100,000 Radios Per Store: The Quiet IoT Revolution Reshaping Retail

This episode is brought to you by Progressive Insurance. Fiscally responsible. Financial geniuses. Monetary magicians. These are things people say about drivers who switch their car insurance to Progressive and save hundreds. Because Progressive offers discounts for paying in full, owning a home, and more. Plus, you can count on their great customer service to help you when you need it, so your dollar goes a long way. Visit Progressive.com to see if you could save on car insurance.

Progressive Casualty Insurance Company and Affiliates. Potential savings will vary. Not available in all states or situations. Not every sale happens at the register. Before AT&T Business Wireless, checking out customers on our mobile POS systems took too long.

Basically a staring contest where everyone loses. It's crazy what people will say during an awkward silence. Now transactions are done before the silence takes hold. That means I can focus on the task at hand and make an extra sale or two. Sometimes I do miss the bonding time. Sometimes. AT&T Business Wireless. Connecting changes everything.

This week on the Mr. Beacon podcast, we're talking to one of the best connected, most experienced, most insightful executives in technology and retail. Thaddeus Segura, who's Vice President of Product at Vusion Group, who are leading the way in electronic shelf labels. Now... We talk about AI because he's done some really interesting work there, but there's a massive wave of disruption that is not AI.

that's impacting retail. And it's the deployment of IoT in massive quantities, huge numbers of devices. Those devices are electronic shelf labels. And that means... There's a level of connectivity in stores that we have never seen before.

A killer application, dynamic pricing, more efficient deployment of that pricing information is just the tip of an iceberg. And in this conversation, we're going to... unpack that and see what the opportunities are, what the issues are, and how this is going to impact how we shop and how we make money out of making that shopping experience better. for consumers. Enjoy the conversation.

The Mr. Beacon Ambient IoT podcast is sponsored by Identive, whose IoT solutions create digital identities for physical objects, enhancing global connectivity for business people and the planet. By Bleakop, who enable physical products to communicate with cloud applications using Bluetooth Low Energy. And by AmbientChat.ai, we connect people, places, products, and brands with a digital assistant using AI and Ambient IoT.

So Thaddeus, thanks so much for joining us on the Mr. Beacon Ambient IoT podcast. Thank you for having me. I appreciate it. It's a return visit, of course. The last two times, we were both part of Williart. And since then, we've left and you've gone to Vision Group, which is a super interesting company. Tell us who they are.

So Vision is the world leader in electronic shelf labels. So this is an electronic shelf label here. The way this technology works is it's built on a display technology called EEK. And what's really cool about it is we pull... like different colored molecules to the surface. And then once they're there, you can basically remove the battery and this is persistent. So once the image is set, you really don't consume any power.

basically allowing them to operate for a very, very, very long period of time on a very small battery, which is kind of the goal of ambient IoT. Like we wanted to remove the cables and data and everything else. Absolutely. Directionally still aligned with all that, but that is our primary product. What I do at Fusion specifically is pretty much the use cases past that. So once you have a network that's already deployed...

How do you harness the BLE signals? How do you deploy computer vision? How do you deploy different AI models? So there's a lot of overlap between what we were doing before and what I'm doing now with just a little bit more computer vision involved. Absolutely. It's a fascinating area. And I really feel like, A, your company is seeing great success. And, you know, retail and IoT technology, technology generally is a fascinating area because we all shop. great way of influencing people's lives.

Your company is just on a storm. The Walmart have announced the deployment, thousands of stores with your technology, and it's going to change the way people shop. in the United States. We've been basically dealing with 20th, arguably 19th century labels for a long time. And now suddenly the information that shoppers see will be very dynamic. And so I want to explore.

i want to understand more about the company but uh the technology and also yeah okay it's an esl but what about all of the things around it what does it unlock what are you seeing in europe what are you seeing in the in the states so there's a ton we can talk about. And also, I do want to get into how AI and so forth hits this. Fusion Group itself, let's just kind of flesh that out a little bit.

