Pod Insights SaaS Interviews with CEOs, Startups, Founders Gather AI: $15M Revenue With Drones and 170% Net Retention - Sankalp Arora About this episode How do you build a $15M revenue business by teaching autonomous drones to find missing sneakers in warehouses, and get customers to more than double their spend every single year?
Sankalp Arora is the founder and CEO of Gather AI, a physical AI company that uses autonomous drones and computer vision to track inventory in real time across warehouses. He holds a PhD in robotics from Carnegie Mellon, built the world's first safe autonomous helicopter for DARPA, and today has 30 to 40 customers paying an average of $500K per year with net revenue retention of 170%.
You'll learn:
Why Sankalp spent three years writing code with zero revenue before getting his first customer in 2021 How a $2 warehouse pick turns into a $15 pick because of one inventory location error, and how Gather cut that by 70% for Barrett The land and expand math: customers start with 5 facilities and expand to 100 within three years at $500K per facility network Why 170% net revenue retention is possible when your product is embedded in daily warehouse operations How Gather AI built neural nets that can read barcodes a standard grocery store scanner cannot, with no human annotation required Why Sankalp passed on big-name VCs and chose Smith Point Capital, founded by Keith Block, former CEO of Salesforce The forklift vision product that turns any warehouse into an Amazon Go store using off-the-shelf hardware from Best Buy How Sankalp was carrying suicidal thoughts since his teenage years and only discovered life without them a year and a half ago, while building a company past $9M in revenue The Series B math: $40M raised at roughly $270M to $400M valuation selling 10 to 15% of the business at $15M in revenue Why Gather AI is touching just 0.1% of its addressable market today, and what 1% would actually mean in dollar terms Connect:
YouTube: Nathan Latka https://www.youtube.com/@NathanLatkawatch
Gather.ai: gather.ai https://www.gather.ai/
Founderpath: founderpath.com https://founderpath.com/
Transcript 0:00 Speaker 00 Where are you sort of sitting in this space?
0:01 Speaker 01 I think physical AI is at the same point where ChatGPT was in around 2015, where it's just finding its roots.
0:10 Speaker 00 What's the average customer maybe paying per year?
0:12 Speaker 01 Our average deal size is about half a million dollars per year.
0:15 Speaker 00 But what does your largest customer pay today on an annual basis, or range is fine?
0:19 Speaker 01 I would say about three to four million dollars a year.
0:22 Speaker 00 How many individual customers are you working with today?
0:25 Speaker 01 About thirty to forty logos, independent customers.
0:29 Speaker 00 Hey folks, my guest today is Sun Kupora. He is building Gather.ai. It's a Pittsburgh-based physical AI company that uses autonomous drones and computer vision to deliver real-time inventory intelligence for warehouses, serving a variety of customers like Geodes, Axon, and others. He holds a PhD in robotics from Carnegie Mellon University, where he built one of the world's first safe autonomous helicopters for DARPA and has raised 74 million bucks to date. We'll talk about a Series B, which was led by Keith Box Smith Point Capital. Sun Kup, you ready to take it to the top?
0:58 Speaker 01 Yes, sounds good, Nathan. Thank you for having me.
1:01 Speaker 00 All right. Tell us, for folks that are not familiar with sort of the work Nvidia is doing on SDKs, foundation models to program the physical world, Jeff Bezos is reportedly raising a fund to invest in this space as well. Obviously, he's got a lot of warehouses to manage. Where are you sort of sitting in this space?
1:17 Speaker 01 Yeah, so I think physical AI, as it's being defined today, is robots operating in physical spaces, of course. And historically, the approach to that has been to design specific autonomy stacks for specific use cases. And what the latest and the greatest of industries trying to do is build a foundational model, much like ChatGPT, that can do anything for text purposes, can do anything for physical spaces. And I think if you get an equivalence, I think physical AI is at the same point where ChatGPT was in around 2015, where it's just finding its roots.
1:59 Speaker 01 But people now have belief that it can deliver that that wasn't possible before. And we somehow sit in the middle, where we have a mixture of foundational models and classical AI techniques, which are probabilistic robotics-driven, to actually make these robots safely work in physical live environments and provide value today, instead of providing value seven years down the line.
