Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Making Manufacturing Smart | Vipin Raghavan @ Haber
420 Plays2 years ago
Even though India has been slow when it comes to modernizing machines, Haber might change that. With pathbreaking AI solutions, the Robot-as-a-Service startup is set to revolutionize the manufacturing industry. Listen on as Vipin Raghavan speaks about the compelling value proposition of Haber!
Know about:-
- Product understanding and features
- Fuzzy algorithm
- Variable frequency drive
- Future plans- solving for sensors
Recommended
Transcript
Introduction to Founder Thesis Podcast
00:00:00
Speaker
Hi, this is Vipin Raghwan. I'm the founder and CEO at Haber. Hi, I'm Akshay. Hi, this is Saurabh. And you are listening to the Founder Thesis Podcast. We meet some of the most celebrated sort of founders in the country. And we want to learn how to build a unicorn.
Challenges of Modernizing Manufacturing
00:00:33
Speaker
It is much harder to build atoms than to build pixels. Or in other words, manufacturing physical goods is harder than making software. We've been running factories for over a century because of which it is seen as a slow-moving space. But the advent of new age technologies like IoT devices, sensors and machine learning has the potential to disrupt traditional manufacturing practices
00:00:55
Speaker
and make factories more efficient, safe and environment-friendly.
Vipin Raghavan's Entrepreneurial Journey
00:00:59
Speaker
This is exactly the insight that led Vipin Raghavan to start Heber. Vipin is an unlikely founder. He worked in the finance function for most of his career, but got bitten by the entrepreneurial bug when he joined Zynga, the gaming company. In his last job, he was visiting large plants across the country where he saw first-hand the immense need for modernisation in Indian factories. And the rest, as they say, is history.
00:01:24
Speaker
Here's Vipin telling Akshay that about the journey of building Haber.
Founding of Haber by Ex-Zynga Team
00:01:28
Speaker
It was a group of us from Zynga that started Haber. Even though in between, I had this other company, EcoLab. So Priya, who's now our chief operating officer, and Prashoon, who was our chief technology officer. So three of us knew each other from Zynga. Yeah, the two were also in Zynga, India.
00:01:46
Speaker
Yeah, all three of us were in Zynga. So Zynga is kind of where a lot of the foundation was set. So we used to have these conversations over coffee or beers about interesting product ideas. How do we disrupt this or that? So everything from online grocery stores, to robots, to new gaming companies, to whatnot.
00:02:07
Speaker
But me, Priya also wanted to do something that had a positive impact on society, not just, okay, we can build this very large company by disrupting this market. There was one piece of it, but we also wanted to do that in a space where we were having some kind of positive impact. So that was kind of the common thread between the three of us, not to say that we had halos around our heads or anything like that, but we wanted to do something meaningful.
00:02:33
Speaker
So that's how the team came together.
Disrupting Indian Heavy Industries
00:02:35
Speaker
So what we wanted out of our startup was something similar. And then going back to Ecolab, I went to one of these very large, well-known steel plants in India, and they would have one fatality every year. And they were proud of that. They were like, we're very good at safety. We only have one fatality every year. I'm like, are you actually kidding me? One person dies because they came to work for you every year and you're okay with that. And then this is something that having better technology could
00:03:03
Speaker
solve. And it was not that they didn't want to do it. Those solutions were not even available. So everything from safety to environment to improving productivity, doing more with less, starting out with the same amount of raw materials and producing more finished product or doing the same operation with less labor or machines not breaking down, reducing your maintenance costs. So there's tons of just immense opportunity. Even you need 10 acres to build this plant. No, you don't. You can be more efficient. You can build it on an acre. So
00:03:31
Speaker
So any one of those spaces, you could touch and you can disrupt. Essentially, you wanted to bring Silicon Valley style disruption to heavy industry sector in India.
Impact of Small Changes in Factories
00:03:42
Speaker
Yes, and going and visiting these plants, which I think the reason that you don't see more startups in the space is because of that. You don't get to go to these places. They won't even let you in. You have to be there for a reason. You can show up at the gates. You can't get past the gate. To get a gate past these places itself is hard. Plus they're in the middle of nowhere. It's not like, okay, you're on your evening stroll and you're going to walk past a sea plant. That's never going to happen.
00:04:06
Speaker
So you have to deliberately go there and then somebody has to let you in and post that these are these massive complexes. If somebody doesn't exactly show you around here, you're still not going to know where to go, what to look at. So it's huge barrier to entry for that reason.
00:04:21
Speaker
this opportunity and we were all very excited about specifically having an impact on energy and water use. So we quickly discovered that these factories use more energy than all the cars on the road or all the homes use more water than all the humans put together. So this is where most of our resources are going and not in cars and in homes and whatnot. And so if you could even make a small change here, you're going to make
00:04:51
Speaker
massive impact at a macro scale.
Haber's Scalable Solutions
00:04:55
Speaker
So like, what did you, what to do? Like, did you first get like a pilot customer and then figure out with them what you want to build? We were like, okay, whatever we did had to have scale. And we saw this problem of.
