iPhone in a Shoebox: Building AI for the World’s Dirtiest Industry

0:00There's so much scope of optimization for founders. Just don't think like, you know, there's just one way which is generated. There's every, and you don't have to be another, like, you don't have to build a god brain for the physical space either. Just focus because intelligence is hidden in people's experience in closed door sub economies. It's not accessible on the internet. Build relationships, go in, use technology, learn. and you build incredible companies out of it.

0:32Ambarish Mitra: Ambarish Mitra co-founded Gray Parrot to bring computer vision into recycling facilities so operators could understand what's moving through their plants and what valuable material they're missing. But knowing how to build AI didn't mean his team understood the waste industry. We talk about earning access to their first facilities, collecting training data with an iPhone in a shoe box, and discovering why a working product wasn't enough to convince customers to pay. I'm Walter Thompson. This is Fund, Build, Scale. Rish, thanks very much for joining me today. I really appreciate the time. Thanks for having me, Walter.

1:09What is your role there at Great Perrin, and how do you spend most of your time? So I'm the co-founder, but I'm also the chief product officer. So I am building the product and also lead the technology. so um and it's a large-scale computer vision uh model being developed for the waste industry uh where we classify and analyze all waste streams in real time um and and keeping in mind that the waste industry didn't have uh any kind of analytics uh before gray parrot uh came into the picture so So not only for like waste as a unique industry, but generally a physical world, physical analytics didn't exist in the scale at which Gray Parrot is doing it for the waste industry.

2:06What's your day-to-day look like? What do you spend most of your time doing? I mean, most of the time, of course, I spend a lot of my energy hiring. You know, we are always bringing talent on board. beside doing product management the CTO of the business works with me closely so analyzing what's being on the tech I'm also very involved in multi-generations of our hardware so what's coming inside the hardware and it's a constant optimization so like our version 5 we call it GP5 is out in the market but we're looking at version 6 and are we going to continue with the same camera, same GPU. So there's seven, eight different streams of work streams, which is looked at iteratively besides the usual finance meeting, exec meeting, board meetings, et cetera.

3:03Ambarish Mitra: Usual executive overhead. Yeah. So before Gray Parrot, what were you and the team building? And how did that tie into what Gray Parrot is doing now? So yeah, the Gray Parrot But initial founding team and early employees came from my previous business where we were building the world's first visual search engine on a large scale, not only to recognize everything in the world, but to make it editable using AR, which was a very complex and large thing to do. and we build a fairly large model of 1692 categories of everything in the world. And in sort of 2014, when the first version of the product came out, it could almost recognize most things in the world as you see it and tell you what you're looking at.

3:53And also it created the world's first visual knowledge graph where not only you can look at what you're looking at, but what its relationship to rest of everything else in the world is so um so if you look at a tennis ball it will of course tell you looking at a tennis ball but it can give you nodes of are you interested in tennis or wimbledon or roger federer or things like that and expand upon so you can start with one single point of curiosity and then infinitely navigate into several points but something sparks your curiosity so we had this massive experience of having built this in an era when this wasn't very common um you know and then dominant networks were 3g and 4g and much older generation of phones uh including we built for like even blackberry and windows phone to give you context of how old the technology was.

4:56And at some point, this learning was very critical to us founding Gray Parrot because we knew how to build large-scale vision models. And the post-consumption world is equally very complex. Just to give you an example, in a perfect world, a Coca-Cola can or a Pepsi can looks the same on your kitchen counter in an NFL stadium in a restaurant. But in the world of waste, that product has 20 ,000 faces, you know, one single product because of the nature of the crush and how sometimes it's unrecognizable. It's semi-recognizable. And there's no, it's not a finite number of permutation combinations you can play with how it'll.

5:52So how do you build a tolerant model which will recognize everything a human throws away in a fast moving conveyor belt with low lighting, etc. So because we had built the first business that already dealt with several complexities, not the ones with Grapear and Fest, but at least the foundations were at play, we were also motivated to solve this problem.