I read somewhere 30 years. Is it really 30 years it's been going? 30 years. I mean, ESLs have been around for that long. I'm surprised. Yeah. Back in the day, they were small LCD segments. Imagine your old calculator or your old alarm clock. So very different technology. This is now like an image. It's like a set of pixels, essentially. So very different technology. We use the same communication protocol they used in submarines. So unbelievably slow, but very robust waveforms.

And a lot has changed since then. Like this current technology is probably closer to about 15 years old. So things continue to adapt and change and adding colors and other capabilities are constantly being introduced. Very good. And so I assume that the core of the business, ESLs, is where a lot of the revenue is coming from. What sort of volumes are we... talking about the numbers i saw were unbelievably big um considering i can't remember the last time i saw an esl in san diego yeah

They are very large. We're talking like tens of millions to hundreds of millions a year. Yeah. To your point earlier, we've been growing at about 30% year over year for over a decade straight. So trajectory has been fantastic. And to be honest, Europe is pretty far ahead. I mean, there was a lot more sustainability plays in Europe. And so I think there was earlier adoption of the technology. The US is just getting started. And we're also seeing markets like the UK really take off.

So we're just honestly at the very beginning of the journey.

But to your point earlier, it's what happens after you have these deployed and you already have a network in the store that things get really interesting because you can tap into that network. Yes. So these things, you talked about the submarine technology to propagate the signals. What are they using now? So we have two protocols. We have an HF protocol, which is like our own proprietary.

But I think the future is really Bluetooth. So we were the first to actually do this over Bluetooth. It is way faster, so it provides way better latency. A good example of that is that there's LEDs on these tags. And you can basically pick... from a pick list or shopped from a shopping list. And if you're wandering out an aisle and you can't find the pasta sauce you're looking for because they all look the same, you can press a button and light up the LED on there. But...

A lot of stuff has to happen between that button press and going to the cloud and actually talking to the label and doing that in a way that doesn't kill the battery. So moving to Bluetooth has really unlocked a lot of that. And then once you're on Bluetooth, you can talk to devices.

So suddenly this phone can talk to this label. So if I have a shopping list in my app and I come within, let's say, 30 feet of this label, I can send the API call. And by the time I'm approaching the section, it's already lighting up because I know what's on my shopping list.

yeah so and that's not like those are things that are live now that we're basically implementing with major retailers this is not science fiction this is not a roadmap like those are very real features that are already built That's so cool. And I remember we were interested in an adjacent use case where you're using the lights to help. You know, basically, you've got the new guy who's working at some grocery store and...

If you can speed him up, then basically less out of stocks, better shopping experience, better productivity, all that sort of stuff. To what extent are... You said it's being used. Are we in the early stages? How many people actually live with using these lights to guide store team members and shoppers?

Yeah. Where we see the most adoption is on pick to light. So omni-channel fulfillment. Yeah. And that can be either for your own people or for third parties. So you can imagine exposing that to someone like an Instacart just as an example. Yeah. So that is already pretty much fully deployed. We're talking about millions of API calls a week. So that is the primary use case. And the reason it works well is that you have a list.

And typically it's an ordered list. So you don't have to do beaconing to anything that could potentially be on there. Yes. To your comment earlier, like what we were trying to solve before with stocking is much harder because if you have a cart, you can work in any order.

So trying to do that seamlessly without having to introduce a scan becomes a lot more difficult. And that was what we were trying to solve for at Williard. Yeah. So in Europe, continental Europe, countries like France, this is pervasive, right? In the UK, you had the announcement with Morrisons and the big nationwide chain, and Walmart are doing it. I'm sure you can't comment the specifics of Walmart, but why...

Why do you think the US took longer? And this is not the only area where the US has lagged. Like, you know, NFC payments is like another classic example. Why? Yeah, I think... the cost and availability of labor is the primary driver. Like the whole UK market's on fire right now because there have been recent changes to the cost structure for retailers to employ people.

You can imagine how much labor is associated with actually changing these tags. But there's always been like a give and take. I think the primary barrier for so long is that we have many very large retailers in the US. So the capital expenditure like that.

the one-time investment you have to put in is a really hard thing to get approved. Yeah. Even if it makes sense, like there's still at the highest levels, like that emotional barrier to writing a check that large. Yeah. Career decision, career decision for them.