00:05:09
Speaker
energy and water use. So there was one problem which was around safety. So remember, we also wanted to do something with a positive impact, not just helping these factories make more money. So we're like, okay, we can help their employees be in a safer environment, or we can help these companies use less water, use less energy, take out
00:05:29
Speaker
or CO2 from the atmosphere do that kind of thing. And we decided to narrow down on that space, the saving of energy and water, because that also had a P&L impact because it was costing them. If we saved a guy from losing one of his fingers, maybe that didn't translate to so much value for the company.
00:05:46
Speaker
So safety is another space that should get disrupted in my opinion. There should be a way to translate that into commercial value. So we didn't go down the safety route. We went down the route of saving water and energy because it was also attractive to these companies. Not so much to the Indian companies. So we went from, okay, this is the problem we want to solve. The product that we built to solve this problem should have scale, right? We don't want to build custom products. We want to build one product which can be plugged into as many places as possible.
Development of Alexa ELIXA
00:06:15
Speaker
So we started studying their processes. What we saw was there was this area where the way they control different aspects of the plant, they would collect like a sample of their slurry in a beaker, walk it over to an outside lab, do a bunch of measurements with benchtop instruments. And then a human subject matter expert would look at this data and then they would determine how to tweak their process to take corrective action because from what other readings they got.
00:06:40
Speaker
So we said, can we build something to automate this loop and by more efficient control of the process, help these factories consume less power, consume less water. So we built a product called Alexa ELIXA and it stands for Electronic Live Information Exchange and Analysis.
00:06:59
Speaker
So what it basically does is the whole sampling process, it is automated, basically think of it as a small robot doing the action of collecting and measuring. And then that data gets applied to an algorithm, which is inside of our product. And then the algorithm takes decisions as to whether to switch on and off a valve or turn a pump on and off or to make a pump go faster or slower. So that's a loop that we have automated with Alexa our flagship product.
00:07:26
Speaker
Like help me like understand what is, who is this manufacturer? What is this slurry thing? Like, like this, that is like, what is the problem that they're facing with an excellent software? So like think of you ordering coffee from Starbucks or Caribou or wherever you get your coffee in, in, in the paper.
00:07:46
Speaker
It's called cup stock. Now that is made from wood. So they don't go in chopping down forests. So there's wood grown as an agricultural crop. So there are farmers who grow wood and then that wood gets converted into pulp. You get this brownish kind of wood and it gets turned into kind of whitish kind of pulp. And then that pulp gets converted into paper stock. That then becomes the cups. And then those are the cups that we get at Starbucks.
00:08:13
Speaker
Now in that process, the pulp gets carried through different stages of the manufacturing process before it finally becomes the cup stock. So what we are sampling is that pulp. So pulp is nothing but 99% water and 1% wood fiber.
00:08:29
Speaker
So that's what we are sampling at different stages of the manufacturing process and measuring for, think of it as converting brown pulp to white pulp and then a white pulp, which is not as strong to a bit more stronger kind of white pulp. So each stage has a purpose.
00:08:47
Speaker
And what Alexa is doing is in the conversion of this brown pulp to white pulp, it's sampling both of these and then saying, okay, how do I consume as little resources in that process? So in terms of how much steam gets added, how much chemicals get added, can that be done with less steam, less chemicals and still get the same whitish pulp? So that's why our customers are subscribing to Alexa because they get
00:09:12
Speaker
savings in the form of a steam equals energy savings. Or they're consuming less chemicals in the process. So the traditional approach was manual tweaking, like someone would come do a test. Yeah. Someone would grab a sample with the brown pulp and the white pulp maybe once every hour. So they're not getting continuous data. They don't know what's happening in between. They just assume that there's no change in between. So they collect the sample
00:09:37
Speaker
They walk it over and these are usually massive complexes. So they walk it over to an onsite lab that takes time. The sample collection takes time. Taking it to the lab and measuring takes time. The analysis takes time and it's dependent on human subject matter experts who are extremely rare. These are not industries where we are sending a lot of talent to anymore. So the talent that they have is extremely rare as well. So the people who can look at this data and make
AI in Factory Automation
00:10:02
Speaker
sense of it.
00:10:02
Speaker
and then determine if they had to tweak something, add more steam or reduce the amount of steam or whatever the corrective action they had to take. So as you can imagine, no continuous data, any kind of analysis is delayed. By the time they figure out that they need to take corrective action, they've missed the opportunity to take corrective action at the right time.
00:10:20
Speaker
they end up consuming more steam or they end up making pulp that is not as white. So they have to throw it away and make new pulp because Starbucks is not going to accept because you want your cups to look exactly the same every time because they got to print the green Starbucks logos on them. So the cup is kind of brownish and you print the green logo. It's not going to look like the green logo anymore.
00:10:39
Speaker
So for all those reasons, they need consistent quality manufacturing as well, and also controlling the resources that are getting consumed in the process. So that's one example of where we plug in an LXL. What industries would be covered by this? So essentially, these are an industry which has a process in which there is water involved, like the top of paper.
