E105 [AI-Translated] Aepsy x Kinastic | Physical AI $47B | a16z DOJ | ExpertVision AI | Gravis Robotics

00:00:00: The original podcast was recorded in German.

00:00:03: This podcast And welcome to Burn Rate, the VC Insider podcast.

00:00:16: I'm sitting here with Guy Gifredi talking about the startup scene with a focus on venture capital.

00:00:21: episode one hundred five is coming up and we're recording on Friday August twenty first at eight thirty in the morning.

00:00:28: thank you very much for the positive feedback on our new format.

00:00:32: We've received quite a bit of feedback since um last week were very happy about that.

00:00:40: This podcast is sponsored by our partners Omnium and Upscaler.ch.

00:00:44: Guy, good morning!

00:00:46: What are today's topics?

00:00:47: Today

00:00:47: we have some really exciting topics on the table.

00:00:51: in the news We're discussing The Acquisition of Kinastic By APSI Two Swiss Startups.

00:00:56: Then we Have The Rise Of Physical AI Investments In The VC Space On The Agenda And... The Third Piece Of News Is From America.

00:01:04: Andrejsen Horowitz Is Being Investigated Because They Sit On The Boards Of Competing Companies.

00:01:10: Then we have Alejandro and Mate from Expert Vision as guests.

00:01:14: They currently have a campaign running on Omnium, our partner.

00:01:18: And finally We will discuss about the transaction of The Week.

00:01:22: This time we have Gravis Robotics Their two hundred million mega series A with Softbank in the lead On the agenda.

00:01:28: Good,

00:01:28: let's get to the news of the week.

00:01:30: We'll start with an exciting piece of news from The Swiss Digital Health Market.

00:01:34: Epsi is acquiring Kinastic.

00:01:36: You already mentioned it guy?

00:01:38: We know Epsia has a platform for psychological support With more than four hundred psychologists.

00:01:44: That's impressive!

00:01:45: Kinastic meanwhile comes From the corporate health promotion space And serves around one-hundred companies with offerings Around exercise nutrition mental fitness workshops campaigns and so on & So forth.

00:01:57: Through the acquisition, EPC wants to bring psychological support and prevention together under one roof.

00:02:03: Going forward, The combined offering will reach more than a hundred fifty partner companies in Switzerland.

00:02:09: Kinastic's team and platform will be integrated into EPC over the next few months.

00:02:14: That sounds strategically logical guy.

00:02:16: AFC has psychological care.

00:02:18: Kinastic has prevention But is this actually an outstanding M&A case for you now?

00:02:24: Or simply two relatively small Swiss providers that are getting a little more scaled together.

00:02:30: Maybe just a quick comment on that before your allowed to answer.

00:02:33: AXA invested in Kinastic, um... In twenty-twenty-two and works with Epsi.

00:02:39: I assume that AXa naturally played relevant role in the combination.

00:02:43: We know both companies very well And have also accompanied and observed them over years And I do think it makes quite a lot of sense that APSI has now acquired Karnastic.

00:02:54: It's pretty logical, especially in B-to-B sales.

00:02:58: as an HR department you don't want to somehow buy five or more different tools one for mental health then one for exercise and one for nutrition.

00:03:10: If APC really manages to turn this into one product with one contract and approaches companies as a single point of contact, then the value for customers increases quite significantly.

00:03:24: And the most exciting part of The Acquisition is probably not necessarily Kinastics technology but also the roughly one hundred existing corporate customers and sales relationships.

00:03:37: For me This is a very good example of the fact that with these kinds of SaaS platform models at some point, the question comes up.

00:03:46: Hey!

00:03:47: Do I build the next product area myself now?

00:03:49: Or do i simply acquire the customer's team and product at the same

00:03:54: time?".

00:03:55: That's exactly what I was getting at.

00:03:57: why does Epsy acquire Kynastic et al instead of building this prevention offerings itself?

00:04:04: I mean workshops content or health-related challenges don't exactly seem like rocket science technologically.

00:04:11: I can't assess now how much they have invested in the technology, but very likely the technology really isn't.

00:04:36: Kinastic was already founded in twenty sixteen, and as mentioned has around one hundred corporate customers.

00:04:43: And building that base yourself can take years.

00:04:46: it did for Kinastic as well.

00:04:48: If Epsi now takes over these customers and can simultaneously roll out its existing mental health offering to them then of course thats significantly faster than organic growth.

00:05:03: Interesting metric for me would therefore be the purchase price, it will certainly interesting to find out what that looks like in relation to acquired recurring ARR.

00:05:15: Without that number, we obviously can't assess whether it was financially a good deal.

00:05:21: But strategically, it does make quite a lot of sense.

00:05:23: Max do you think will see deals like this more frequently now?

