fire.
And that alert would just kind of beep at the driver, right?
And a lot of organizations have automation.
So if they see that alert happen, it would go up to the back office and someone would call the driver.
Now that automation goes to an AI agent and the AI agent calls into the cab using the speaker that's in the camera.
And it will say, hey, Sean, like, it looks like you are getting a little sleepy.
Like, do you want me to kind of talk you through into your next rest stop?
And it's amazing.
I mean, it saves lives.
Like, it's like, it's this incredible thing that.
What's been helpful is like people, it's basically augmented coverage in a way that doesn't feel threatening to a back office employee.
It feels fully augmenting.
And that I think has been a really powerful use case for us.
And it's been interesting kind of like figuring out the right prompt and context and like actual experience for a driver there.
Like you don't want to be accusatory.
You don't want to be.
kind of like overly confident because a yawn is, it could just be a yawn, like, you know, and that's a very serious thing to feel like you're calling in about.
And so we've had to really iterate on that prompt to make it really comfortable and make it sort of like a neutral experience for a driver.
So that's been a, that's been a fun one to play, to work through.
Okay.
I actually have a little experience working with somebody who is creating AI solutions in like the trucking industry, but more on like, like the, The human resources side, right?
Like the payroll side.
They had a lot of unintended benefits of like a phone agent that they had.
Were there any unintended benefits of like the solutions that you've put forward where it was like, oh, we didn't even think of that case because we didn't find that discovery.
But yeah, it's not.
But like our users ended up finding that for us.
Yeah, yeah.
I mean, I think one of the things that makes our.
product experience different is it's not a set of pre-built agents.
Like we do not sell, we have templates, right?
And the templates are meant to be starting points for our customers.
But like when we forward deploy with our customers, we hear them say, you know what?
I really like that capability, but I need it to be specific to this use case.
And like our philosophy is the, like the software needs to conform to the business.
The business should not conform to the software.
And and so we built like a studio suite where, you know, we hear we have a geofence agent and it can basically do anything when you cross a geofence.
And so some users are like some customers are asking us, hey, when I get.
within one mile of my final destination, I want to have very specific instructions for that last mile to the driver.
And that was a job that a dispatcher would have to call every driver for every stop.
And now they can fully automate that.
Others will say, I want to make sure that my workers are compliant.
And so when they are out in the field, they have to have safety cones around their vehicles.
And when they...
when they get to their location, I want to make sure that the agent is prompting about that.
And so it's just a very generic kind of like capability set.
And our customers are, you know, really just tailoring it to their needs.
I mean, this is a thing when you find when you're working with huge enterprise companies that all have various kind of needs across, you know, different industries.
So it sounds like you built really effective feedback loops.
Yeah, very effective feedback loops.
Like we are agents are in the traces every day.
We are kind of like pulling pulling out the the gold that we can and iterating, you know, really quickly on this stuff.
So we're ripping a ripping some quotes here, right?
We sell trust, not products.
Yeah, the software has to meet the business, not vice versa.
So so loving that.
I'd go back to like the trust piece, because I think, yeah, especially now, you know.
AI tools are maybe doing the opposite, right?
So in the scenario that you were mentioning with the sleepy driver agent, it seems already you kind of have two different personas involved, right?
You have the back office worker and you have the driver.
I'm curious, how do you wean over trust for both of those workers who might be threatened by an AI agent coming in, even if it's, you know, the benefits are semi-obvious?
Yeah, I mean, it's it's it's it's tough.
You do have to prioritize.
Right.
So first, you have to win over the trust of a back office employee who needs to turn the alert on and set up the automation and the workflow.
At the same time, like they will hear feedback from their from their frontline workers and they will their their workers will say this is annoying or, you know, I'm getting alerted for things that aren't that are false positives or whatever it is.
And so.
It's, you know, what you find when you kind of work in this experience is like the more you make drivers happy and the more outcomes that you can deliver against, the happier everyone is.
And it does come back to ROI.
I mean, that's been such a big topic.
within AI for the last year or so.
And, you know, what makes, I think, Samsara's AI somewhat different is it's very targeted.
Like it's not like a generic kind of harness where you can kind of do anything under the sun.
It is, we have these templates, we have these capabilities.
Again, we sit with the customer and we build these workflows and we model out the ROI on the other side of it.
And so they're very, like our buyers are very old school and they have, you know, they have, very thin margins.
And so they are not kind of out here token maxing and just kind of telling everyone in their organizations to use all the workflows.
They're like, let's deploy this surgically and make sure we are getting value out of this.
And as we are, we will add more workflows on top.
And so that's really been the motion for us.
So I met all that in another conversation about ROX.