00:11:02
Speaker
Yeah, not necessarily. So we focus on what we call process manufacturing, right? So these are continuous 24-7, 365 day plans as opposed to discrete manufacturing. So if you think about putting a car together, that's discrete manufacturing, right? So these are process industries, which you have raw material kind of continuously flowing through a process and the finished good coming over the other side, which is also continuous and not a unit like a car. It's a roll of steel or it's a roll of paper or whatever it is, right?
00:11:31
Speaker
So those are the industries. So anything metals, steel, alumina, things like that, pulp and packaging. So think about everything from a tissue to the boxes that you're getting your Amazon deliveries to your Starbucks cups, that industry.
00:11:46
Speaker
food and beverage processing. So sugar, breweries, dairy, all of those. So those are the industries that we are currently focused on. So oil and gas would be another space where we could apply elixir. We're not looking at oil and gas. How do you get the data real time? Like this would essentially need you to build a device which can test real time.
00:12:09
Speaker
Yeah, so just like our phone has sensors in it, so Elixir has got its own sensors. So the sample is flowing through like a flow cell with a few sensors. So this is why I was calling it a Fitbit for factory. So it's measuring these critical parameters in real time. And then the sample just goes back into the processor. So it's continuously sampling and there are sensors which are measuring critical parameters and that's the data that's getting applied to the algorithm.
00:12:34
Speaker
So you need to build it separately for each industry, but what are those sensors doing? Help me understand the tech behind it. Would a set of sensors that work in food and beverage also work in pulp? Also work in oil and gas? Also work in metals?
00:12:50
Speaker
For the most part, yes, because the parameters that we're interested in are kind of similar, right? Obviously they're very different industries, but temperature is important. pH is basically something that tells you whether it's acidic or basic. pH is extremely important. And then other things called conductivity, alkalinity. So these are the same. So you're still concerned about the same things. And then there's like an
00:13:14
Speaker
optical sensor that's basically shooting out light at different wavelengths and it studies the light bounces off the particles and comes back and then we can study different things. So for the most part, 90% the sensors are the same. There are some unique sensors required for some unique applications, but for the most part it's the same.
AI's Broader Applications
00:13:34
Speaker
So temperature, pH, optical, these are like the three mainstream sensors, which are like 90% of the cases covered by these. Yeah. And then there is flexibility. So when I say pH, it's like an electrochemical sensor. It tells you multiple things. It tells you pH, ORP conductivity, many things. And then the optical sensor also. So a lot of things that you measure are through optics. You're just shooting out light at different wavelengths and like the color is appropriate or not.
00:13:59
Speaker
a lot more than just color. So even the tests that you do for coronavirus and those are also optics. You're adding a reagent and then looking at how much sort of the color changes and that's what you're getting. So optical sensors, how most of the world works.
00:14:16
Speaker
Just to get a little more nerdy, like give me through an example that it could be like say a juice or a pulp or whatever that through the sensor we can figure out if this is going wrong or not and all through the optical sensor like a little more.
00:14:33
Speaker
Yeah, sure. Go back to that example of making the pulp for the cups. So you're starting out with wood, right? So there's variability in that. It's a natural resource, right? So inherently you have some variability. So you're starting with some kind of variation, right? And then you're adding
00:14:52
Speaker
heat, you're adding pH shock and that's how you get the tannins and things like that out and you get what you want, which is whitish kind of pulp. So pH shock is worth like through some chemicals. Oh yeah, you are giving extreme pH. So you're taking it down to very acidic pH and then taking it up to very high pH. So when you give it pH shock, the things that are
00:15:21
Speaker
kind of not so well chemically bonded, those things tend to kind of loosen out, right? So, so if you think about soap, soap has a little bit of pH shock. So you get the dirt out, right? You have the forming and then you have a little bit of pH shock, which is what gets your dirt out. So pH shock kind of set basically separate. It doesn't.
00:15:39
Speaker
It can destroy things as well, but you can get a separate thing. So in the case of wood, you have cellulose and you have tannins and things like that. So you want to remove these tannins and you want to retain just the cellulose. So the fiber is what gives it the strength. So if you have tissue, you get the strength. So this is just the fiber, which is cellulose and hemicellulose.
00:16:00
Speaker
without getting into too much of the chemistry. So basically the way you're doing it is by adding heat and then moving the pH up and down and then some other things as well. And then what comes out is the pulp that you desire that you want to make the cups with. So because it's starting out with wood chips that are introducing variability, right? So you have some level of variation always happening in the process, right? Yeah. Yeah. The input is not standardized because it's a natural product. Yep. Yep.
00:16:29
Speaker
And you're starting out with wood chips, and also you're adding water to it. The water also comes with some amount of variation, right? So wood plus water becomes your pulp, and then you're increasing, decreasing the temperatures, so on and so forth. So because you're starting with this rational resource, you inherently have some variation. What Alexa is doing is predicting...
00:16:47
Speaker
what the part that's going to come out the other side is, how wide is it going to be? How strong is it going to be? How are cellulose? It's going to have all that desirable characters and then figuring out, okay, how do I, if something is expected to go out, let's say an excess expecting brightness to go down, right below a certain threshold.