00:05:26: In other words consolidation among Swiss digital health and employee benefit startups

00:05:32: probably not only in this area but generally.

00:05:35: That wouldn't surprise me.

00:05:36: I mean many of the companies started out very specialized in recent years, mental health fitness nutrition.

00:05:43: there were quite a few benefits providers or prevention.

00:05:46: And for the employer, this fragmentation is actually unattractive.

00:05:50: they want as few providers as possible and one solution that covers a lot right?

00:05:56: At the same time fundraising has become difficult on the startup side.

00:06:01: That makes M&A more attractive for smaller providers instead of working alone again to achieve scale.

00:06:08: A merger with a peer can make quite a lot sense.

00:06:12: That's why I also find this deal interesting, and perhaps EBSI Kinastic is less of a huge acquisition in-and-of itself but possibly very good example how the Swiss digital health market will consolidate over next few years.

00:06:26: Let us move on to second piece news.

00:06:28: one number that really stands out this week.

00:06:36: physical AI startups in the first half of twenty-twenty six alone.

00:06:40: That's across more than five hundred twenty one deals, that almost four times as much as in the second half of Twenty-Twenty Five and around eighty percent more then in the First Half of Twenty Twenty Five.

00:06:52: And even crazier!

00:06:53: in the entire three years from twenty-twenty two to twenty twenty four combined, it was only just under forty two billion dollars.

00:07:02: So importantly by physical AI we mean among other things robotics autonomous vehicles aerospace drones industrial automation or sensors.

00:07:12: so all topics where we also invest at coio capital our fund which we run together.

00:07:19: guy first question for you Are we really at the next big wave after generative AI here?

00:07:26: In other words, is VC money now moving from software into the physical world.

00:07:31: That's the question.

00:07:32: I definitely think that trend is absolutely real.

00:07:36: You can see it in numbers and you also see if you study them a little that many are betting that AI will enter the physical.

00:07:46: But as always, you do have to be a little careful with the numbers.

00:07:49: I mean... A huge part of the capital comes from just a few mega rounds.

00:07:54: Waymo alone raised sixteen billion US dollars in February and then there are five billion for Andrew Peter Teal's start-up Then two billion for SHIELD AI And somehow just under two billion.

00:08:06: and Waymo alone accounts for almost a third of the entire physical AI funding volume.

00:08:29: The thesis behind it is actually clear.

00:08:37: Software and AI infrastructure are now good enough that you can bring intelligence into real machines, processes much more

00:08:43: easily.".

00:08:45: But thats exactly where I'm a little critical too right?

00:08:51: You should never invest in hardware startups, high capital requirements long development cycles complex supply chains.

00:08:58: So why is that suddenly attractive again now guy?

00:09:00: I remember that period very well too.

00:09:02: we've always looked very closely at hardware topics across all funds and it was difficult to find co-investors because everyone rejected them immediately.

00:09:12: But it does seem that the economics have improved significantly.

00:09:16: now, I mean the investors quoted by Crunchbase argue precisely that compute and foundational models have become more accessible simulation works better and sensors as well as hardware has become much cheaper And As a result of course smaller teams can now develop much faster and bring products to customers earlier with less capex required at the start.

00:09:40: And you also have to say that the business model has changed somewhat.

00:09:44: Hardware is no longer simply sold as a single sale, where you have to sell another robot every year.

00:09:50: but hardware is now increasingly combined with... As we know it from software usage based pricing or outcome-based pricing.

00:09:58: That means you don't just sell once But can build an ongoing software and data layer on top of it and generate ARR for AVC!

00:10:06: That suddenly becomes more interesting because In the best case, you combine the defensibility of hardware with the economics of software.

00:10:13: That's obviously incredibly exciting and The decisive question remains can the company actually get into production?

00:10:19: And also reach paying customers?

00:10:21: according to investors that exactly where the focus is now shifting max if You look at this from the startup perspective Is physical AI Now an area that founders should jump in too.

00:10:31: the way everyone suddenly founded a genai start up two or three years ago.

00:10:35: I'd be very cautious there.

00:10:37: The opportunity is obviously huge, but physical AI is probably the exact opposite of an easy trend.

00:10:44: you should simply jump onto.

00:10:46: With a classic AI software startup, you can build and test the product relatively quickly.

00:10:52: With robotics, aerospace or industrial automation... ...you often need hardware expertise You need access to real customer environments You need production And great deal of domain knowledge.

00:11:05: I think that the strongest modes emerge in companies that control several layers themselves and are vertically integrated.

00:11:13: You hear that again, for me the winner probably won't simply build an AI wrapper but will control technology data and a relevant part of physical system.

00:11:25: That's why I would say Physical AI is one of next major VC categories But it'll be harder to build a genuinely good company there than the hundredth AI SaaS case.

00:11:38: Good, let's move on to the third piece of news and this is an exciting piece from the VC world.