6:19Ambarish Mitra: I mean, but that's a very different problem than trying to create like a taxonomy for like the visual world, right? So, I mean, like the idea of, you know, how do we identify this token from beginning to the end of its journey? Like the whole, you know, after the bin, I suppose. When did you start to realize that that would actually be something more important than just kind of helping brands, you know, sell or advertise something? Yeah. I mean, when we were, me and Michaela, my co-founder was sort of the chief commercial officer of my previous business. and we we used to talk about this that actually we we can recognize everything in the perfect world and it really hit us hard when we realized that the world is optimized and particularly we were based out of silicon valley then it was optimized around this last mile which became the last meter last second delivery where It's all about goods and services being consumed and through various digital journeys.

7:26And once those goods reach your home and you consume it, and once it enters the bin, we realize that's where the digital journey ends. And that's one of the big reasons we have a waste problem, because we actually haven't measured truly in real time the waste world. Hence, we have a waste crisis. and a problem that you haven't measured there's no way you can solve it you know only then you can see that how much you're reducing how by what degree you're making it more efficient so there was there's no sense of tracking unit economics like it's very macro it's not bottom up you know it's like really top down there's a people look at the heap the user experience to understand what's going on but no one knows in any given container what exactly is in there so we thought that if we bring our expertise we call it the post-consumption world if we map it then many people can navigate it through this and and and and we can be one of the dominant players driving it but most importantly we have to map it that's when we decided we're going to create this world model of all physical objects that has been consumed and gone through the bin but tolerant enough that it's original shape and size totally being morphed doesn't matter you know if it's a coffee cup to whether it's a standalone standard coffee cup to a starbucks cup or a branded ice cream tub to a takeaway tub you know like non-standard to standard doesn't matter irrespective of materials irrespective of shapes if we can recognize and classify through various attributes you know because something can be something can be a material it can also be a brand it can also be a shape it could also have contaminant it can also be the quality of it you know how shredded or intact it is and then there's several other attributes, more intangible attributes.

9:39Ambarish Mitra: Are you building something that's hard to explain? It can be a sign you're working on something interesting, but it also raises the risk that you're missing out on fundraising, sales, hiring, and media opportunities. I help early stage founders sharpen the narrative. What matters? Why now? Who needs to care? And why they're the right team to make it happen. If you're preparing for a raise, a product launch, an important customer meeting, a panel discussion or even a hiring push, I can help you pressure test your story before the stakes get higher. Visit the Fund Build Scale website to learn more or email fundbuildscale at gmail.com.

10:16Ambarish Mitra: So what's possible now that wasn't possible before with Gray Parrot? Like what are some of the things that, you know, CPG companies could do or waste processors could do or even, you know, if I have an environmental regulator, like what are some of the things like with this data I could do now that i didn't have before you're spot on with that question uh walter in fact they're incredibly summarized for me we have three we are try silver tri-stakeholder company you know our stakeholder number one primary is waste processors you know we call them morse material recovery facilities and for everybody's knowledge like majority of uh stuff from your recycle bins ends up here.

10:59And in these facilities, the destiny of a product is not decided in your bin. It's decided in this facility, not in a truck, not in a curbside pickup. In this morph, it's decided whether it's going to go to be recovered or recycled, or it'll end up in some kind of a landfill, or it'll get burned somewhere and some kind of incineration that's usually the three outcomes and um and we thought we can make the highest impact here you know people have asked us this question why didn't you do a smart bin and and those look they'll look exciting but they're like a bit boutique you know how many bins will i go and knock on doors i have to become a consumer product company then because for us we almost wanted to be like a toll gate technology you know like entering a city and leaving a city If you want to count cars, like leave it there, you know, entry points and exit points in the city.