Exactly. So I think that we're seeing that shift now. And I think that we're, you know, past the inflection point, you always need someone to move first. And the US had happened to be Walmart, which is the biggest. So we're not complaining about that. And yeah, we'll be fully deployed with Walmart by the end of next year, all stores. Wow. That's a large number of stores. So how many in a typical store?

Um, doesn't necessarily need to be Walmart, but let's say large big box grocery store. How many ESLs are there in a single store? You go over a hundred thousand. Oh my goodness. And in this. Walmart specifically, we actually have a different product I don't have in front of me called EdgeSense. So to reduce the radio noise, because you can imagine 100,000 radios in an environment, would destroy the BLE band. So we put everything on a rail.

They all share a battery and they all share a radio. So that brings down the component cost for label and it reduces the signal noise by an order of magnitude. So the rails go on the edge of the shelf. The tags go into the rail. And that basically becomes your network. We fix those as we install them. So we capture the location. So that's the input for triangulation and wayfinding and everything else.

And then in the future, you start clipping in things like proximity sensors, cameras, potentially RFID, whatever you want can feed from the same radio, the same battery, as long as we're really intelligent and pragmatic about how we use the available power budget. I could tell that your marketing people had a great time making the videos for Edge Sense. It's like...

super high production. I'm like, by the end of watching two of them, I wanted to buy an Edge Sense rail for my house. It was like so good. I remember my onboarding. It was like the first thing I saw. And I was like, oh, this is polished. This is nice. But yeah, it's a great product. And it was like a step change for us. We were the first to consolidate everything into one device.

And it really unblocked us because you can't deploy 100,000 unique devices in an environment like that. The BLE channel is just too congested already as it is. So you have to start getting really intelligent about how you... use those few channels that are available.

So I want to unpack a little bit more all the things that you can do when you go from no radios in a store, which is kind of the way it was maybe 20 years ago, to now where you have tens of thousands of them and what can we actually do. Let's just for people to really understand what is the ROI on putting this technology in? Why would a very cost-conscious company like Walmart write what must be?

a huge check to deploy all of this technology in every store what's the benefit yeah so i think there's two things for walmart so first is if you use any tags at all whether that's standalone or edge since you save a ton of labor on price changes. And you also accelerate how quickly those price changes are executed. So to like double click into what the flow used to look like, a merchant or someone at the office would decide to change a price.

That would have to get queued and that would have to get sent down. Then the store would have to go through and put special labels into the printer and print those out and then wander around the store. or use a handheld hip printer and find a thousand items across a three-acre area and do those one at a time. Now, a merchant changes a price, they press a button, and then it updates.

It's that simple. We typically do it overnight. We don't do surge pricing. We're not watching people oh no i was thinking surge pricing no no certain pricing we're not you can have people racing around you you've got five minutes to pack everything in the store and the price isn't going up yeah so fun no no no no surge pricing that is not the goal and honestly like okay it wouldn't even work from a technology perspective because the update of the screen is what drains the battery

Oh, okay. So if we were updating the screen multiple times a day, it would completely change the function of the device. And yeah, we would never go that route. Okay. So we went back to kind of the ROI, the core use cases. And so just...

not having people with a stack of price labels to printing it out, all of the things that go wrong. We've all seen it. People go to check out, this is the wrong price. And then suddenly the queue's getting longer and it's just... chaos the manual way so we automate it it's less labor less error and you can do it more often not every hour but

A bit more often than you could. A great example is tariffs. I think in the first 45 days of the year, we saw tariffs change 30 times. If it takes you an average of 14 days to execute a price change... uh you're gonna are always be behind like we even did the math for like what it would cost a retailer just to have taken 14 days and it was in like the hundreds of millions of dollars in terms of lost margin so

You definitely want the ability to be agile without it introducing incremental labor to go through and execute the price changes. Yeah, makes sense. So what beyond that then? So with EdgeSense, you start getting into like planogram compliance. And what I mean by that is that a lot of retailers send down like a centralized planogram of what should go in each area.