00:11:45: The US Department of Justice is investigating Andries and Horowitz because partners at the fund simultaneously sit in boards that now compete with one another.

00:11:55: Specifically it concerns data bricks.

00:11:58: Ben Horowitz sits on the board of Databricks and a sixteen-z partner, Martin Casado.

00:12:06: What's interesting is this when Andreessen Horowitz originally invested according to TechCrunch The two companies were not yet direct competitors.

00:12:16: only later did Databrick expand its offering more strongly toward data pipelines And connectors in other words directly into five trans core market.

00:12:27: According to the report, the investigation has already been ongoing for almost a year and is based on Section VIII of The Clayton Act.

00:12:35: An antitrust law that's more than one hundred years old... ...and intended to prevent so-called interlocking directorates between competing companies.

00:12:45: Guy!

00:12:46: Is this really an Andresen Horowitz problem or it simply unavoidable when a large VC finances hundreds of companies over the years?

00:12:55: and their markets eventually start to overlap.

00:12:58: I mean, Andreessen Horowitz is known for investing in fast-growing companies then staying on the board a long time.

00:13:05: The fact that there are portfolio overlaps And they become competitors Is almost unavoidable.

00:13:13: Startups change products They pivot and expand into new market.

00:13:17: If you're a large fund Investing hundreds of companies At some point Some will simply be competitors.

00:13:25: But you also have to see that an investment is something very different from a board seat.

00:13:29: A normal investor who's simply on the cap table receives updates on the numbers and perhaps, on the current market.

00:13:36: but as a board member of course You get much deeper insight into the strategy The product roadmap pricing potential acquisitions future expansions And of course you actively influence all of that as well?

00:13:48: That's precisely why it becomes problematic when two partners from the same fund or the same firm sit on boards of competing companies.

00:13:56: Then at very least, a question arises how do you ensure that information doesn't reach one company and how they remain separated inside Andreessen Horowitz?

00:14:08: Exactly!

00:14:08: Because Andreessen could now argue these are two different companies' partners... ...and we build so-called Chinese walls between them.

00:14:19: Is that enough in your opinion, or would one of the two partners have to give up their board seat in a case like this?

00:14:25: If we put on The Governance Lens for a moment then I'd say very clearly if companies now truly become direct competitors.

00:14:33: Then One Board Member stepping down from one of these companies will certainly be cleaner solution.

00:14:37: A Chinese wall can certainly help internally, but it's still the same company or the same fund on both sides.

00:14:44: And these partners know each other.

00:14:46: they might go out for a beer or dinner from time to time and then those are all off-the-record discussions.

00:14:52: that where becomes quite

00:14:53: problematic.".

00:14:54: This is another interesting aspect we should perhaps briefly look at here as well.

00:14:59: What does this actually mean?

00:15:03: Top funds don't simply sell founders' capital.

00:15:06: Hey, we're investing.

00:15:08: Instead they say hey!

00:15:10: We are investing and joining the board helping you with strategy hiring fundraising and M&A process.

00:15:19: If a fund at some point has to say, hey we can no longer remain on your board because Especially in very concentrated markets, it's definitely relevant.

00:15:45: We're already seeing an AI that investor loyalty has become significantly weaker and many large funds today invest simultaneously In companies that are in some cases directly fighting each other And for founders That means you shouldn't only ask who gives me the best term sheet but also what Other companies does this investor have?

00:16:10: Where could conflicts arise in two or three years and what information does the investor receive from us?

00:16:16: Of course, you can also argue the other way around.

00:16:19: It can also be an advantage if an investor has very deep industry expertise.

00:16:24: that's something You can't forget either.

00:16:26: you always see it from the critical side now But you can Also argue differently.

00:16:30: That's clear.

00:16:31: but all-in-all I don't think large VCs will suddenly stop financing competing companies But board seats could certainly be allocated more consciously and perhaps the amount of strategic information founders themselves share with their investors will also change.

00:16:48: I do believe that, so you certainly have to be careful there!

00:16:52: The interesting thing about this case therefore isn't only Andreessen Horowitz.

00:16:57: If the Department of Justice actually takes action here, it could have consequences for the entire Silicon Valley model in which large funds finance more and more companies within the same markets.

00:17:10: We're curious to see how this will be decided...

00:17:13: And what would certainly also be interesting is seeing as in many cases where you had two such large competitors and suddenly end up dealing with antitrust questions.

00:17:25: Is this an isolated case now?

00:17:27: Or are there really five, ten or twenty funds that have a problem like

00:17:32: this?".

00:17:32: With Omnium you close your financing round faster and more successfully.

00:17:37: Our partner Omnum is an innovative platform for growth financing.

00:17:41: Omniam is the one-stop shop for financing rounds because they offer a completely digital lean and cost efficient process including the legal framework Investor KYC and Payment Processing.