00:17:06
Speaker
let's say there's a minimum threshold of brightness and you don't want it to go below that and it's predicting that an hour from now that's going to happen. So what Alexia is doing is predicting something in the future and then it's taking an action right now, a corrective action right now to prevent that bad thing that might happen in the future.
00:17:21
Speaker
So we're giving it the controls to move the levers. So it's then increasing the steam or reducing the acids or whatever it has to do to then meet that. So that's one thing that we ask you to do. The second thing we're asking you to do is get me consistent brightness by using as little steam and as little chemicals as possible. So it's giving you consistent brightness, but it's also doing it with the least amount of resources. Essentially, it's highly fine-tuned.
00:17:50
Speaker
With very simplified fashion, yeah, that's what it's doing. You think of it as like an autonomous driving car, but it's some sort of predicting what's going to happen with traffic or by looking at things in the present and what it's seen in the past. So it's going to be like, okay, down the road, I'm expecting this guy to cross the street. So I'm not, I'm going to start decelerating right now so that I don't have to slam on the brakes when I get there. So that, that type of thing.
00:18:13
Speaker
So how did you learn what these numbers would mean? So you had a device which was throwing out numbers based on these sensors, like some temperature numbers, some pH numbers, some numbers which would come from the optical sensor. How did you figure out what these numbers would mean?
00:18:29
Speaker
they would predict. How did you build the intelligence? Sure. So also have this AI background. So going back to my engineering, one of the electives that I took in the final year was AI and neural networks. Back in those days, there was no application for it. It was mostly just theory.
00:18:47
Speaker
So I was really intrigued with how AI works and fast forward to my MBA. The reason I got into that startup was also I participated in this business plan competition and my business plan was called mdnetwork.com. The idea was a neural network that physicians would subscribe to
00:19:08
Speaker
And they would have their patients come in and put in all their symptoms. And this neural network gets smarter and smarter over time and then becomes smarter than physicians and can then predict very accurately the disease state with the symptoms fed in. So that was the business idea. And John Papajar, not the pizza guy, but the VC guy, offered to write me a check if I drop out and actually start this business. And I didn't know that. And I wasn't sure what he was even talking about. I was like, what the hell are you talking about? I'm here to get my MBA. That was before dropping out of school.
00:19:38
Speaker
Yeah, exactly. And I didn't have in my family, in my friend's network, I didn't have the right kind of support or people who guide me. I didn't know what the hell this guy was talking about. And I said, I'll think about it. But I said, I was just there for the business plan competition. So I won the business plan competition and got like this opportunity to go work at a startup.
00:19:58
Speaker
So that was also an AI-based, even before there was applications for AI.
Virtual Sensors and Process Optimization
00:20:03
Speaker
So the idea was an AI-based medical diagnosis tool. So in addition to having this entrepreneurial hitch, I also had this AI hitch. So fast forward to Haber. So I was looking for AI to solve the problem of not having sufficient human experts available.
00:20:22
Speaker
and also the speed at which human experts can perform. So that was the other problem that we were solving. So the way Helixa also works is you don't need to necessarily measure everything. So if I know, for example, if I know the angle of sunlight, I can probably tell you with a high degree of accuracy what the temperature is without having to measure temperature directly. So we figured out virtual sensors, right? So there are a lot of things that
00:20:51
Speaker
Alexa can predict which it's not directly measuring with a high degree of accuracy. So the AI is the secret sauce behind the product. We did solve a bunch of other things, but kind of making that human subject matter expert obsolete is the biggest value that we are adding. And although I don't want to use the word obsolete, let's say augment, making that human subject matter expert available
00:21:15
Speaker
to every factory all the time, which in the past, even today this happens because we haven't taken over the market. We still have a very small piece of the market. People would just fly in. So you would have like an expert in Japan and there's only five of those experts globally. So this guy would take a flight from Tokyo to Delhi and
00:21:35
Speaker
literally parachute into a plant and look at this data and solve it. And why can't they do it remotely? It's very hard for a human to just look at data on a spreadsheet and they would want to come visually see it, not just through video. They would want to get a feel for it. And this is how they solve some of this. Yeah. It's an investigative process. Exactly. These people are extremely rare and they're getting rarer by the day. So unless
00:22:01
Speaker
industry puts a lot of these things in software, they're going to have a hard time even existing a few years from now. So you would actually be getting data from multiple stages in the whole process and also you would need data at the end that this is meeting desired standards because that is how the AI will get trained if it sees input data and then it sees output data.
00:22:27
Speaker
So now we are plugging elixirs into small sub-processes within the plant. We haven't put an elixir that controls the entire plant into it. So if you think about plants, it's just sub-plants within a plant. There's multiple sequences of processes, some in series, some in parallel. So we plug in elixir into those discrete processes and it's just figuring out what does this process meant for? What is it doing? There's a goal for the process.
00:22:56
Speaker
So we're giving Alexa some goals. The way Alexa sees it is it's a bunch of time series data points because everything is just moving in time, some of which it can control, a lot of which it cannot control, and some of which are desired parameters. Like for example, I want brightness to be as stable as possible. And then it's just predicting brightness. It knows what brightness is right now. It doesn't know what brightness is in the future is predicting it.