And that will have the number of facings and where the tags should be located. And what is a facing? Facing is like how many unique. So if I have, you know, a can of Red Bull, if I have two next to each other. there are two facings on the shelf. It's not about the depth, it's about how wide it is. And that's a very relevant question. There's actually a lot to balance there. Because the more facings you have, the more holding power, but also...

There could be some efficiency gains or losses in terms of how you stock from the case directly onto the shelf. So there's a lot to actually optimize even in that one little lever. So if the store starts to deviate from that... even if they go from like two facings to three facings. If the case pack is coming in a certain shape, it could actually be slower to add an additional facing. And so you really want to maintain the planogram integrity, but historically there was no good way to do that.

With EdgeSense, we're constantly listening to the tags and the rail to understand was the tag removed or did it go offline? And what we find is that a lot of tags get moved very, very often. And it's a signal that the mod is changing or someone down the ladder has basically pulled a label and spread and filled. And we can send signals to send them exactly to that particular location of the store. Go to Zone A.

IL-17, the fourth section, the 35th label, and correct it. So what we've actually seen is a huge impact in the planogram compliance and what we call modular integrity by unlocking these use cases. That makes sense.

So what about the information on the label? I guess one of the benefits is you can go from simply having a price and maybe a short description to something that could drive Lyft that will entice. people with more information. But of course, you need a bigger label that costs more. So what are you seeing in terms of the adoption of larger color advertising space versus?

something that just kind of automates the update of prices it's a great question because i think retail media is like a going to be a very exciting place over the next couple of years so we're seeing i think steps toward it with existing labels, QR codes to send people directly to the website, plugging into POS data to understand when an item is out of stock. And you can even update it and say expected stock coming in at this date.

We've even seen some people basically tap in inventory information. So in one customer I know of in particular, they actually say, we believe we have six on hand. And so if a person is helping someone... They can say, oh, it's not here, but we have six. They're somewhere in the store. Let me go look for them. So the nice thing is it's basically just a screen. You can put whatever you want on there.

whether it's a qr code whether it's a product rating clearance rollback whatever you want to highlight really you have complete flexibility but in the future i do think we're going to see more and more like full color displays More things that are replacing paper. So you think about ads that stick out in the middle of the aisle. I think we'll see a lot of that start to get moved over to things like e-ink as well. Yeah.

And I've been thinking in my mind during our conversation, big box retailers, grocery, that sort of thing. But what are you seeing in terms of adoption across different categories of retail? assuming that... You know, an electronics retailer might well see significant value with larger products that are more complex and being able to overcome all the challenges you have with the people selling the product not really.

knowing about it and all that can you talk a bit about that it's it's much broader than just big box and grocery so then they represent a big portion of the business but you have things like diy which is like home improvement convenience stores

because you can imagine there's not a lot of labor that's available. There are industrial use cases, like we have some large displays that actually have cellular tags that go into cars to display the price, like the big sticker that would normally be on the window. And also keep track of where the cars are at in the lot. Because to be honest, it's actually a hard thing to solve for the dealerships. Yeah. So there's like a really broad assortment of use cases.

pretty much anyone that displays crisis we're seeing demand in those environments and of course apparel i think that was our first major u.s customer was kohl's and we had those those signs that years ago very good so

Let's talk about some of the adjacent products that complement the ESLs. Perfect. So I think what I'm most excited about is computer vision. So we have a solution in-house called Keptana, and it's both hardware and software. And we basically have the ability to deploy cameras flexibly. Like we've talked in the past that if you have to run power and a data drop in a retailer, it's unbelievably expensive.

from $1,500 to $4,000, takes weeks of approval, there's liability, and it really kills most projects. If you can deploy things flexibly on battery, all of a sudden you can do all sorts of use cases or previously cost prohibit. So we're seeing a lot of success deploying these cameras. They do a couple of things. First, we obviously identify out-of-stocks.

So that is the most obvious use case you can imagine. If something's out, you can send a signal to someone to replenish it automatically. So where is the camera facing? Often on the other side of the aisle. So you'll have cameras on both sides of the aisle. They're very small.