00:17:53: Anyone raising on Omnium has the choice of doing a private round, using only platform for their investment opportunity or they choose Private Round Plus with additional access to ten thousand registered investors.

00:18:06: anyone who wants high-profile rounds with broadest possible reach chooses Public Round where offering is made accessible to all investors.

00:18:14: With omnium you close your financing round faster and more successfully.

00:18:19: Do you want to scale your startup quickly and successfully, not just grow?

00:18:23: Then the Upscaler program is exactly right for you.

00:18:26: It gives you access our large pool of founders, entrepreneurs & suitable investors.

00:18:31: More than fifty startups have already benefited from more then five hundred workshops And scaled their businesses nationally and internationally!

00:18:39: Do also wanna benefit From this personal and individual support.

00:18:43: You can find all information at www.upscaler.ch.

00:18:48: Today we're once again presenting a project from our partner Omnium.

00:18:52: On omnium investors have the opportunity to invest directly in selected Swiss companies and growth projects, And The Expert Vision AI Campaign is currently live there.

00:19:02: Expert vision AI is a swiss tech company that uses computer vision & artificial intelligence for theft prevention in retail.

00:19:10: Put simply expert vision turns the stores existing surveillance cameras into intelligent sensors.

00:19:17: The AI analyzes video streams in real time and can, for example detect when products are not scanned at self-checkout.

00:19:24: Merchandise is concealed large quantities of a product are taken from a shelf once or someone leaves the store without paying.

00:19:33: The goal isn't simply to document a theft after the fact but alert staff while there's still time to react!

00:19:44: ExpertVision states that it does not identify individuals and integrates with existing camera systems.

00:19:51: Today we're speaking with the two co-founders Alejandro Garcia, founder & CEO of Expert Vision AI With a strong background in computer vision, AI and technology And Mate Varga founder and COO who brings many years of experience from the retail environment, previously served among other roles as digital director at Coca-Cola.

00:20:17: Alejandro Maate welcome to The Burn Rate podcast!

00:20:21: Thank you for having us.

00:20:23: Let's start with a very beginning.

00:20:25: How did two of you meet?

00:20:27: What was your specific moment when you realized that retail theft could be addressed much more effectively with Compute Division and AI.

00:20:37: Yeah, I've

00:20:38: been working in computer vision for a bit more than seven years as you mentioned And i realized already a few years ago that the automation of the analysis of video streams would actually help industries.

00:20:52: overall This has been picking up!

00:20:55: I started working on my first project at Compute division.

00:21:01: One of our customers was Coca-Cola.

00:21:04: There, we had the mandate to actually bring first autonomous store in Switzerland for employees.

00:21:13: And it happened that Matte was a digital director at the time working on this field and so we worked together more than one year In order to bring these into operation.

00:21:25: A few years later Mate and I, we realized that there was in this journey something more urgent to address.

00:21:36: And it's a theft prevention for retailers.

00:21:39: In the year of twenty-four, We decided together to fund an expiration in order to help retailers prevent theft.

00:21:47: Very interesting!

00:21:48: Mate you know the retail industry very well.

00:21:52: How significant is the problem with theft actually?

00:21:56: And drinking for retailers today?

00:21:58: And why are traditional measures such as CCTV, security tax or security staff no longer enough?

00:22:05: Yeah.

00:22:05: That was actually one of the main reasons I felt that it's a huge problem and shrinkage is an inventory difference.

00:22:13: This is something what every retailer has this big supply chain operation all together and shrinkages acceptable.

00:22:20: It's part daily business.

00:22:22: The problem is that what we have seen, it has been growing and doubled in the last five years.

00:22:29: And if you know grocery retailers which I know quite well they are working on a very slim profit margin right?

00:22:37: Three to five percent of their revenue to shrinkage can be twenty per cent.

00:22:48: That's why they invest now more and more in prevention.

00:22:52: And we're asking what about the existing solutions, right?

00:22:57: Let's quickly have a look!

00:22:58: The existing solutions are very passive.

00:23:02: you can put everything behind the counter.

00:23:04: it is working at a small store or you can lock down every thing on a cabin.

00:23:09: It happens actually but that prevents sales.

00:23:13: So it's not gonna work, you need someone to open for.

00:23:16: You can put down everything in the locker or put a cable on it but its preventing again people shoppers who experience their product.

00:23:24: so they leave retail store.

00:23:26: and of course you can put cameras.

00:23:29: many retailers started putting cameras.

00:23:31: watching camera is also job and watch upto five-to eight cameras at same time.

00:23:37: even one hour later.

00:23:39: your tired there just economically not viable to have forty cameras, twenty-forty cameras constantly watching and having a real time impact.

00:23:50: It's more for reviewing the videos if you've spotted something that can be used.