11:55We thought of that sort of traffic system for the journey of the commodities, for these recyclable commodities which people throw every day. So in a MRF, and MRFs, they are very manual processes, very human heavy, still in sortation and quality control and also they have a lot of mechanical stuff mechanical parts which are not digitizing they can just sensing and feeling and and they don't do color detection or they cannot do quality control and they miss a lot of things on the day so we thought we can build this cognitive layer which can understand 100 % of the waste flow in what's the total net positive in and net positive out and what's left which is residue and and is there any good stuff in the residue which they can mine again and then there are other good stuff like is the contamination and quality of their bail because if the quality is high these morse can set it at a premium because the contamination reduces the premium so it becomes unattractive but there was no way to measure so the trust was earned through perception and past experience so by us digitizing this it became certifiable it became transparent and and and it also became real time and we took guesswork out of it and and just to give you some sense the manual audit processes less than one percent you know so you know we could we basically made it 100 percent uptime so it made a very big difference secondly these are very capex heavy environments so they knew their multi-million dollar plant how is it running and is the throughput where the failure points they started having a huge efficiency gain so morse were a customer number one number two very recently became brands because we call them producers because they are the material producers they have the complex supply chain they bet on materials sometimes with five ten year horizon window like products you are having a shelf today was not thought about it today the supply chain for that was established ages ago so for them to turn the ship something that will impact us in three to four years or five years, the decisions have to be made today.

14:31We thought if we take the learnings from the MRF and send that information upstream, we could have a massive impact. They could choose more sustainable products and be more answerable. And also they do not know that sometimes we have our own, in our own experience, we have material biases. We think plastics bad and paper is good or wood is good but these are i'm not taking any side these are biases but in reality how things operate in a morph is the reality of life of waste you know right right not what you think is good or bad so if something doesn't get recycled it'll end in landfill that's basic you have to remember that so and we we shared this intelligence to them which is an eye-opener for them and lastly is the policy makers in general they are introducing policies like extended producer responsibility and few others and they were there's an element of guesswork or industry standard interviews where they get some information but because we have real real reports and real factual information we can inform them to introduce actual practical policy that could really reduce you know plastic pollution and increase sustainability but also make material recovery and recycling a profitable business you know which is one of the big goals here that's interesting i mean i like this from speaking

16:11Ambarish Mitra: as a consumer the reason why this technology like appeals to me theoretically is because i feel as though there's like a lot of um a lot of nonsense like where corporations trying to shift their burden onto me the end user for recycling and reducing everyone's carbon footprint and it's ridiculous there's no possible way that i could make a meaning i mean if i do recycle i compost and i sort everything but me as one person i i'm not i'm not the problem the problem is the upstream I believe, and the manufacturers. So I'm glad to see someone. And sometimes our data shows sometimes, because even we were a consumer once, like you and everybody else, and we've learned from our own data that it's not really that binary.

16:52It's not like brands don't have to take a bottle and they've, because brands differentiated themselves with their uniqueness of packaging and color and many variants, which makes their products quite a bit unsustainable for overall. But also it increases their cost. They're using extreme polymaterial carbonates, which then multi-materials, which now they have three to four different individual supply chains to put it all together. So we have enough research to suggest that actually sometimes 15 % change in packaging can lead to 60 % increase in recyclability. So it's not like you abandon and move from one end to another, like, you know, make your product go from strong, sturdy packaging to some flimsy paper, not waterproof.

17:51You don't have to make that extreme change. It's just the sleeve length, you know, from full sleeve, make it quarter sleeve with the same branding, you know, and don't have a sleeve like a bottom to the top fully like choking the bottle that reduces recovery by 80 90 percent you know so we've almost had the it's it's not like zero or one there's a lot of shades in between which doesn't make them compromise their brand values and standards but definitely has a huge impact on outcomes here's my impression of like what it looks like inside

18:25Ambarish Mitra: a waste management facility there's it's noisy it's very loud it's filthy it's huge um it it's not really well lit everyone's wearing like you know kind of protective gear and like you know earmuffs and there's conveyor belts with like trash waste being moving at like high speed and someone is trying to pick out like the stuff that doesn't belong and they're doing it manually and they're probably missing a lot of stuff but they're you know they're picking out you know know trash or soiled things and flipping them out of the bin and that's kind of pretty accurate and it's total chaos and also just for listeners mindful it is every city's 24-hour pulse it's not like it can stop you know you're all consuming with throwing bins and there'll be another mountain tomorrow which has to be so there's a time pressure so it's a bit chaotic too it's not like they'll leave it and we'll deal with this tomorrow tomorrow there'll be more so so the first time you walked into one of these places um like seeing walking in cold and seeing it for the first time yeah what was your how are you thinking oh i can solve this so very naively i did ask them where are the cameras because we our background was not hardware you know we were forced to build the hardware because they said like why do we need to have cameras we have cameras as security for the facilities as an entry exit.