2:25 Speaker 00 And in your opinion today, if you looked at every engineer in the world, every person building something, who's the best sort of person to build this? I mean, I think deeply about what must be happening in Ukraine with these folks on the front line, programming physical spaces, commanding to the drones, you know, having them, you know, hit and protect certain cities, things like that. I also think about, obviously, someone that might have built the, you know, inventory, you know, robotic system to manage Amazon warehouses to get delivery time down by a millisecond because then we spend more. I think about someone like you, work with DARPA, right? You know, great university going to. I mean, why is your background and your team sort of best suited to win this space?
3:03 Speaker 01 Two things. The three co-founders, Daniel Gitash and myself, like you mentioned, worked on the world's first fully autonomous helicopter, won national awards for the work we did with aerial autonomy. So we had deep expertise in physical AI for aerial autonomy. And my thesis focused on how to make robots curious. And they're curious here in warehouses about boxes, barcodes, inventory and workflows. So that that makes the technology alignment quite strong. And the second thing is, over the last six or seven years, how our stack has evolved. Now our stack is the only stack in the world that you can put on a moving
3:47 Speaker 01 camera that you buy out of Best Buy and turns it into an autonomous data gatherer. So that's the technical side of it, where we have our data mode and a tech mode of given our technical backgrounds. But also, over the years, we have assembled a bench of people from Amazon, from Uber, from C-Grade with deep logistics background. So our product has evolved to support logistics players in how they want to be supported. So I think those two combinations give us an unfair advantage to to serve effectively in this space.
4:24 Speaker 00 And so with that background, again, I've got your website now pulled up here. We'll talk about the 4 million Series B here in a second, but you use sort of the terminology physical AI for intra logistics. Is your plan here to sort of be Switzerland and open source foundation model that anyone can use? Or is it really to try to, you know, walled garden, sort of Apple App Store approach and keep it all internal?
4:46 Speaker 01 Neither. I have a very different view on it, if you allow me.
4:49 Speaker 00 Please.
4:49 Speaker 01 I think our view is squarely focused on how do we deliver value to the end customer? Which means we can have really good foundational models, but the customer could not care less. They really care about how does it hit their top line and bottom line performance, which is either how does it impact their labor? Or how does it impact their shipping rates? Or in at the rate at which they are shipping out goods, facility throughputs. So we provide the value to the customer as an end result, which is mostly actually accessed by them through our web dashboards and insights, not through our physical AI. Our physical AI is a mode to get them that that value.
5:30 Speaker 01 So we really think of ourselves as the solution that generates the end value and not just provides the empowering technology to do so.
5:39 Speaker 00 So you're, there's a trend in AI right now, which is charge against the job to be done, not necessarily for credit usage or seat based approach. Am I hearing you correctly? You're charging Barrett Distribution Centers on some job to be done. There's not a massive engineering team there calling, you know, you know, tool calling your different foundation models to do work. You're, you're doing the work for them.
6:00 Speaker 01 That's right. That's right.
6:01 Speaker 00 Interesting. Make this hit home. Can you use this real example since it's on your website? So, you know, they have to store 400, 500,000 serious baggage shoe box. Is it so, okay, so they're a shoe company? Tell us the work you're doing for Barrett.
6:13 Speaker 01 Okay. So Barrett, we are actually in multiple kinds of facilities. They are one of the leading third party logistics players in US. But for this specific example, do you know Stadia Goods? They sell, it's a, it's a marketplace for vintage shoes.
6:26 Speaker 00 Spell it stadium?
6:27 Speaker 01 Stadia. S T A D I A.
6:30 Speaker 00 Okay. Go ahead, I'm pulling it up.
6:32 Speaker 01 Yep. So you get all these collective. Collectible sneakers that range anywhere from a hundred dollars to fifteen thousand dollars. And that shoe facility that you were showing images of holds about half a million pairs of these shoes. And when customers pay, you know, even in this case, when customers pay, let's say one thousand dollars for their shoes, they want it next day. But when a person goes there to pick this pair of shoes, which is a case, and they don't find it there, they basically have to pull a software and on and say, oh, I can't find the shoes there. So let's spread out and find that pair of shoes that needs to go out. And suddenly a pick that was supposed to cost two dollars ends up costing ten to fifteen dollars.
7:15 Speaker 01 What we came in and were able to do in partnership with Barrett was reduce that those kind of location errors by seventy percent.