Not much larger than an ESL. We even have versions that are like hidden behind the rail or embedded into the end caps that are in development right now. And the idea is that you don't really have to do anything. You could just plug in and use the existing network. take the pictures, and then send them up to the cloud for processing. Here at Blue Apron, we know exactly how hectic school nights can be. That's why we created Assemble and Bake.

Delicious one-pan meals that make family dinner simple. Just assemble the pre-chopped ingredients and put the pan in the oven to bake. Then you're free to help out with that last-minute diorama. Shop, assemble, and bake at blueapron.com. Get 50% off your first two orders with code APRON50. Terms and conditions apply. Visit blueapron.com slash terms for more. This episode is brought to you by Progressive Insurance.

Do you ever think about switching your insurance companies to see if you could save some cash? Progressive makes it easy. Just drop in some details about yourself and see if you're eligible to save money when you bundle your home and auto policies. The process only takes minutes and it could mean hundreds more in your pocket.

Visit progressive.com after this episode to see if you could save progressive casualty insurance company and affiliates potential savings will vary not available in all states. So you're not streaming video here. You're just taking snapshots and uploading them. So you can see out of stocks. But I'm now thinking about power.

you know, even if you're just taking snapshots, that's going to consume more power. How do you address that? Is it viable to put solar on these things to start to kind of ambiently harvest the power? Absolutely. Power budgets are number one concern. So we never stream video. So we're not watching customers or anything like that. Because the most important thing for us is how do we maintain the battery life over as long a period of time as possible.

So most retailers are somewhere between like three and 12 pictures a day, depending on what kind of cadence they want. And of course that can be configured by which area of the store you're at. Plastic containers and home lines doesn't sell as quickly as cereal. So you don't need 12 pictures a day. But to your point, we do have technology and there's even a version of the EdgeSense rail that we produce with a partner that has integrated light harvesting.

that's optimized for indoor lighting conditions. And that's basically allowed us to completely autonomously power the rail. And those same technologies can obviously be applied for things like computer vision or anything else that we're developing. I think where we are really interested in exploring that technology is also around CD edge compute because the bulk of the power budget comes from actually transmitting the image up to the cloud.

So if you could do some of the processing, like extract the only things you care about, like do object detection at the edge and only send the things within the bounding box to the cloud, you can save 70, 80% of the data. because you only care about certain parts of the image. But then you have to offset that peak power pole to actually run the edge compute. So those are things we're thinking about. That is interesting.

So I do want to go back to the product line and your strategy, but just a very practical point. It's clear that... This can get complicated and that the opportunity is enormous strategically. This is not just, let's get some different kind of labels for our products. This is... strategic. How do you go to market and how do you help your customers?

you know, unlock the full potential, put the strategy. It seems like that you need to really be management consultants, retail strategy consultants, as well as... experts in e-ink and radios and power consumption and all that sort of stuff i actually think that's one of our competitive advantages so i normally draw it out as this flywheel where we have hardware

We have software and AI and we have applications. And normally if I'm a major retailer, I would go to one provider for hardware, a different provider for software, different provider for the application layer. Then they all have to talk to each other and I have to coordinate and keep them all in line. And it's really difficult to manage. For us, we build hardware, we build software and AI, we build the applications. We're able to do some really creative things because we can pull the levers.

and control the roadmap of all three of those groups. So the retailer just comes and talks to us. They just tell us what they want. They don't need to understand power budgets and Bluetooth channels or anything else. They say, hey, this is the problem I...

I want to solve, this is how much money I have to solve it. And then we basically build the solution in the end for them. And do you get third party systems integrators involved or are you just doing all the work yourself? Definitely. I mean... What I optimize for is speed to market. I think that agility is the most fundamental form of competitive advantage. IP, price, all those things are really hard to actually defend in the long run.

But if you're the fastest and you can execute well at speed, I think that's what really makes you sticky and really dangerous. So if there are third-party solutions that will help us accelerate, and there's a few I have in mind that are too early to mention. But we're working with some people right now that are allowing us to save a year in terms of development. And that's really, really significant. So absolutely, if third parties make sense.

great, but we like to still be the interface so that the customer doesn't have to start managing all these different relationships and get right back to the status quo. Okay, so you would typically prime the project and have the other work with the other folks. Very good. Okay, well, let's get back to the stall of Vision Group products. So there's computer vision. What about other things? I'm interested in help.