00:23:24
Speaker
and then saying, okay, if brightness is going to drop below the sudden threshold or it's going to be too bright, how do I get it back within that range by these three levers that I have right now? And I only have these three levers. So how do I tweak these three levers so that I get that? So it's just working in that sub process and so on and so forth.
00:23:42
Speaker
So the next level of this stuff is then a little project we are working off to the side, which is kind of building this digital twin of the entire factory and having Alexis talk to each other and telling each other what to do. Right now, we're just using data from different Alexis to kind of strengthen the algorithm. They're sharing data. They're not talking to each other and telling each other what to do. So that's like the next stage of it.
00:24:06
Speaker
And that will enable a lot more. And now the thing about this, let's say this manufacturing plant has five stages, right? So we are individually optimizing each of these five stages. Now imagine if all five annexes are talking to each other, you could take it to the next level. Maybe you don't need to make it as bright in stage one. You can make up for it in stage two, or you can make up for it in stage five. And then we can figure out the most optimized way of doing it across as opposed to like in individual stages. Okay.
00:24:35
Speaker
So right now, like a stage one, Alexa and a stage five Alexa, then data would build the algorithm because you can see that in stage one, if temperature is off by five degrees, then stage five brightness goes down. I'm highly simplifying it, but
00:24:52
Speaker
Absolutely. We are doing some of this offline because we have Elixis live in certain plans where the sequences are in parallel. We are doing this offline, meaning we have our subject matter experts looking at this and being like, oh, here's an insight.
Evolution of Alexa ELIXA Hardware
00:25:08
Speaker
And then taking that back to the customer and tweaking the goals.
00:25:11
Speaker
Now instead of taking it to 60 brightness in stage one, let's only take it to 55 and then adjust it in stage three, that type of thing. But that's happening offline. It's not happening in the true AI sense. So that is something that we are currently working on because again, you got to be able to scale these things. How do you build a digital trend of any factory in a way that anything is, things are configurable. These mathematical operators, AI operators can kind of work together in harmony.
00:25:39
Speaker
It's a complicated problem where we are solving it and we have a great team to solve it we have what should take the product to that level is gonna be amazing amazingly effective right. I'm guessing that the evolution must have been like this the initial product would have been one in which those sensors.
00:25:56
Speaker
are giving real-time data and there are some hard-coded rules that in this stage temperature should be within this range pH should be in this range or whatever tests are coming out there would have been some hard-coded information that okay these are the ranges to maintain if there is deviation then
00:26:14
Speaker
press this lever or do this action. So yeah, at first our first POC was a combination of a subject matter algorithm, subject matter expert based algorithm plus fuzzy. So, and then it would test which was better, was fuzzy better or
00:26:32
Speaker
The SME algorithm was better and we would just use that. So there was like this decision tree. How did you build fuzzy right in the beginning? Because you probably wouldn't have had enough data to build fuzzy. What does that mean, fuzzy? We built a generic fuzzy, right? It wasn't like specific to the problem. We just had a generic fuzzy algorithm.
00:26:55
Speaker
What does it mean? What is a fuzzy algorithm? Fuzzy means, okay, I have this lever. Let me start by going to the extremes of this lever as high as possible, as low as possible, see the impact of it, and then kind of strengthen the algorithm over time and not necessarily like a time series AI model that will understand seasonality and times of the day and different things like that. So fuzzy is more.
00:27:19
Speaker
Think of it as running random experiments and then just strengthening that and figuring out, okay, between these parameters, I have these kinds of strengths. So that was how we started out, coupled with the subject matrix word algorithm, and then kind of a decision tree, which would say, okay, which one can predict better? And it would just use that to control, right?
00:27:39
Speaker
So your customer allowed you to run like this? Because we were giving them a much superior product. We were giving them real-time data, which was visible to me. At that time, we hadn't yet built our online dashboards yet. So there's a screen. So Alexa has this device that's about the size of a small refrigerator. So that's how big the device is. It has a screen also.
00:28:04
Speaker
So they could even go at the screen and look at the data. I mean, day one, obviously, even that did not exist. It was just seeing data on a terminal. So then we had the screen. They could see what was happening. That itself was a step above. And secondly, there was some kind of action being taken, right? Now the concern was how good was this action, right? But then this action was controlled by how much of a change could Alexa make. So that was also controlled. You can only move between 40 and 50.
00:28:34
Speaker
you can't take it down to 20 or you can't take it down up to 100. So you can only go between 40 and 50. So that gave the customer enough comfort that, okay, you can't mess things up too bad. So we started out with many tight guard rails that gave customers... We had
00:28:49
Speaker
an early adopter, Padamji, which is a company here in Pune that makes some of these filtration products that then 3M uses. Everything from your masks, your filters that go in cars. They also make hygiene products. So the paper that goes in diapers, tissue paper, those kinds of things.
00:29:06
Speaker
like a fabric company basically. A pulp and paper company. But the end use of these paper products is filtration, hygiene, things like that. So in India pulp mill, where they make the pulp, is where we applied elixir. That was the first use case of elixir. And in parallel, we went live at Imami. Imami is a large consumer company. And Imami is also one of India's largest packaging producers. So
00:29:29
Speaker
They, the boxes that the fair and handsome comes in, they'd make those in-house and then they sell those other packaging for other people to put their products in boxes. So these were two early adopters and they were keen on applying technology in their factories. They gave us our first break and yeah, from there, there was no looking back. So those installations still existed to this date, right?