I think it's something the product community struggles with in general.
Yeah.
I'm curious, you know, in your experience without going into your company secrets, how you shape those ROI conversations with these enterprises where maybe they're using more of a traditional finance model and you're trying to bring, you know, the kind of 2026 AI world to that.
Yeah.
I mean, some of our, you know, I think, A, you have to explain the kind of unit economics of the problem.
Like, and I've been kind of building out our pricing model for our AI features at Samsara.
I think customers respect the transparency.
And if you tell them, like, this is our, like, tokens cost money.
And like, we work with the frontiers and we get these models.
We try to make this as efficient as possible for you.
Then you can have like kind of a basis of a conversation that's built on reality.
From there, I think it's, you know, we really focus on those first few use cases that are really concretely delivering ROI.
So an example would be, we have a feature called a ride-along.
And a ride-along is a thing that historically, if you're running a fleet safety program, you would have a kind of a coach sit with a driver and like with a kind of like a, you know, clipboard, like...
kind of monitor all the things and say, okay, like you're passing here or not, right?
And it's a very long process.
It takes two hours to do.
Like you're like basically, you know, you have to analyze it, put it together, like coach it.
We do that all with AI.
Like we basically can do a, you know, a ride along in minutes.
We can synthesize what we saw.
We can put together a rubric.
We can score it.
We can send it to a coach.
We can send it to a driver.
All that happens immediately.
There is very clear ROI there.
right?
Like you are saving hours and hours and hours of work.
You would be paying the person to do that.
And you couldn't scale that program.
And it's very effective at driving down crashes.
You couldn't scale that program without AI.
And so these are the kinds of use cases that we hone in on where it's like, this is a very, very clear replacement rate value.
But then there are also use cases that are just like, augmenting or scaling and their use cases at the same time.
And it's kind of like the PM has to use a lot of imagination to figure out like, all right, where am I going to calculate that ROI in a way that the customer will understand?
Yeah.
I want to pull on this thread a little bit because you've hit on two things that I think will be kind of important for our audience, right?
Is a lot of times in product management, we are somewhat upstream or downstream of like the pricing model.
We don't actually get to like a lot of our listeners that we think about working for large enterprise corporations, the pricing model is kind of like out of reach for us.
So I want to hear a little bit more about that, but especially what are the challenges that you've had with the pricing model and maybe like the changes in AI token costs?
Yeah.
I mean, the main thing that our customers were not happy about was an open-ended model, like a consumption-based open-ended model.
Like our customers, they need to build a budget.
request funds like have a predictable spend level with their vendors and we're you know trying to kind of meet them where they are and so we've built i i think there's a reason why all the frontiers are basically packaged this way as well they have like effectively like sass tiers with usage underneath and so we've built something very similar and i think it just gives people the predictability that yeah if i if i give you this much per month, I'm getting this many credits and I can draw down those credits and I can do whatever I want with them.
I have a lot of flexibility in terms of how I use them.
But like the transparency that we've also built into the system to know exactly where you are, which users are using the most credits, that sort of thing.
Like it's a model that the frontiers have really, I think, done a good job with.
And we've sort of picked up that model and said, yeah, that's the right way to kind of package all this stuff up.
Yeah, you've mentioned that like your customer base a little bit, some of your customer base comes from the old school.
So I imagine this is kind of like a difficult conversation to have because you're not only talking about pricing, but you're also talking about brand new technology and how it's being used, like what the benefit is.
Are there any ROI arguments that you've had that like haven't held up?
And like, what did you learn from those?
Yeah, it's a good question.
I mean, it's not that they haven't held up.
It's that it's just hard to like, it's hard to point materially to like this specific use cases.
creating this ROI, right?
So an example would be like a driver profile.
We have a ton of data about a driver and we use AI to synthesize that and come up with a quick profile.
Like the synthesis alone probably saves, you know, this much time and that sort of thing.
But like we have that use case multiplied in a hundred different places across an organization.
And so there's a lot of these kind of like summary.
synthesize, send the PDF use case, which sure, we could like kind of do the math and back into ROI and say like, yeah, this is actually saving you time and therefore it's important.
But I think everyone kind of intuitively understands that we just, we don't need to like do the arithmetic for them in those cases.
And so there's an element of ROI that's just like, yeah, it's just like greasing the wheels across the organization in every single corner.
And therefore, we're okay with a baseline level of sure we're going to pay for it because it makes us faster in some cases we are very explicit about roi and so in others we are we sort of like take the company stance that ai is an accelerant across the board right so you yeah you mentioned something that i've just kind of been puzzling over i think you know a lot of us in the proc media when we do have those roi conversations yeah i was like hey I want to solve this problem.