00:23:56: So more or less.

00:23:56: existing solutions are very expensive to build maintain it.

00:24:01: imagine RFID tags.

00:24:04: what they put on every single product is huge workload.

00:24:08: Honestly, when do you get the alert?

00:24:10: The last moment that a person is stepping out of your store.

00:24:13: So if they don't even have time to react and need to maintain this sometimes take it away before you sell it.

00:24:20: so big work but not proactive or helping them.

00:24:26: That's where we said that's right for real-time detection and alerts to prevent ideally prevent or facilitate security teams.

00:24:37: Okay, perfect.

00:24:38: Very interesting Alejandro.

00:24:39: maybe you can go through an example with us?

00:24:42: With the supermarket that has already existing cameras what happens once export division is connected?

00:24:49: and how do you turn these normal cameras into intelligent systems?

00:24:53: yeah That's a very good point because as Mati mentioned before most of the retailers have already cameras in their stores.

00:25:01: so We are already working with that infrastructure, which is dedicated only to record video streams.

00:25:09: In the event something really big happens then it's when they react but this is typically too late.

00:25:14: So basically we use these infrastructures in order to analyze video streams for real time.

00:25:21: For that typically we place a small server on the store To reduce latency and deliver alerts at fastest available times.

00:25:31: Typically, with a small server you can handle about ten to fifteen cameras which is normally sufficient to cover the most problematic areas.

00:25:40: Then obviously this solution can scale depending on the needs of specific customers.

00:25:47: We process those video streams and then when something suspicious is detected for example someone taking lot of perfumes in very short amount time we deliver a short video and a little bit of additional metadata to the security personnel in order for them to take action.

00:26:07: Or store managers, depends who is responsible for this?

00:26:11: And you call it sort of like the virtual RFID tag or the goods.

00:26:15: how does that work?

00:26:16: I mean an RFID tags physical device and cover up with video.

00:26:21: Yeah so we have coin disterm as mentioned virtual RFI D. actually We have patent pending on these technology that we have developed, and basically allow us to deliver the same functionality that RFID tags today without their constraints.

00:26:41: And with all of its benefits.

00:26:43: so basically were able follow a product from the shelf

00:26:47: to

00:26:48: the cashier or to the self-checkup point and determine if this product eventually might be stolen.

00:26:55: And that requires different aspects.

00:26:57: So for example, identifying a person with different cameras in the store, identifying when a person is actually grabbing a product and going to an exit

00:27:07: etc.,

00:27:07: etc.. These all together are what we put into many of our applications.

00:27:13: Some other applications do not require this core functionality but many of them yes!

00:27:18: This is basically what Defensit asks us also from other vendors only focus on delivery alerts in a very specific area.

00:27:28: The other advantage of the technology is that it allows us to deliver the technologies into different places, like for example at the aisles and checkouts, cashiers or exits which I think are interesting for retailers because typically they want solutions that can cover theft in all their aspects.

00:27:47: With this technology two things are obviously critical accuracy and privacy.

00:27:51: so how do you minimize false positives, meaning situation where a completely innocent customer is flagged as suspicious while at the same time ensuring that the tech remains privacy compliant.

00:28:06: Yes!

00:28:06: That's very good point and something we put lot of attention especially because requirements from GDPR and street data protection law.

00:28:16: Here we have to bear in mind that yes, while we put a lot of emphasis and here.

00:28:22: We are also trying to differentiate in this front or their vendors unfortunately they deliver such an amount of false positive that many retailers cannot use them And we put the lot of emphases into bringing these number two very low rates typically below one percent depending on the application.

00:28:42: Then the other aspect is that these alerts are only alerts.

00:28:47: They always are monitored by a human, by person who's actually determined.

00:28:51: We're only bringing information to this person going to decide.

00:28:56: we bring evidence and then they can't decide if it was actual case or not.

00:29:02: How do you build your system?

00:29:03: You aim at having real-time alert all of the time so almost real-Time.

00:29:11: Nevertheless, we see that in many cases retailers either are not ready to handle real-time or they don't want to have a different process.

00:29:20: They're reviewing the alerts on later stages.

00:29:24: so it's no problem if they review and any time they prefer before acting with their customer especially Switzerland is very important.

00:29:35: perception of the brand the alerts and what.

00:29:41: we help them, they don't need to search for the proofs.

00:29:45: We send them on their phone or laptop so that they can review

00:29:49: it.".

00:29:50: Okay I'm very curious then!

00:29:52: The system detects a potential suspicious event right?

00:29:55: What happens next inside of store?

00:29:57: So who receives the alert?

00:29:59: What information do you get from there?

00:30:01: And how does an ideal process look like at this point

00:30:05: onwards?".

00:30:06: As I said, we are aiming to send the alert as early as possible.

00:30:12: You know?