00:29:51
Speaker
So these installations that go back to 2017, so it's almost been five years. So those installations are still there. We've changed the hardware as with newer generations of hardware, but those installations and what they're doing, still they're doing that, right?
00:30:06
Speaker
Predicting, controlling, saving both Padamji and Imami lots of energy, water, more consistent in the case of Padamji, more consistent quality pulp. In the case of Imami, it is kind of like in their power plant, so their power plant is using less coal. It's a massive savings net, right?
00:30:24
Speaker
Can you talk about the evolution journey of the hardware? Like currently you're saying it's like a refrigerator, a science box. So what did it start with and how did it get to where it is today? What did you learn along the way or what mistakes did you make along? Yes. So fundamentally the hardware, the components have not changed a whole lot, right? There is, it's got like three sections, right? The top section behind the screen is the brain, the processor memory, the IO cards and whatnot.
00:30:50
Speaker
The middle section is kind of odd electrical stuff to talk to different components in the factory because not everything in the factory is a smart device. So taking like an electronic signal and converting that into an electrical signal. So there's like a piece in the middle that does that. And the bottom third is the sensor array. So it's like a little flow cell. Think of it as a cylindrical tube where the slurry is passing through and sensors mounted in there. So
00:31:18
Speaker
Obviously, the iterations are just making it better, more efficient, using less sample, less cleaning cycles, more efficient cleaning cycles. Those are the iterations that we've gone through. But in the top, the brain itself, our latest generation, which is what we call Alexa 4, it's much smarter. It can run heavy AI models on it, things that we were earlier running on the cloud and we would push down pickle files. Now those things can happen in the edge itself.
00:31:46
Speaker
It has more memory, right? So it keeps more data, which also helps with some of the AI functionalities and security. So security is a big piece that we have enhanced quite a bit. So now we have security in the hardware itself, right? So Alexa photos is impenetrable, right? So we're going to run a hackathon as soon as we didn't want to run an online hackathon.
00:32:08
Speaker
So I think now that things are starting to open up here in India, we want to run like a hackathon where we put an Alexa in the middle of the room and anybody who can hack into it gets like a $10,000 price.
Interactions with Factory Systems
00:32:20
Speaker
So security is something that we have significantly improved with the latest generation. And things like, so having strong processor memory, now we can do computer vision. So computer vision was Alexa 3 and prior to that we couldn't do computer vision. So you mean like the optical sensor?
00:32:36
Speaker
And optical sensors is just shooting out light and then measuring different energies. That's very small data. I'm talking about high quality video camera. So taking optics to the next level, basically. So with video data, you can do a lot more.
00:32:53
Speaker
So we haven't rolled out any computer vision based use cases yet, but yes, so that's something that we want to do. And is this like, how do you build this? Like, how did you figure out building this? Did you find suppliers in India? Did you have to go to China and figure it out? Yeah. So, so, so when we were building Elixir, the architecture right back in 2016, there was no edge computing device.
00:33:18
Speaker
There was no edge-combing device off the shelf that one could buy. So we built the entire electronic architecture from scratch. So what is it that we want to do? Select the processor, select the memory. We also did analog to digital conversion, digital to analog conversion. So there's a lot of electronic architecture that went behind it.
00:33:37
Speaker
And then electronics are just devices, little chips, and you've got to make them alive. So voting the operating system, the embedded software layer, all that architecture was built in-house. And the key thing is these are 24-7, 365 day. Our product has to work. It can never break.
00:33:54
Speaker
So that was the hardest problem to solve because what happens is you get memory overload, you get processing overload and all things heater and it breaks. And it used to happen in the beginning. We used to have, we would have to, Alexa one must be like hand assembled kind of a, or close to that.
00:34:11
Speaker
Yeah, so we didn't print the boards. We kind of etched the boards at a local supplier and then we just hand soldered the parts. But very quickly we went to kind of a robotic board building, soldering etching and all that. Because that was another reason why things could fail. Because you can't sold manually, soldering is very hard. And we used to do it. Myself used to sit in our homes and be doing the stuff. And we didn't have the best equipment also.
00:34:39
Speaker
Well, so the electronics got solved in terms of the quality of building the electronics that we fixed very quickly once we went to automated building of boards. The software, kind of the hanging of the software. You can just think going back even a few years, how frequently would your cell phone crash? Even the good ones. So we had to solve for that. Our device could never crash. So that was a very hard problem to solve that we solved. So obviously today, Alex has never crashed. Did you build your own operating system?
00:35:08
Speaker
So it runs on Linux and then we ported Linux and the embedded layer is all in C. So meaning moving the code and managing memory, all that is done with C. We wrote the code in C. Yeah. So like I said, you got to use memory in an efficient fashion and back into it, the things have kind of advanced since then, right? Back then getting 2GB memory or 6GB or whatever was not easy on a board. And it was prohibitively expensive because you got to think about how much the box is going to cost as well.