19:46But this is trash. We don't have cameras on. We're not mining gold here.

19:52Ambarish Mitra: Right. So, and that made me think that where am I going to put a compute device here? You know, many, so the more city bound morphs are very stacked up. Some processes happening on top of each other and in more rural parts of the world, it's laid out like a football stadium so everything is very linear on a ground floor but no two murphs are alike in a way so and it's very hard to figure out what's the start and end point because there's a lot going on so it is very overwhelming for anybody you know it's one of those places where if you it makes you think you know like there are people in the world who've been to Himalayas or been to Amazon or been to Greenland.

20:39This is a bit same experience, but on the other side, it makes you think, oh my God, this is what we do every day. This is the almost shadow of consumerism and it's total chaos. And from our technical point of view, not only it was technical challenge, it was humongous. And we didn't know what to do, where to start. When we had not even

21:00Ambarish Mitra: develop one model to detect even seven different things forget about everything yeah how did you figure out whether the people in waste management were having the pain and the problem you thought they were having how did you get those first conversations yeah when we got we we knocked on many doors and many of those doors closed uh of course like any any any entrepreneur in a founder a journey you need to have persistence you need to have some level of self-belief first of all you need to love the problem so as we spend more time in it more the team and my co-founders wanted to solve it because we saw the opportunity we saw there's no digitization we saw lack of transparency we saw over reliance on human input but the problem is too big it's like this is one problem which connects every human to this problem on the planet.

21:54Waste is being produced by everybody in every city, every place in the world. So the scale of the problem was so big that we really wanted to solve it, except there were several barriers besides the industry itself not even adopted internet at a full scale at that point. They jumped in a big way into AI with us. so we had to convince them multiple we had to give these scenarios that what if we could recognize and tell you in real time how could you use this information it wasn't that apparent till you always need to have an ambassadorial ambassadorial client on the other side and we we found someone in our journey who believed in this and was forward thinking in the group and they said if this works if and if it works this way and it can create xyz scenarios i see it can change the industry we need it can you do it as it's not we need some time we need a lot of data there's no there's no data available on the internet for this it has to be primary can we set up a we can we set up something here is it working set up we don't have hardware so we literally put an iphone in a shoe box later on except the phone as many of you know it heats up the camera heats up the processors it was shut every few minutes so we had a we had a shift system of phones to be replaced oh wow like manually slapping them out yeah literally because we also and this is a positive part of the industry there is so much product that flows through like even even 36 hours would be like millions of images to give you data you know so these are these these plants they process so much if you choose a big city how fast is that conveyor belt moving like how many like three three to four meters a second two meters wide wow okay and we do not miss anything on it like it's totally if you look at it even as a human being you get a headache within 30 seconds it will create like a fuzzy layer in front of your eyes and it'll be like a blur uh and you won't be able to one won't be able to detect accurately what's going on so it's like it's continuous it's fast and in many places it's inclined and the lighting is very poor and it continuously vibrates right so we had to have like some level of autofocus and so from a computer vision problem limit was most unoptimum environment.

24:33Ambarish Mitra: I mean, this is a kind of a bit of a tangent, but like, could I adapt this or retrain this for sorting, you know, Yukon gold potatoes or strawberries or something else besides bottle caps? The thing is, we are a bit, due to the nature of the model, is a bit tied to not the objects, it's tied to the environment of moving conveyor belts. So it's a bit like, it's not a thing that you can go and take pictures of a bin and do it. It's too small. It's now like an engine with a big appetite. So only large chaotic. So if you have a construction demolition site, you're taking down a whatever, 50 floor skyscraper, and there's so much pipes material, this model will work.

25:17You know, if you're doing mining, if you're doing large scale harvest, you know, like for, and then you want to do quality control. You create, all olives look the same, but there's A, B, C, D variants. So it'll all come in a chaos. Chaos, moving at high speed and finding variations. It'll thrive in all those environments. Returns, you know, big retailers have conveyor belt of returns every day.