My back, the napkin, you know, back, the envelope math is the ROI will be this.
And you mentioned the part of evaluating it with your customers.
So would you mind like elaborating on how you go through that evaluation process and kind of that feedback loop to keep driving value?
Yeah.
I mean, it really is going through the use cases specifically, deploying those use cases.
Oftentimes we deploy those use cases in trials.
And so like for our...
like AI detections that run on the edge.
Like they all go through a free trial for 90 days.
We, we evaluate, you know, behaviors that were happening at a certain rate before and behaviors that are happening at a certain rate after you can view crash rates.
You can see accident rates inside of, you know, inside of construction sites.
There's like a bunch of different factors, but it boils down to like.
I mean, this is like, it's somewhat basic stuff.
And this is kind of like my highest level take on the way that product management has evolved over the last, I don't know, what, year and a half?
It's sort of all the things that, like from Good PM, Bad PM, which is like my Bible for PMing, which was written probably in like the late 90s.
You read that and you're like, yeah, this is the job.
This is still the job.
Like you still need to like identify value, understand the market, like prototype quickly, work with your engineering team, be able to tell the right story, tell the ROI story to the customer.
You still need to do all those things.
You just have a different set of tools.
And so that's kind of the approach that we've been taking.
I like that take that the job really hasn't changed, but the tools have.
So knowing the job hasn't really changed, but the tools have.
I want to kind of circle back to, you know, we kind of...
Thanks for taking us down that, the pricing and the pricing model thread that we wanted to pull on there.
But I want to kind of get back to, you know, the workforce that you've had to work with, knowing that they are in the physical world and we're kind of introducing these new capabilities kind of like around their world.
What's it like, kind of like, how do you approach that kind of what I'm going to assume is kind of like a resistant workforce?
You know, before I worked at Sims, I worked at Shopify and we we I ran the point of sale business there.
And it was a similar like it was a similar kind of problem to approach in that we had retail workers, large kind of retail brands.
And so you could think of like, like, I don't know, Parachute Home or all like any of like the digitally native brands that are opening stores.
Right.
And they had.
hundreds or thousands of retail workers.
And the thing that upset our customers the most was feeling like they lost control.
And this is the thing that like drives me nuts about Claude.
When I'm using Claude and it's like, do you want me to send a Slack message?
And like, I have a draft already written in Slack.
I'm like, hang on, I'm going to like put you in a box and put you in a corner because I don't want you speaking for me.
I don't want to feel the loss of control.
And I think...
For frontline workers, it's the same.
They don't want to feel a loss of control.
For their back office managers, they don't want to feel a loss of control.
They want to feel like the tool is working for them and is driven through their desires, right?
And so it puts a lot of onus on like product marketing and how do we tell the story and how do we launch and is our launch cadence predictable?
And are we doing the right education all the way through?
And do we have the right...
settings in place such that every organization can roll this thing out in a way that makes their back office look like they're in control.
And the more that we take, if you bite into that, or if you're trying to go too fast, or if you're trying to deploy really quickly, you will, it will come back to haunt you.
They will revolt.
And I think I see this in the tooling that I use every day.
And it's the thing that we just, again, the product is trust.
And so what are the programs you have in place to reinforce that and make your customers absolute stars, make them look incredible inside of their organizations?
That's a big part of kind of how we think about building product.
I mean, that is...
Tremendous.
I often talk with colleagues about, you know, sometimes when we're building tech, we think of like the tech is the thing, but like really we're kind of tool makers, right?
Like we're building stuff to help people do their jobs.
That's the goal.
And I think as a product community, sometimes you maybe lose sight of where we actually fit into this ecosystem.
So I think, you know, that's just such a critical aspect of this job is the jazz piece.
You gave us a lot of quotes today.
So I want to like, I want to go through some of my initial takeaways and how it kind of ties back to some of the things that we talk about at ITX.
We've said we sell trust, not products.
Right.
And we strongly believe in like this concept called the relationship ladder where you move from, you know, no trust to trust, to loyalty into advocacy and advocacy is where those are the behaviors that we want to inspire.
Right.
So great products really end up creating effective changes in behavior.
Right.
That are going to be measurable in some way.
So I just love that concept of we sell trust, not products.
That really ties into what we do.
Also, you know, early on in our conversation, we talked about feedback loops and feedback loops essentially are the result of like effective advocacy, right?
You're not going to get effective feedback if people don't trust that you're going to use it for good, right?
And if you're not getting that feedback, you've probably got apathy and we 100% want to avoid that.
Another thing that I really liked is that everybody got a new title today.