00:35:38
Speaker
memory and processing have become cheaper over the last few years, which is what is enabling us to do computer vision and things like that. So yeah, solving the embedded layer, I would say more than the AI, that was a really hard engineering problem. Obviously, you don't get a lot of credit for doing that because nobody sees it. Oh, your product now fails. Good, because that's what you expect. You should never fail. So it does what is expected. How does it manipulate the levers? Like you said, that it can control
00:36:07
Speaker
So how does that happen? Yeah, so it's a two stage process, right? So the first stage is predicting, right? So it's predicting what brightness is going to be sometime in the future. Then it is predicting what brightness is going to be sometime in the future.
00:36:23
Speaker
Let's say there are only three parameters. So it's looking at this cube that has three parameters, and then it's finding the right spot inside that cube, which will then solve for the brightness problem. But it's not a nice three-dimensional cube. It's this multi-dimensional thing, which we cannot comprehend. So it's looking at this multi-dimensional space and finding that right intersection of all the parameters. And actually, it sometimes would find
00:36:47
Speaker
maybe two different solutions and then it has to choose which one to go for. So that's what it's essentially doing. So the first stage of the algorithm is predicting what is brightness going to be 90 minutes from now. So time series data is like a movie, right? So every frame has some dependency on the previous frame and what happened previously and so on and so forth.
00:37:07
Speaker
So if you know what has happened up until now, and you've seen these patterns previously, so if you're watching Game of Thrones and you've seen the first season, you can kind of predict what's going to happen in episode four and season two. It's what people do in equity also, like their technical analysis. So based on how that stock price has moved over time, you kind of try and predict. Yeah, their technical analysis is now the models are more in a combination of fundamental and technical, which is where AI can do a better job.
00:37:36
Speaker
So what it's doing is it's watching this movie and it's seen this movie many times before and not just in this theater, it's watched this movie in other theaters. So remember the data is getting shared on the cloud from different audiences. So it's finding common patterns between those.
00:37:51
Speaker
And then it's saying, okay, I've seen all the stuff. I want to predict what's going to happen in the future. It's predicting that. And if that prediction is not a desired outcome, then it's saying, what can I change right now? How do I change the plot right now to get my desired outcome in the future? And that changing the plot is...
00:38:10
Speaker
Let's say some processes might have 10 parameters to as many as 100 parameters, right? So it's looking at this thing in 10 different dimensions, 100 different dimensions and finding the combination. So some things it cannot control, right? So out of those 100 things, let's say it cannot control 88 things. It can only control 12 things. So those 88 things are given to it. And then it's saying, okay, no, these 12 things that I can change, how do I change it? So that the professor doesn't get caught in money heist in the next episode. We are a fan of money heist.
00:38:42
Speaker
If you like to hear stories of founders, then we have tons of great stories from entrepreneurs who have built billion dollar businesses. Just search for the founder thesis podcast on any audio streaming app like Spotify, Ghana, Apple podcasts, and subscribe to the show.
00:39:03
Speaker
But how does it change? Does it need to do some physical manipulation? Well, how does it then talk to the plant? Yeah, so the plant is running on its own systems, right? So there are these things called DCS, Distributed Control Systems or SCADA, Supervisory Control or Data and Data Acquisition Systems.
00:39:21
Speaker
So these are, you might've seen them in movies and like nuclear plants or in NASA control rooms, a bunch of screens and a bunch of people just staring at these screens. So you go to any plant, you're going to find, they don't call it a control room, but it's kind of like a control room. They would call it like a DCS room or a SCADA room.
00:39:37
Speaker
People like to hang out there because it's air conditioned and you'll find these people looking at it. So that's where they control their plans from. So then Elixer is talking back to the DCS, right? And sending. So, so the other thing that we've solved for is this protocol conversion, right? Because different DCS's, SCADA's talk different languages. TCP, IP, Modbus, Profibus, ProfiNet, whatever. So it's translating into whatever protocol that DCS understands the language. And then it's setting those instructions. One change it from.
00:40:07
Speaker
50 liters per hour to 48 liters per hour. That'd be the instruction that it tells the DCS and then the DCS will control the pump. Which previously would have been this guy sitting in a DCS room and entering data into like a little text box or in slightly more primitive plans, literally somebody walking to that pump and running a knob. And even in more advanced plans, you might
00:40:31
Speaker
see things where you have to send somebody to a pump or a valve and they might have to manually control the valve or the pump. But so, Alexa can only work if the plant has a DCS slash kind of this manual kind of plants like
00:40:46
Speaker
Yeah. So that's that middle section of Alexa, right? Let's say the plant is not very advanced. It cannot understand digital. It's not a digital plant or certain parts of the plant that are not digital are still analog. So that's where it's taking the middle section converts the electrical, the digital signal into an electrical signal. So it's mostly what is called a four to 20 milli ampere current loop. So you adjust the current between four and 20 and depending on what the device does, let's say it's a pump that moves between
00:41:15
Speaker
20 liters and 200 liters. So four corresponds to 20 and then 20 corresponds to 200. So it will send like a four to 20 milliampere signal and then basically control the power that's going to the pump and then that then translates to the speed of the pump.