25:46Ambarish Mitra: Indeed, yeah, I'm dropping them off all the, I go to the post office and most everybody in the post office is returning something. All right, getting back to the customer discovery process. Who opened those first doors for those initial conversations? So we had one customer in England and one customer, you won't believe this, out of South Korea. And we traveled there all the way to get to meet and greet them. And they shared a lot of data with us. And then one outside early days, Heathrow Airport in London, massive morph. and not with the promise that they will become our clients, but just to test what we are saying, there's an upside.

26:28And then for a certain amount of time, that is for them to get the benefits out of that for no cost to them. But we did develop our product. We were not like, we know it better than you mindset. we were very humble about that we know computer vision and large-scale vision-based model development, but we have to understand waste and waste constitution and how they perceive what's... Because the consumer's lens of waste couldn't be more different to waste professionals' view of waste and what they think is valuable. So taking that all into the development of the product was massive, steep learning curve for us.

27:20And then it had to be translated into the product, how it understood it. So it took us almost two years to reach level where the industry thought this qualifies. Now you qualify to get into, we start taking your data seriously. And then since then, it's been a constant journey. And now we're a little over seven and a half years old. And I would say in year four, we reached where we got the confidence of the market. And now we are strong enough that we can be trusted by policymakers and others as a credible source. And we often reported as an important source of data in very well-known publications.

28:08Ambarish Mitra: At what point in that four-year process, were you kind of trying to also validate the workflow and also the pricing and their willingness to pay? yeah so initially when we solved the problem we didn't see the willingness to pay because they were comparing and we were also now when i look back we were wrong about it because we already see that when you do full installation when you tell the full story like from the beginning of every fiber, every plastic, every metal entering to where it ends. And let's say there are 10 steps in between at every step when we do that level of installation, which we couldn't do in the early days because we were just wanting a chance, just put us over the HDP line or put us over the aluminum.

29:00They won't put us everywhere, you know, because that's a risk for them too. You know, we are a new company and imagine we have to be tested for electrical efficiency and many things because there's a hard weather so they have to do their own risk assessment too you know you can't just walk in and take over a plant you know they have to stop shifts also for you to install them and stopping a shift in a waste plant is a very big ask like and there's a health and safety angle and all those things so we we learned that actually the value prop is partially understood and the roi is not understood well if you just put it in one part it's like saying you're doing analytics on your web page and we do home page only or we do part of the payment page randomly you okay you get a story but you don't get the why fully you get okay people drop off here but you don't know what led to coming here what happens there where did they fail you get a lot of information in one single point so we started really charging better and understanding our fit and once we started doing bigger installations that's that's where the story changed and in the we did a lot of freemium models for quite some time in the early days we were doing a lot of single installations or two to three or three to four only in the u.s market very like i would say the last two years where we've seen massive growth uh where installations have been in quantums of 15 20 at least always like seven eight and above where a much more detailed story comes through and the roi is like on your like obvious it becomes a brainer that they cannot live without it today you know i'm also curious are customers how into

30:54Ambarish Mitra: when they talk about what they like what they love and how great it is are they talking about like the technology are people talking about this great ai tool or are they talking really a bit more about just like it works it does the thing i need to do i think that in a way that's good thing because i think we can get carried away i like that about the waste industry the a bit straightforward people, simple. They're actually quite numerically savvy because they know their numbers. So you have to meet and beat it, basically. It's counting oriented and they do a lot of weight estimation, which is one of the key things.

31:30So it's like, and you remember that the story of AI is also changing in this very period. So imagine when Great Pirate is entering this world in a big way, 70 there is no like uh open ai and anthropic and the big names of the world right so the ai literacy in general is not there but there's a lot of okay some level of skepticism so it's not like they're not personal users of ai you know who think oh my god i know this works in my life so it'll improve things so we had to sort of climb out of not only in general remove that friction but i had to do a lot of customer education where today the customer is self-educated and they think oh i've heard of great use cases everywhere and if this works it'll be great there's a certain optimism towards it but in because by the time typical ai had become a bigger movement we were already quite settled but in our early days they had certain kpis if we met them it's working, you know, because this industry works with machines and they just saw us just a smarter machine and which helped them analyze count better accurately without human intervention, but they can only check it against a human, human counting, gray parrot counting, gray parrot never stops counting and gets it right, basically, you know, and that's how we were tested.