00:30:12: We have to wait until something is happening... ...we need to reach a trigger event which is triggering the suspicious case.

00:30:20: It can be sometimes at the checkout because it's a check-out or walk out case and when that person seems to walk out

00:30:27: but

00:30:28: giving enough time for the retail operation to react on it.

00:30:33: To your question about what's happening next, there are different options.

00:30:37: There are kind of human options.

00:30:40: either they have a security Personal in place and bigger stores.

00:30:45: It's typical And those people would get the alerts They will read it or look at it and react on it In smaller setups like small franchise stores where the franchise owner Or this store manager is in charge then these people acting on it.

00:31:04: And we are going towards a more autonomous store directions in general, so we're also working on autonomous solutions where our event can trigger alert in the store like the beeping of the gates for the ASTECs.

00:31:22: We can do just earlier with alerts some kind of serials or any kinds of lights or even like an AI-generated interactive text in an autonomous store.

00:31:38: Very interesting and theft.

00:31:40: big topic, how does the customer actually measure the value of your technology?

00:31:48: How did you demonstrate that it's cheaper to use expert vision than just having things done in pilot projects maybe also with real clients already?

00:31:58: Yeah, also a very good question.

00:32:00: And the answer is typically when we work with a new retailer We offer a pilot so both sides have time to actually work together in order To bring this system into optimal performance.

00:32:14: Typically you do that at the beginning.

00:32:17: Once the installation has happened running In couple of weeks You'll be able to adjust or fine tune the solution for them.

00:32:28: also on their side.

00:32:30: that depends on the size of a retailer, but they typically assign different people in order to receive those alerts.

00:32:38: And at the end-of-day, the retailer what they care about is it helping me detect more cases than I was detecting before and this typically we do within weeks.

00:32:49: so... In matter of weeks are able already detected more cases that they were doing before, even with security personnel watching cameras.

00:32:59: So in one specific case here for example department store in a retail chain here in Switzerland In just two weeks They were able to detect three times the value of The cost off of their operational fee That we have on them and therefore For them, it's clear that the value is there.

00:33:25: And that's why they're already planning about a scale in this solution to many more stores.

00:33:31: Okay very interesting and how do you make money?

00:33:33: So what's business model of expert vision

00:33:36: right.

00:33:36: so The business model is very simple.

00:33:39: we its assaults business.

00:33:41: Right.

00:33:41: We have little technical component like hardware component which is managed by the retailers themselves.

00:33:48: They have their camera infrastructure in place.

00:33:52: So our hardware component is a server which can be managed by the retailer, and the rest are our SaaS solution.

00:34:00: They're paying yearly fee based on their needs.

00:34:03: we created modular system.

00:34:07: some retailers have self-checkouts then they take module for self checkout.

00:34:10: others don't need this so that you build up kind of shopping list way what they needed subscription fee per store, and that's it.

00:34:25: In terms of competition AI and computer vision are developing extremely quickly And Of course you're not the only company looking at this problem.

00:34:33: What is your technological or commercial mode?

00:34:36: Why can't a large retailer Or an established security provider simply build similar solution themselves?

00:34:44: We saw some intense, a few years ago like maybe four or five years ago from large retailers to try and start working in this area.

00:34:56: But what we have also seen that period is they stop doing those exercises.

00:35:02: Retailers are retailers but not computer vision experts And for them trying to build these technologies is typically very short life exercise.

00:35:14: Obviously, there are other companies that are working in this field.

00:35:18: As I mentioned before, one of the key differentiators is these aspects where our core technology has the capability to detect theft cases for external offenders and internal offenders at the aisles, checkouts, self-checkouts or exits while others only work on some aspect.

00:35:38: Anyone who would try be at a level we're today they would need at least two years.

00:35:46: And you know how fast computer vision is changing today, which is opportunity and risk.

00:35:51: Opportunity in the sense that there are many other applications that are opening up very quickly... The risks obviously means we have to keep innovating so as to ensure that we're on top of the range….

00:36:03: That's where we are now!

00:36:05: We are at same level than any competitor in the world.

00:36:10: Now looking at the two of you, both founders with a very different background.

00:36:15: Alejandro, computer vision and technology expert with extensive experience in retail operations.

00:36:22: So how do you divide on a daily basis?

00:36:25: The responsibilities between the two?

00:36:27: view where do feel that skill sets complement each other particularly well?

00:36:31: And we're not?

00:36:33: Yeah, so naturally I think it was quite clear that Alejandro will be focusing on the technology part.

00:36:40: That's what he has been working for a long time.

00:36:43: He is dealing with the technical team and our CTO.

00:36:47: they are working hands in hand to develop their product.

00:36:50: My focus is more about building up relationship between retailers The whole sales process And marketing.

00:37:01: it's quite naturally fit together.