Are you doing anything in the content space? A little bit, yeah. So we do have a full line of like full color displays. We do have some wired displays as well. So we have a whole product line called Engage. And that is all geared out around retail media. So the e-paper displays go as high as 72 inches. And these are battery powered and you could just put them up anywhere without running power or data. So you could imagine the flexibility you get from those things.

But if you're going to deploy retail media, you need a content management system. So we do have a content management system in-house. We're not generating the content ourself. We're mostly getting it from third parties. basically managing it, making sure it's the right format, pushing it down. But of course, there's always conversations around, you know, how could we use AI to facilitate these processes? Yeah. But I think it's too early for us there. We're not quite there yet.

Well, it seems like everybody does have an AI strategy or a reason why they're actually an AI company. We don't make X. We're actually an AI company. Where do you guys stand on that? Well, we're a little of everything. I mean, I think that we have historically been a hardware company. And I think now we've pivoted to more of like a...

a holistic technology company because a lot of our revenue now is from the cloud and from other value-added services. But I think the aspiration and where everyone's trying to get to is to be an AI company. Where I think we're unique is that I think the barrier between where retailers are now and where they need to get to to be like an agentic store is that they need better signals of what's in the building. Yeah.

Talk about this life cycle of retail where 15 years ago, stores were just stores. And then Omnichannel came along and stores had to operate as fulfillment centers. And some retailers leaned into that, and most did not make the investments required to understand what they owned in the building and to optimize processes.

And now they're fulfilling through third parties and their customers are paying twice as much for the same goods. And it's created a lot of pain and frustration. And it's almost too late to go back and fix some of that. But now I think people are understanding that if I want to be a gym teacher, I want to use AI.

I need inputs. The only way to get those inputs is to deploy some sort of sensors in the store to be able to understand what you own, how many there are, where they're at, how customers are behaving, where your associates are. I think where we're really uniquely positioned is that when people understand that discrete list of inputs they need, we'll be the best person in the world to provide those signals to them so they can truly unlock the intelligence that they see as so alluring.

i mean the thing that i find really interesting about uh vision group and what you're doing is You're kind of in the end zone. I'm not a sports fan, so I'm going to mix these metaphors up. Basically, just as the goal is about to be scored, you're kind of there and you have a play in optimizing... the retail shopping experience as it is universally but also i think um you know the thing that i find interesting is

your potential ability to talk to apps, to have these labels talking to apps, and maybe it's QR codes, maybe it's Bluetooth. But what I would love to see, and tell me if I'm just... smoking something. But what I want to see is the kind of organic specialty... food experience, but scaled massively from the likes of a Walmart, where you can find out where this product came from, how it was handled, and enjoy the story of the product. And also just...

get better information about what it is you're voting for with your wallet. And the use case that I would love to see someone do is you have tags on... on cases of product, maybe it's cardboard cases, maybe it's plastic crates and you're tracking them coming from the back of the store to the front of the store. And then people take all of the apples out and they stick them on the shelf. And somewhere there's a relay of the ID that's on the crate that...

and a serialized identity gets passed on. It's like a relay race and you're kind of right in the final lap of the race and you get it. And then I come walking along with my AI app and I... pick up these signals and I can then tell this amazing story about what farm it came from and how far it traveled and whether it's really organic or not and all that sort of thing. Is that possible or am I just like hallucinating?

Technologically possible, very difficult logistically to implement. So we do have a product that has a label embedded into a plastic tote for a very similar use case to what you're describing. So we have temperature sensors in here and we have local memory. So the beauty of that is that you can have the tag on there. You can harvest lettuce in a farm.

and put it directly in there and start tracking temperature without infrastructure on the tractor or in the field because it's really hard to do. Then as it moves through the network, anytime it hits an access point it's compatible with, it'll download the temperature.