32:58Ambarish Mitra: And also another kind of subject shift a little bit but like you had you had been working on visual search which is you know that's software and yeah and then gray parrot basically you have to have something you know in the in in the in the facility that lets yeah it's hardware it's a you're making something that that has to go someplace it has to be resilient a very tolerant hardware well usually hardwares are so uh in beautiful places they're air-conditioned yeah well it's something you said in a prep call it's like it's like it's like the opposite of an apple store as far as like you know it's it's filthy.

33:31Ambarish Mitra: How did you and the team transition from being comfortable, like doing visual search and software to being a hardware company that had to perform? I mean, a lot of it, initially, everybody who built hardware in our business had never built hardware. So our first two generations were actually genuine startup culture of 3D printing parts in our office. And also COVID hit us so working remotely but it also introduced some level of skills within the business that was interesting about covid because we it forced us to build things portable because we had to ship things to our customer there's no face time or meeting them so it forced that it needs to be light we had put some incredible sort of now when i think about why did we do it i think it was a bit driven by it has to be portable it needs to be light to be shipped and it needs to be installable by one person but maximum two it can't it can't be heavy that it requires an army to do it and it needs to be self-installation self-understood and needs to be plug and play and everything else we automatically the machine takes over we put these product slash engineering difficult boundaries around it which forced us to innovate and and innovate survive you know because had we not done it uh if our thing was like 60 kilos like shipping cost alone would kill us and no one would install it on the other end and so and then if they had to figure five things out on top of it that's you are introducing more frictions to the industry you're trying to get over the line the industry is not themselves motivated to get over the line so major product innovation had to happen but also a lot of the because we learned to build hardware so when we actually hired hardware people we came from a position of knowledge that we can do this you know because and of course our product was not usable from deployment at scale, but it was usable for pilots.

35:41You know, we got images to build the libraries, to build the models, which is the first two, three years work. And then we said, OK, I think the market's going to buy into it. Let's create CE certified or UL certified proper because these are electronic products. So create proper products. And now we do the whole thing is done in-house by us. Yeah.

36:05Ambarish Mitra: What percentage of the customers who were on board for pilots or proof of concept actually converted and became like long term paying customers? I would say 85 to 90 percent are customers still today. Yeah. I mean, something else I learned from our conversation is that. I mean, I would say there'll always be few who would want it at that free price point because they helped you start it. At some point, you have to, we honored that for some time. And at some point, we had to walk away. Wait, that's the 15 %? They just wanted like. Yeah, yeah, yeah. It's not like no one said we don't see the value.

36:41But there were some, let's just say there were like commercial breakdowns rather than anything. High subjections. Interesting. Yeah.

36:50Ambarish Mitra: Massive global industry. but there's not that many players like crowding the field to try to like, you know, own a huge slice of it, which means it's a kind of a tight vendor ecosystem. How did you, how did you, how do you break into that? Plant visits, conferences, just cold calls? What was the? So the industry is very tight, in a way, tight in the sense that I would say like there are under a thousand vendors globally though globally the industry is big it's like almost 1.5 trillion dollars uh but it's a bit country dependent and bit continent dependent too and a lot of the larger players have grown by acquisition so there are like several i would say like at least uh 10 are double digit multi-billion dollar entities you know and then there are a few in 100 millions and and a lot of mom and pops across the world.

37:46And I would say the market size would be like 5 ,000, 6 ,000 plants globally, but growing because it's a big need because consumerism is growing every day. So, but these conferences, which have now become bigger in our journey already, they were even way smaller before. You get to meet everybody, you know, the big guys, the small guys. And in our case, because we were category creator, we literally created this category. We didn't call us an AI. We call it waste intelligence. So it's implied that there's AI in it. So waste intelligence. And we were really quite a big news for quite some time because of the uniqueness of it.

38:31So it goes without saying that if your product is unique and it's authentically disruption, it's not like I use the word disrupt here, not instead of innovate, because it has to be substantially different to what the market's seen. And they said, OK, wow. So we were a major disruptor. We got that attention. Now, then you have to prove that disruption. That's what takes some time because are we the hype or this is a real deal? And then we became the real deal. That's where a lot of, in that valley, a lot of people fall because it's easy to do a demo video on YouTube and get people excited. But another thing is to install it.