00:37:03: Yeah, that brings us to your current funding round on Omium.

00:37:07: so why are you raising capital now and how exactly do you plan to use the funds?

00:37:12: And yeah what are two or three key milestones you want to achieve over the next eighteen-to twenty four months to bring export vision to the next level?

00:37:23: So um... You know in first year we took time to build up the technology and products to a level that would be ready for running in stores.

00:37:35: Last year we managed already to have first pilots, converting those pilot into paying customers.

00:37:42: That was very big milestone for us ensuring this product market fit.

00:37:49: It was super important.

00:37:50: and this year what we have seen is that, uh We started working with large retailers.

00:37:55: That are obviously very keen to get these problems solved as fast possible.

00:38:01: So...that's one of the reasons I mean why we're doing the fund raising.

00:38:06: so..we think it's right moment because now in that scaling phase Now the retailers asking okay can you install half our stores?

00:38:16: and we have to bring the product now into this enterprise level readiness, so that fans are going to mainly be used in order start these scaling phase.

00:38:28: So great!

00:38:28: Alejandro Mateo thank you both very much for joining us and giving such an interesting look at export vision AI your technology and plans of future.

00:38:39: For everyone who would like to learn more about Expert Vision AI, the investment campaign is currently live on Omnium.

00:38:45: There you can find more information about a company and funding round.

00:38:49: an opportunity to invest will of course include the link To that campaign in the show notes.

00:38:57: So let's get to the transaction of the week.

00:39:00: And from a Swiss perspective this really isn't exceptional financing.

00:39:03: around Gravis Robotics, an ETH spin-off from Zurich has completed a Series A round of USUSD two hundred million and the investor is none other than SoftBank as the sole investor.

00:39:16: According to the press release this values Gravis at around USUSUSD one billion post money making it new Swiss unicorn in the round.

00:39:23: For

00:39:24: a Swiss startup that's obviously enormous signal.

00:39:27: And its also not simply large software round but capital intensive robotics and deep tech case.

00:39:34: Gravis builds autonomous systems for heavy construction machinery, so for excavators or earth-moving machines as they're also called which are used on construction sites in quarries or large infrastructure projects.

00:39:47: And Gravis I find interesting that not simply building an autonomous excavator but rather a retrofit solution means existing machines can be retrofitted with some hardware and software.

00:39:57: That is the so-called Gravis Rack.

00:40:00: It's a system with sensors, computing power and autonomy software that is mounted onto the existing machine.

00:40:07: From the customer's perspective That's obviously extremely important because construction companies already own many very expensive machines in other words A large machine fleet With an enormous acquisition value.

00:40:19: And if autonomy only works fully for a few new Machines then The barrier to entry Is obviously extremely high.

00:40:26: But If you can retrofit existing machines from different manufacturers, then the market naturally becomes significantly more interesting and adoption much more realistic.

00:40:38: As already mentioned ultimately Gravis is about what's known as earth moving so it's about digging excavation grading truck loading or stockpile management as its also called.

00:40:51: That may sound less glamorous than, for example humanoid robots which are very trendy right now but economically it's extremely relevant.

00:41:00: Almost every infrastructure project begins with digging somewhere and that is exactly where many of the construction industry structural problems lie.

00:41:08: Productivity has been an issue for years.

00:41:11: skilled workers are scarce good machine operators are hard to find And construction sites are complex and often dangerous working environments.

00:41:21: That's what makes robotics on construction site so difficult, this variety?

00:41:26: A factory is actually relatively easy to control.

00:41:30: but a construction site isn't.

00:41:32: the ground changes depending upon weather.

00:41:34: that changes as well.

00:41:35: The machines move.

00:41:37: people simply walk around meaning the construction workers and the site owners.

00:41:41: there are rocks Then there are utility lines in the ground, There's a slope and suddenly there is mud next to dust.

00:41:49: So it's very diverse environment which robots have find their way around.

00:41:53: And machine actively changes its environment because it digs or moves material That an extremely demanding environment for good autonomy stack.

00:42:03: Exactly And Gravis was founded in twenty-twenty two as an ETH Zurich spin off from the environment of Marco Hutter's robotic systems lab with Hans Peter Fessler.

00:42:13: As PRP, He was CEO of ABB Switzerland and with CEO Ryan Luke Jones, and CTO Dominic Judd.

00:42:24: The team brings a great deal of robotics and autonomy know-how.

00:42:27: they combine that with the very clear industrial problem And from my perspective That's precisely the difference compared With many Robotics visions you hear again and Again Gravis isn't trying to build A general purpose robot but is starting with a very, very specific application where the economic benefit is immediately understandable.

00:42:46: Exactly you can directly calculate that benefit its higher productivity less dependence on scarce and by now also expensive machine operators And of course greater safety as well.