So everything you're describing, like all the way to the sales floor of the store, we've done pilots with this technology. Like it works. We have continuous data. Nothing is lost. We can store weeks of data on there. Even if it's sat.

in a DC for a month, that isn't an issue. The challenge is that when you actually try to roll this stuff out in a supply chain... totes are like a shared asset across multiple retailers across multiple geographies yeah so one tote could go to retailer a go to country d go to retailer c go to a wash center and so there's no way to deploy a controlled pilot

And when you're building up the use case for technology, you have to do something small, prove that technology works, and approve a concept or economics to show that it actually provides the value. Yeah, this is where platforms come into play. This is why your product, ESLs, is so interesting because it goes in on one use case and then potentially you can use it.

for another and it's so difficult to introduce a new technology platform but if there's a killer use case for one thing then suddenly you can use it uh for this other thing and i'm not making much sense here so let me unpack this like ifco you know they announced this deal they're buying hundreds of millions of bluetooth tags from our sponsor identif and so you know there's going to be a lot of crates with bluetooth that's going to last for years and years and so you know

That's going to be there. So the question really is, is it feasible for a device like your devices in the rail and in the label to read Bluetooth signals from crates that are... are being rolled out onto the shop floor? Or is it just too much power to reliably listen? The simple answer is yes, it could be done. And especially with technologies like light harvesting, because...

A lot of stores, especially like on the top shelves are quite bright and you actually do get pretty good power. Yeah. The challenge is always just low level integrations because. There's still barriers between having a network and being able to tap into it for individual use cases because we're all also still trying to optimize for privacy and security and everything else.

So it's complicated. And complexity is the enemy of these things. So you kind of have to do it in steps and stages, don't you? And that takes time. Let's just make sure that we haven't missed anything in this conversation. I do want to get your perspective on Europe versus the US. On one hand, Europe... is ahead it's been ahead in payments it's been ahead in uh sales what are the sorts of differences you you see in the geographies that your very large company services yeah so definitely

payment sustainability so anything related to sustainability feels so much more visceral when you're in executive meetings with european retailers yeah like it doesn't seem like a kind of like afterthought or oh it's nice to have or oh we'll do this and take credit for it if it has a sustainability benefit yeah it's very much like a consideration so if you gave them two solutions and one was slightly more expensive it had

significantly lower carbon footprint a lot of times that is what will win so i do think there's more of a focus on sustainability i do see less centralized data So in the US, a lot of things like planograms, pricing, assortment are dictated by like a core office or even like a regional office. In Europe, I see a lot more autonomy.

So a lot more flexibility, but that also means worse data. So there are some major retailers that don't even have clear visibility into the assortment of what a local store carries. They know 80%, but 20% is... their best guess of what might be in there at any given point, which is actually really, really surprising to me.

And then the last thing, and you can debate if this is ahead or behind, but I actually think Europe is in a much better place around AI regulation because I think that they see some of the risks and potential negative ramifications that could come from... completely unregulated ai and so i think they've done some stuff proactively that has probably impacted like the speed of innovation but i think will be much better for people in the long run yeah i do think that europe has a so

It has an advantage in context. The U.S. is inarguable, like the model development and the trillions of dollars that are going into that. No one's going to... beat that you know china's uh giving us a run for our money and uh so forth but In Europe, with digital product passports, standards, legislation, I just don't see that happening over here, and it is happening over there.

I actually think when AI meets data and context, which is what digital product passports are, then all sorts of things will become... possible i i just hope that europe uh you know sees it through rolls it out uh leverages it but uh i think there's huge opportunities there i don't know whether you agree or i agree yeah and

I used to think of it as like linear, like who's ahead and who's behind, but now they may just take different paths. Yeah. We may just be on a path in the US of we're not going to regulate and we're going to let businesses grow as quickly as they can. despite what happens or it will regulate way after the fact when the pain is very visceral and very real. Yeah. And I think Europe is just taking a different route and I think I see those things continuing to diverge more and more, but, you know.

There are opportunities on both sides. So Thaddeus, you took time off between Williart and Fusion Group. What did you do in that time? I did a couple things. Of course, I went deep in LLMs and small language models. You know, all of the work seemed to be flowing there at the time.