7:23 Speaker 00 And that's because of this drone we see here in the graph? This is you basically going and finding the SKU that was purchased on new sneakers to get to the delivery time faster?
7:31 Speaker 01 Yes, but it's even better than that, that those drones are continuously flying, taking images. So you not only need to find that targeted shoe, you actually know at any given point, how your warehouse is laid out and how things are moving. So you can make decisions in real time on how to do better picking, how to do better sorting. And that seventy percent decrease in inventory misplacement errors for someone like Barrett was the difference between keeping the Stadia Goods account and making them happy versus having them move to someone else because they came to Barrett because they were unhappy with their last three PL and how their inventory was maintained.
8:11 Speaker 00 Guys, remember, I am not just a YouTuber, I'm investing in my third fund. We've deployed two hundred and fifty million dollars into five hundred and fifty software companies so far. Again, at FounderPath.com, if you're interested in capital, I would love to cut you a check because I know you're investing in your education, you watch my show. So sign up at FounderPath.com and when you get the onboarding email, I reply and I see all those. Just reply and say, Nathan, I found you through YouTube and I'll make sure to prioritize you. I would love to cut you a check. Check out FounderPath.com. This makes a lot of sense to me. Thank you for educating me on how this works. How do you price? This thing, what's the average customer maybe paying per year? Is this sort of enterprise motion with million dollar per year accounts? Or is it more sort of bottoms up? I imagine it's probably more enterprise, but please you tell me.
8:54 Speaker 01 Actually, it's a bit of both. When we, so just one thing I'm really proud of, I'm proud of the, of the thing that we've built, our team, is our net revenue retention annualized is one hundred seventy percent annualized. So it is best in class. So we start with, you know, five to seven facilities in a network, which is usually our average deal size is about half a million dollars per year in terms of subscription.
9:21 Speaker 00 For those seven facilities?
9:22 Speaker 01 Yeah, five to six facilities, I would say, yes. And then that ends up by second year going to fifty odd facilities. And by third year, we are looking at hundred plus facilities and taking that forward. So it starts with a little enterprise deal, but expands into network wide fairly quickly and fairly often.
9:40 Speaker 00 This is a traditional land and expand by, but coming from the guy that built the world's first autonomous helicopter, which we love. This is your engineering sort of sales kicking in here. What is, don't name the customer, that wouldn't be appropriate, obviously. But what is your largest customer pay today on an annual basis? Or range is fine.
9:57 Speaker 01 I would say about three to four million dollars an year.
10:01 Speaker 00 This is so impressive. How many, I'm just not familiar with how many warehouses exist in the world. You have one hundred and seventy percent net dollar retention. How many customers are in the world that you think you could get, you know, on Gather AI and deliver that three to four million dollars of sort of contract value annually?
10:17 Speaker 01 Oh, so this is very interesting. So looking at warehouses in US, let's not even look at the world. If you look for warehouses that are above hundred thousand square feet.
10:30 Speaker 00 Okay.
10:30 Speaker 01 That's about hundred and fifty thousand warehouses.
10:33 Speaker 00 Wow.
10:34 Speaker 01 And for us, for example, to serve a decent chunk of that business so that we are generating a hundred million dollars worth of revenue on our side. Because that usually means that we are generating a value of about three to four hundred million dollars, sorry. Hundred, hundred million. Dollars of ARR on our side, that is usually means we are generating three to four hundred million dollars of revenue for our customers. Is just serving about hundred customers with ten warehouses each. So that's just point one percent of the market that's available. So this, we have an opportunity here to really serve a massive number of customers to alleviate their day to day, day to day pain points.
11:18 Speaker 00 And how many, you just gave the point one percent, Nar, but how many individual customers are you working with today? Like, like a Barrett, are we talking like ten or a thousand or a hundred or?
11:27 Speaker 01 Oh, yeah, we are, we are, I like to say we are just, just starting out. This is still very much day one. So I would say about thirty to forty logos, independent customers.
11:36 Speaker 00 Okay. And I don't want to put you on the spot, but I mean, can I multiply the thirty logos times a half million bucks a year? That would put you at about a ten or fifteen million run rate today.
11:44 Speaker 01 Let's say roughly right.