39:2030 machines which you have to make, manufacture, send, install. They all need to work. and and if any of them become in any way one percent hazardous no one will take you after that you have to be solid about it so these are 20 30 million dollar plans so we had to really getting to meet the people was good but i think the industry has its some barriers to entry because of the closed nature of it. Meeting people is easy, but to develop products and people allowing you in and then reallowing you in when your product is ready, there's like seven, eight steps to that. There's a high barrier to entry over there.

40:11Ambarish Mitra: How long between day zero and actually serving your first customer? in in our timeline i would say serving in the sense of just proving that something works to the customer yes uh was definitely around 10 months first point yeah okay and so how long between from and from that point to series a how much time elapsed from that point i would say so we we initially started with a pre-seed within the first three months of the business we got and then we another 10 months later we raised a seed and i would i'm just trying to remember our series a was 2023 so around another two and a half years yeah to three years almost yeah and i would say that's like slightly slower than usual in my experience having done multiple companies just because the customer adoption cycle was very slow very slow what had to become true before great parrot was ready to raise the series one thing i would add i didn't mention sorry to interrupt your question it's important so there are many industries if this was algorithmic trading or e-commerce you could check out your funnel of any optimization sometimes in seven days, 20 days, 30 days.

41:34The customer set here wants minimum six months data to make a call before they think, oh, you qualified this stage, now do the next one. So every six, if two six months cycle is one year, go on and just test. So even that doesn't assure you you passed it.

41:49Ambarish Mitra: What was that change specifically? But like, we're ready to talk to investors now for series A, like what had you proven about the technology or what had you proven about customer demand or that sort of thing? So there are three things we were proving. First of all, we, I mean, all our new investors then didn't need to know our background in detail, but they already knew. So we could show them that we are solving an infrastructure problem in a very cost-efficient way. So we process a lot of images, you know, just so you know, we process as many images as a major image social network per day, same amount at this point like we could be like doing around three three and a half billion images a day okay and and we do that at one ten thousand one twenty eight thousand cost of a social network so from an engineering delivery point of view was an attraction to tech we see already that okay whatever the use case this is a good good there's that and why why we innovated that way because we were in a waste industry which doesn't pay you for performance they're dealing with trash it's not ads that if i have my if my my serving cost is two dollars i'm going to charge someone eight dollars no i have to charge flat fee that means i have to always innovate to reduce my cost it forced us to figure stuff out there you know why we do it and how much of the stuff is done on edge versus cloud the whole trade-off the the second point we're communicating do we have traction with early customers who are willing to vouch for us you know it doesn't have to show numbers you know like or you don't have to say like x-ray video of course those presumptions were made in the business case etc but they were able to voice and talk about it and we needed one big player minimum big means well known globally well known so we had that under our belt that someone was really i think that made a very big difference.

43:54And number three, that can we, in our case, we were challenged on that. Can we scale manufacturing? Should this go big? Like, you know, because this is not alone a software story, though everything is done on device and cloud on the software level, but the hardware is the enabler. Without hardware, we cannot process it. So I'm also curious.

44:16Ambarish Mitra: I mean, I don't think. Many people said no to us because of hardware too. Interesting. Yeah, because they're software play only. Yeah, right, right, right. I mean, and a lot of vendors, a lot of VCs, I can't imagine they're any more expert in waste management than you were before you started this. So they're coming at this cold as well. But did you have to like, did they fully understand the TAM? Or did that seem to... No, no one, I think there was a major knowledge gap in general. Like, you know, people don't realize waste is everything, you know, and also understanding not only that, that the waste's biggest protagonist is plastics and plastics is deeply connected to crude oil.

45:02Like there's deeper connections to the industry, heavy industries in general, which is like the global economy, because the recyclable material doesn't have to become bottles again. And they can literally fuel into anything else which is being made out of these polycarbonates. So both sides, and of course, we were positioning it also as a large data play, but we didn't have much data at that point. But the vision was there that there's two things. One, we are optimizing waste. But secondly, we have the consumption pulse of nations and cities. and we haven't figured out how to use that optimally, but it must be of value.