00:42:57: Technologically Gravis combines sensors computer vision LiDAR GNSS Machine control and AI models.

00:43:05: so the machine doesn't just have to see where it is but also understand what its currently doing.

00:43:10: Let's take an example, It has to understand how the ground reacts and What movement makes sense next?

00:43:17: Gravis works with simulators for this.

00:43:19: And another important point Is that gravis isn't going directly For full autonomy meaning without a machine operator behind.

00:43:27: With the gravis co pilot as they call it.

00:43:29: They've developed an intermediate stage in which and operator remains in the machine but is supported by three d guidance has a detection and digital assistance, that of course makes market entry little more realistic as well.

00:43:44: and max who has gravis been working with so far?

00:43:48: Gravis is already showcasing major construction machinery and industrial partners such as Holcomb, Taylor Woodrow HD Hyundai, Hitachi & Flannery.

00:43:59: Holcomb is particularly interesting because there it's not only construction sites that are relevant but also quarries material logistics and heavy industrial environments.

00:44:13: isn't only a construction tech startup, but also a provider of autonomous heavy industry applications.

00:44:20: They talk about deployments across several continents and machines that are already live in different

00:44:25: countries.".

00:44:26: Let's briefly go back to Gravis' financing history.

00:44:29: as always we've put together a quick cap table.

00:44:32: the company was founded in October twenty-twenty two.

00:44:36: In February twenty three they carried out pre-seed round, the valuation was already set at twenty million pre money here and wasn't communicated widely.

00:44:47: In all likelihood Armada investment in the lead here had certainly participated.

00:44:53: In April of this year a seed round of twenty three million took place.

00:44:58: This round led by IQ Capital and Secure Ventures Company and Armada Ventures participated again as well and Holchim also invested.

00:45:09: And that was already an interesting mix of deep tech VCs and construction, industrial exposure with the strategic capital from Holtzium.

00:45:17: The valuation at the time were set a thirty-one million pre money... ...and they therefore had little over fifty million postmoney.

00:45:25: Great!

00:45:26: Now SoftBank is coming in with two hundred million US dollars as apparently the sole investor.

00:45:31: That's completely different dimension now.

00:45:33: According to reports the valuation around one billion U.S.

00:45:37: Dollars Post Money.

00:45:39: What's particularly interesting is that there had previously already been speculation that SoftBank had considered acquiring Gravis.

00:45:46: Now, there won't be an acquisition but instead a very large growth

00:45:49: financing.".

00:45:50: We looked into this a little and the two hundred million will probably be invested in several closings.

00:45:56: In an initial closing now around fifty seven million Swiss francs were invested.

00:46:00: at evaluation of seven hundred ninety million US dollars.

00:46:04: That already represents a sixteen X increase in value for the first investors.

00:46:09: So that's not too bad, a sixteen X within three years and once

00:46:20: From a market perspective, this shows once again as we already discussed in the news that physical AI is gaining significant importance right now.

00:46:29: After the first major wave of generative AI it's increasingly about transforming artificial intelligence into the real world so into machines robots vehicles industrial facilities.

00:46:42: Gravis fits exactly into these thesis.

00:46:45: The construction industry is enormous and not very automated suffers from a shortage of skilled workers, and at the same time faces high productivity pressure.

00:47:18: machine types and application areas, then a potential operating system for earth moving in heavy industry emerges.

00:47:25: That's obviously a major difference between product on the platform which is what they're aiming for here And with the two hundred million gravis now wants to expand internationally build teams accelerate commercial deployments end of course significantly advanced product development.

00:47:41: but robotics doesn't scale like software.

00:47:44: Every new machine, every construction site and every customer brings real operational complexity.

00:47:50: And sometimes it's also very individual... ...and accordingly Gravis now needs many field engineers meaning people who work on-site.

00:47:58: they have to build support safety processes hardware supply chains integration with the machines insurance issues regulatory issues.. ..and very robust customer support.

00:48:08: And on construction sites, the systems have to be extremely reliable because mistakes that are extremely expensive and can also become dangerous very quickly.

00:48:19: an ETH spin-off, robotics roots in Zurich.

00:48:24: You have previous top investors with Armada you have strategic industrial partners on board and now SoftBank as a global investor.

00:48:32: that's very strong validation And it also shows That Swiss deep tech isn't only relevant In life sciences or quantum but Also physical AI and Robotics.

00:48:44: Zurich has a very strong foundation with ETH and the robotic systems lab, as we've seen once again here.

00:48:51: And Gravis is an example of how a global infrastructure case can emerge from that.

00:48:56: so That was it for burn rate The VC insider podcast.

00:49:01: if you want to support our podcast Subscribe to our newsletter and share it with your network.

00:49:06: Thank you very much for listening.

00:49:08: We wish you a great weekend.

00:49:09: take care and goodbye.