You can add forward, forward deployed engineer.
into your title right now, if you're in product and you go and actually talk to your customers, which we hope you all do, right?
You are now, you can add forward deployed engineering to your title and your skillset.
And then lastly, you said that the job really hasn't changed, but the tools have.
And I think that maybe we get lost in the tools too much.
And I think that there's so much that we can do now.
It's actually like, was that an effective use of your time?
Yeah.
Is the prototype or the little app that you just built to do something that maybe only you're going to use and might not ever see the light of day.
Is that, was that actually, was that actually a valuable use of your time?
So, oh, and then the early on you said, don't vape, don't vape and drive.
So that's, don't forget.
Don't vape and drive.
Yeah.
So what's next for you and what's next for like where you see this, where you see the, designing products for like the designing AI products for the physical world?
Like what's next in that, in that realm?
Yeah.
You know, I think there's kind of two ways to answer it.
One from a craft perspective, I think, you know, we are all, we've all been kind of climbing the.
this, this mountain in terms of like learning how to use the tools and all that stuff.
And I think you're exactly right.
Like we're starting to get to a place where maybe I don't need all those agents.
Maybe I need to like, kind of like trim down and focus on the ones that are really adding value and that sort of thing.
And I, and I come back to like, you know, the job hasn't changed.
You do need to just kind of be a connoisseur of all the things that are effed up about.
your customer's experience and go and fix them.
That's like the, that is the job and, and, and building the right thing in the right way at the right time has always been the product person's job.
And it is now more than ever.
I think like, you know, and so, so, so I think that again, the craft is, it's sort of like, I felt like I would get like.
maybe like 20 at bats to do to like make an, like an important product decision every month as a VP.
And now it's like, I can do that.
Like I do that like 10 times a day.
It's like, because the rate of the kind of like actual development is, is, is so much faster and the prototyping is so much faster.
And so not that the job has changed, but just like.
it's just faster.
Every, every step of the cycle is faster.
The, you can listen to gong, all the gong calls in 10 minutes.
You can, you know, you can summarize all the Slack messages in 10 minutes.
You can figure out all the fires in 10 minutes.
Like you can do all these things.
And so everything gets faster.
So on the craft side, like again, not a lot has changed.
The tools are just better and they're, everything is going faster on the product side.
You know, I think there's a lot that we are kind of discovering on the agentic front.
And we are going to see these kind of massive digital transformations of physical operations companies.
80% of the world's workforce are frontline workers.
You know, a very small sliver of VC money ever goes to products that focus on them.
That's changing now as like people are starting to wake up and understand that AI touching the physical world is actually probably the biggest opportunity that kind of that sits in front of us.
But I think the like fundamentals of focusing on how do we make businesses safer and more efficient are not going to change for Samsara.
The like.
products that we launch, the iterations that we make, the, you know, whether it's like helping customers save on fuel or, you know, wearable accessories or asset trackers that make sure that every package gets delivered to the right place at the right time.
Like all of these things are really framed around three things, make businesses safer.
make them more efficient and digitize these businesses that have been kind of traditionally running in, you know, very old school systems for a very long time.
So, you know, it's just the mission stays the same and the tools get better and the craft stays the same.
Last one, yeah, as a former baseball player, you get more bats.
Is the batting average going up with these changes?
I actually think, you know, I think the inputs, it's interesting because I think that like the bats are actually becoming You have to hit more singles.
Like I think you basically, because every one of these decisions comes with so much context now.
And it's kind of like, if you are a curious PM, you can go and get the answer.
Like you don't need, you don't need the kind of to wait on the cycle for the data person to pull the thing, or you don't need to like, you know, have the meeting with the customer to talk to and get the customer feedback on this specific question.
You just.
you have a corpus of like data already.
And so when we are making decisions, we kind of can ask questions in the room.
We do like product pit stops, which are basically like really informal product reviews.
And we do them all the time.
And during a pit stop, like a PM, I'll just ask a question.
A PM will say, okay, let me just get the answer.
And like, you just kind of get that context in real time.
You're like, okay, well, let's just make the decision because this one's lower stakes and we have all this context.
So like, I think we can just like do this like much faster than we were.
And rather than like you go and pull the thing, we'll meet next week.
And then, man, this is a high stakes decision because we don't know.
The risk is like unclear.
We don't have all this context in this space.
We kind of can narrow all of those things now and just like make really clear decisions.
So that's been really fun.
And it has been, yeah, it's been really rewarding.
I love it.
Yeah.
He had a bunch of singles.
That's a way to score a lot of runs.
I think that was a home run.
Arvon, this has been tremendous.
It's been really great hanging out with you.
Thanks for having me guys.
This has been fun.