11:46 Speaker 00 Okay, great. And can you share what growth looks like if you're at fifteen million ish today? Where were you one year ago? Obviously, you're growing because net revenue retention is one hundred and seventy percent, but that doesn't even include new logos.
11:55 Speaker 01 Yeah, it doesn't, yeah. So we grew two point five X last year, year over year. And this year we are forecasting anywhere between two to three X, again.
12:06 Speaker 00 Wild. I want to get your, more your background story. We know about the great work you did at DARPA on autonomous helicopter. But you list many industries on your website, three P L Logistics, Retail Manufacturing, Life Sciences, and four separate solutions, A I Copilot, M H A Vision, A I Vision, and sort of how it works solution. When did you write the first line of code for this platform and what was your first target industry and solution?
12:28 Speaker 01 Our first target solution was helping warehouses get their inventory accuracy much higher, which is our drones flying around in the warehouse taking inventory. The first line of code was for the solution was written right after my PhD defense, so, late twenty eighteen. And it took us three years from then to get to the market, to get the tech ready, because no one had ever built this tech around things that you can buy out of Best Buy. And we really wanted that for scalability, for serving our customers with effective hardware, which is robust, but at the same time, for our unit economics to look like SaaS instead of RAS. So it took us three years to get the tech stack out there where it was performing reliably for our customers.
13:12 Speaker 01 And since then, we've grown dramatically year over year.
13:15 Speaker 00 So 2018 was first line of code. Did I hear you correctly? 2021 is first customer?
13:21 Speaker 01 Yes, that's right.
13:22 Speaker 00 How did you fund the business over those three years?
13:24 Speaker 01 We have, we was, since we were pitching a tech that was not being done in the world, so we knew we needed sufficient capital.
13:32 Speaker 00 And then you just recently announced, to Allah you, I don't want to spoil the announcement, talk to us about the Series B. And we don't love celebrating funding sort of on the show, but obviously you need it to build a revolutionary technology. So talk about how much you raised and then what you plan to use the money for to help your customers like Barrett.
13:49 Speaker 01 So we, as I said, we grew 2.5x year over year last year with those retention numbers that we are really thankful for. But at the same time, the sales motion had scaled beyond founder led sales, beyond me, thankfully so. Because our new sellers were doing much better than I was. And as a result, we had sufficient proof points to say that now we need to pour more gas into this GTM engine and scale, which is where a $40 million Series B round came about with Smith Point Capital leading and taking most of that round. That is of, I think, what is founded by Keith Block, ex CEO of Salesforce and Chris Little.
14:31 Speaker 01 And they, they just saw how enterprise physical operations are going to change. Even if they won't be fully autonomous, they will be fully digitized. And that's what we do, digitize those physical operations. And now it is going towards not just scaling our drone based inventory monitoring product, but also scaling our forklift vision product, which puts cameras on forklift, the same tech stack on forklifts. And suddenly, whatever your forklifts are carrying and wherever they are placing it is digitized. So your warehouse turns into an Amazon Go store without a wide network of cameras that Amazon Go stores needed. So instead of things, we think of it like drones were finding things.
15:15 Speaker 00 Hold on, can you show me a picture? I'm sunk up. I'm trying to find, as you're getting, I'm trying to find a picture. Do you have the forklift anywhere on the site?
15:21 Speaker 01 Let's see. I'll have to check. Give me a second.
15:24 Speaker 00 Okay.
15:24 Speaker 01 I think if you go to. Solutions and go to MHE Vision.
15:27 Speaker 00 MHE Vision. Okay. Oh, here we go. Forklift number two. This is you?
15:31 Speaker 01 Yes.
15:31 Speaker 00 Okay, so you're the software and the hardware.
15:34 Speaker 01 Yes, so the operators get live feedback on whether they're doing the right transaction, whether they're picking up the right thing, whether they're putting away the right thing from the truck to put away, to replan back onto the truck. So as a result, your whole operations becomes digitized. And then we can start providing insights on how to run operations better and not just find your missing things in the warehouse. So we have gone from-
15:58 Speaker 00 But you've made the decision not to be heavy, sorry to cut you off, but you've made the decision not to be heavy OpEx. You're not selling all the, the yellow forklift here. You're, you're basically saying, look, we're going to build software that's so flexible, you can plug it into any off the shelf hardware. So you are hardware Switzerland.