And so I wanted to understand what the limitations were and actually play with it. So I did a lot of experimentation. I think the first thing I found was that it's expensive. Yeah. So when I used to find... tune models at home I might be able to fine tune a decent size computer vision model for you know $50 $100 just to experiment I was trying to fine tune large language models and it was pretty expensive from a compute perspective and that was like

Yeah. And so that tuning is prompt engineering or parameters or what were you doing? No, going deeper than that. Yeah. So I think a lot of people today are taking LOMs and they're plugging prompts in. Yeah. And now there's like a ton of context that's pumped into all of these models. Yes. But I was actually curious about like how I could rewrite it. What I was actually trying to do was I was trying to.

come up with a model that just ran persistently even if someone didn't interact with it and basically like make it reflect and if no one gave it input in 30 seconds it would have introspection so just like you know pull from a different LLM to say like Who am I or what do I believe? But there are certain things hard-coded into the model to make it not think it's a person or not think it's persistent.

And so to override those, you can't just do prompt engineering because they're baked into the parameters. So I had to like generate training data and a backstory and internal monologue and then actually fine tune the model on those things. That sounds really interesting. So did you say you were using two models to interact with each other or did I misunderstand? No, that's exactly correct. Yeah. So I would like generate content.

some of the questions that didn't have to happen in real time from like a larger model and then run like introspection like a smaller model almost think of it as like um different models of different parts of the brain yeah so it was supposed to basically be multiple models interacting with each other and

my thought process was that all of these things together could give rise to something that would feel like consciousness. And I think what's missing from LLM is this ability to reflect because there's a cost associated, of course, if you're just going to run a prompt when no one's watching. Yeah. But I think it's a very, very interesting area. And there lies the ability to start to put... better executive functions in place if you can structure that and harness it.

I mean, at a much more simplistic level, I've been using these coding models, cursor and codex and more recently, anti-gravity, which is pretty good. But, you know, even anti-gravity was having problems. The development work that it's doing is so beyond me in terms of the actual coding that I basically had GPT 5.1 doing code reviews. of the code that Opus 4.5 was doing. And it was almost like, I mean, I was basically just kind of copy and pasting and throwing in provocative questions.

Seeing that was just amazing, that kind of collaborative aspect that you can orchestrate. I completely agree with you. And when I would get into sensitive work that I was like, should I patent this? when I was thinking about founding, I would break the work up and give it to different models so that no one had the cohesive view. It was probably just paranoia, right? But I would give Gemini a bit, give Anthropic a bit, give Tetchum a few a bit.

bring it all back together uh on my own because it's it's also hard if you're doing great work like how much of that do you own how much of that do they have rights to and i think that's still to be defined and i would hate for it to be defined after the fact and you know have there been negative repercussions from that yes i mean we're kind of at the risk of getting into my own business uh my

The motto of the day for ambient chat is own your context. And I feel like both individuals and companies are just. kind of wandering along and blithely surrendering all of their data. And we know that the data is where the value is, and it's kind of being sucked in. And so we have to... be mindful about that. And personally, I think there's a business in building tools for someone to collect their context and have portability.

and make sure that we don't just completely surrender to these massive giants. But anyway, we won't get into any more of my AI paranoia. There's a ton of... questions that I'd love to talk to you about. But, you know, I really appreciate you coming on the podcast. And thank you very much for joining us today. Of course. Thank you for inviting me. So that was my conversation with Thaddeus.

He's an amazing individual, a real model of many things in what you would want to find in an executive in our space. So thank you very much for listening to the show. Thank you to Aaron Hammock for editing and distributing it. And I look forward to seeing you next time. This episode is brought to you by Progressive Insurance. Fiscally responsible, financial geniuses, monetary magicians. These are things people say about drivers who switch their car insurance to Progressive and save hundreds.

Visit progressive.com to see if you could save. Progressive Casualty Insurance Company and Affiliates. Potential savings will vary. Not available in all states or situations. Say what you want about AI, but it's here. And it's helping businesses get more done in a day. Wix's website builder is infused with AI so you can stay ahead. Create a beautiful, functional website just by describing your idea. Track how your site appears in AI search results.

Create custom images on demand or launch an entire campaign in a matter of minutes. Wix gives you AI wherever you need it. Try it now for free at Wix.com.