45:49Ambarish Mitra: What did the Series A actually fund? Was it like product development? Was it go-to-market deployments? What did you use it for? I would say Series A, in our case, funded a majority of the technology. Like we had proven everything. Now let's go and manufacture it in a certified way. and number uh two is uh the clients you have got like uh make them into permanent accounts with annual recurring revenue and make that into a deeper commercial pipeline you know and i think we ended up doing a little more we because we we had 21 countries today so we also were geographically focusing on more than the job england where we were located and even before we did our series B.

46:41We were in the U.S. market for the last two years.

46:45Ambarish Mitra: What's the one thing you think, I guess, maybe a lot of founders misunderstand before they set out to start on this journey and build applied AI for the real world? First of all, I mean, I can give a good convoluted answer because I think about this. First of all, we sometimes get very carried away by too much of a big picture, which is great. and everybody wants you to answer that because they want to know their time and they want you to be excited. But there is hundreds of billions of dollars hidden in the physical economy at every step of a process. You know, just like if I'm, my technology is focused on waste sorting, you know, in waste collection, it's itself, there could be like a billion dollar company doing that stuff if you should.

47:35And if you look around in the room and just see how everything's made, you would almost realize like 80, 85 % of world GDP has zero AI involvement to bring it to you. Zero. We can talk about it all day, but it is literally a pixel-based innovation and hardly atoms have been touched. And everything you use, eat, consume, experience around you, as far as you can see right now, now is atoms and it's hundreds of billions of dollars of capex is built into that and they're not going to overnight change it into the next big thing you know so there's so much scope of optimization for founders just don't think like you know there's just one way which is generative there's every and you don't have to be another like you don't have to build a god brain for the physical space either.

48:34Just focus because intelligence is hidden in people's experience in closed door sub economies, you know, that's not accessible on the internet, build relationships, go in, use technology, learn, and you'll build incredible companies out of it.

48:50Ambarish Mitra: One last question, actually. And this is, it's not really connected to the chat we just had, but metaglasses. And I think it's because I live in Silicon Valley, I'm seeing like they're being really pushed heavily here, certainly. And to me, they seem like a product in search of a solution or a product in search of a problem in a market because there's no demand for glasses that record everything. And it just seems like a big data collection grab on the path of meta and not really a service thing that would help. What's your take on this product, this technology? So I've been involved in building glass-based stuff in my previous...

49:24So we were one of the first companies, my previous company, which made actual the original Google Glass into a visual search glasses. And I was also involved with, indirectly with Snaps glasses and acquisitions and stuff it had done. So in my last life, I was very embedded in stuff happening on cameras and glasses generally. And my view was, again, the scope and use of it in industrial space is so much more and so relevant, actually, in terms of feedback and learning and safety and security and what's in the normal curve, what's not, when things become amber, green, red, you know, like you're in the danger zone.

50:10And there's the amount of work that humans do that requires two hands and computing at the same time because you need some level of feedback loop. So I feel like if it came from that path into some multi-user consumer journey, it would have made more sense i think just capturing photos videos getting map based feedback i'm also a believer that whatever the predecessor technology and in this case it's the phone unless the next interface significantly is better you will have pushback and today no glass has achieved a better user experience and interface experience and zero latency see all of that yet in here.

50:58So I think the product is still doing halfway solutions until it's ready and having its iPhone moment till that point, we won't have adoption.

51:10Ambarish Mitra: All right. I'm glad to know I'm on you and I, I'm in agreement with a visionary in the industry. So I guess I'm on the right track. This is a good thing. Very reassuring. Rish, this has been such a fantastic conversation. Yeah. Thank you for the time today. I really appreciate it. Thank you, Walter, for the good chat. The Fund Build Scale theme was written by Michael Tritter and performed by Michael and Carlos Chárez. Michael also composed Ocean, the track he heard at the top. If you enjoyed the episode, please subscribe and leave a rating. It helps more people find the show. You can get the Fund Build Scale newsletter on Beehive, follow along on Instagram, and if you have a guest suggestion or a question, reach me at fundbuildscale at gmail.com.

51:51Ambarish Mitra: Thanks for listening. Fun Build Scale is a production of Truth and Soul Media, LLC.