16:12 Speaker 01 That's it. Yes, that's true.
16:13 Speaker 00 Smart, really, really smart. Okay, very, very interesting. So, Nicole, do you have a data advantage here? In other words, the more IOT devices installed and whether it's a drone or a forklift that you have installed, you're collecting, I imagine, millions of data points every single month at this point. How have you structured sort of your database to normalize the data, run an ETL process on it and help the next Barrett that signs up have a lower sort of loss rate or inventory loss rate? Like, is there a compounding data machine underneath here?
16:39 Speaker 01 Yes, I'm so glad you asked that question. I think that's what keeps us in the lead in terms of the data insights we offer to the customers. Because we, and we have proven examples, I would, I would request Amanda to share them with you. Now we can, our neural nets can read barcodes that not even a barcode scanner, a line scanner that you have on your grocery store can read. And mostly because the millions of data points, the millions of barcodes and boxes that we are seeing, a bunch of our neural nets are running self-supervised experiments on them. So there is no human in the loop. They improve themselves as they run these experiments. And then a small number that they have trouble with in self-supervised fashion get to people annotating those examples and then feeding it into the trading loops.
17:19 Speaker 00 So I appreciate you coming on today. Is there anything I missed that you want to make sure we talk about here over the last 60 seconds?
17:24 Speaker 01 Yes, there's something that I, it's a bit of a personal message because you said a lot of the audience is existing founders and some VCs. I would say the biggest struggle that a founder- And this is every founder that I've ever talked to is the loneliness and the mental health cost that comes with dealing with uncertainty. So the best thing I would encourage the fellow co-founders is across across the world is to be vulnerable with that side and seek help because I think to me as well that something I discovered has changed the quality at which and the level at which I was performing when I was able to address and seek help in those domains actively.
18:05 Speaker 00 Is there a personal story you're comfortable referencing? Did you deal with a battle of depression?
18:09 Speaker 01 Oh, yeah, sure. Yeah, no, I'm super, super comfortable sharing. Actually, up until one and a half years ago, I didn't know that life, that people were existing that don't have suicidal thoughts. Like, that was my life up until then, and I thought everyone deals with it. So so am I because that's how I.
18:27 Speaker 00 You were having you were having suicidal thoughts as you're passing nine million of revenue. Everyone would say, oh, my gosh, this is amazing. At least if they're just reading your Twitter bio. But you were dealing with suicidal thoughts at the time.
18:37 Speaker 01 Yes, yes, yes. And that was being that was my sort of I was carrying that since my teenage years, but they were getting really louder and mostly because of the sense of responsibility and pressure. And I discovered the right mixture of medicines and discovered that there is life without suicidal thoughts. And I think that just made me a much happier operator and a much happier builder. And that shows in our work and what we're doing.
19:03 Speaker 00 Well, I can't wait to see the impact that you and your team obviously are going to have on the world. To your point, you're like touching point oh oh one percent of your market right now. My only regret is I didn't find you earlier and demand that you let me put in a hundred thousand dollar check because I think what you're building is really compelling intellectually. It's interesting and the growth speaks for itself. So some call up if people want to follow your journey going forward after this interview, where can they find you online?
19:27 Speaker 01 The best place is actually LinkedIn or reach us through our website.
19:30 Speaker 00 Just at a series B of forty million with Smith Point Capital, selling ten to fifteen percent of the business, that would be about a two hundred and seventy to four hundred million valuation, somewhere in that range with seventy five people. But more importantly, he's really touching one of the first founders I'm seeing really touch on foundation models for the physical space, which is the next. Big frontier. He's touching a very small percentage of his potential customer base right now. There's over 150,000 warehouses in the US with more than 100,000 square feet. He's touching just 30 customers today, largest ACV, three to four million as he continues to expand his solutions like MHE vision into multiple industries like retail manufacturing and more 3PL companies like Barrett. Sung Koo, thank you for taking us to the top.
20:08 Speaker 01 Thank you for having me on and letting me share.
20:10 Speaker 00 You won't believe this CEO's revenue. Click here to watch the next episode right now.
Previous Tive Hit $100M Revenue After a Down Round and $10K in the Bank Next He Makes $50m/yr on Cookie Popups, Controls 2% of all Web Traffic Page updated November 6, 2026 · transcript and summary are machine-generated