00:41:29
Speaker
So we have customers who are not just enterprise, more stable, more advanced plans. So interestingly, India has some of the most advanced plans, because it has some of the newest plans, the newest investments in the world. And it also has some of the oldest crap, right? Like stuff that is fairly low-tech, built in the 50s.
00:41:50
Speaker
Yeah, maybe not built in the 50s, but some European plant that was going down wanted to sell it and they would go and pick up this plant and ship it over and then rebuild it here. I just want to clarify one quick thing before we talk. So like you said, this 4 to 20 thing. So it's like when I have a mixing at home, then I can do 1, 2, 3, 4 and that the blade moves faster or slower based on what number I'm selecting.
00:42:18
Speaker
So this is the same thing happening directly from Alexa to the pump, like telling it what speed to pump at. Instead of those buttons getting pressed, that is what the 4 to 20 thing is.
00:42:31
Speaker
Yes. So what's happening is there is something called a variable frequency drive. So that converts this electrical signal to the frequency and then the frequency controls how fast this motor is spinning. So in your blade example, when you turn that knob one, two, three, four, it is changing the frequency and then how fast it's moving, right?
00:42:53
Speaker
Sunilixa does this. What if there was a tap which needed to be turned? That's a good question. What if there was a completely manual thing that doesn't even run on power? We haven't solved for that yet. Do those things exist? Yes, there are some valves which are completely manual. They need to be open with human physical endurance, either turning the muscle power,
00:43:19
Speaker
muscle power. So yeah, those things would need to be augmented before Alexa can talk to them. Those are not very common, right? Yeah. Do those exist? Yeah. You could have a situation where you have something flowing and gravity being used for the force and somebody might manually kind of open and close a valve. Have we seen those? Yes, we have. But then those are not very common.
00:43:42
Speaker
Because these plans are tens to hundreds of millions of dollars of investment. So they have some minimum level of automation, minimum kind of modernization.
Sales and Installation of Alexa ELIXA
00:43:54
Speaker
Okay. And how do you install this device? Like what is the installation process for it?
00:43:58
Speaker
Yeah, so I'll just quickly talk about how we sell and then how we... So how we sell is we send... There is now like some word of mouth people coming to us and asking us, right? Which is how I want all of the selling to be. And you're selling in India or like...? So yeah, we have customers in the subcontinent. So Bangladesh, Sri Lanka, India, most of our customers are in India today.
00:44:20
Speaker
also in the Middle East and Africa. And we just started selling in the US. We hired a sales head for the US market and we've just started selling there. But we don't have any customers yet in the US. We just started the process. But yeah, this is where all our customers are today. So from Qatar to Oman to UAE to Kenya to Uganda to India to Sri Lanka, Bangladesh,
00:44:44
Speaker
And so how we, our team would go talk to somebody at the plant, like a decision maker, usually the plant manager, and then talk about a case study that we had already done in some other customer site. We just show them the case study and whatnot.
00:44:59
Speaker
and try to win the contract. So either they're doing something right now with some kind of inferior technology or they're not doing anything at all. So either we are replacing a cost or we are a new cost. And then our new cost has to be justified by the ROI that Nixx has provided. So we say things like, okay, hypothetically, this is the kind of savings, this is the kind of impact. We're going to save so much energy, which translates to so many hundreds of thousands of dollars. You would need an engineer in your pre-sales process who would actually study
00:45:29
Speaker
and then present a customized proposal. Yeah. So we have a customer success team which does both the pre-sales bit and then the post-contract, the go live process. But even our core sales guys mostly have an engineering technical background. So some of them can do it themselves. Sometimes they would take a member of the customer success team and it's a very small team. And our customer success team is super, probably,
00:45:57
Speaker
Yeah. They're the most experienced team in the organization. So they're very super experienced folks. So these guys would go in and they'd say, they'd kind of calculate the savings and they would say, okay, this is going to be the impact. We showed out a proposal to contract and then we kick off our
00:46:12
Speaker
go live process. So we have a program management team who goes from contract to go live. They work with the customer success team and within the program management team, we have our own installation team. So we have guys who will go to these plans and do the physical installation. We want to get to a state
00:46:30
Speaker
like IKEA where we could just send Alexa in a box with a manual to how to install it. We want to get there, but that's going to take some time. But having enough installation engineers, it's not stopping us from growing. So it's not a problem, but we want to get to that spot. This is a high impact purchase, so they would probably want an engineer from able to come in and
00:46:53
Speaker
make sure everything is installed properly. And definitely in India, they wouldn't want it. The pandemic has allowed us to kind of go a little bit remote on that with minimum presence at the customer plan site and doing more explanation through video. Like you can do troubleshooting remotely. Yeah. So in a way, the pandemic helped us to make that paradigm shift. But yeah, we want to get it to a place where customers can just
00:47:19
Speaker
order elixirs, open the box, configure themselves and even build their own algorithm. So we're calling that project internally Mount Fuji. The reason I'm laughing is because you're in Japan. So that project internal is called Mount Fuji, which is, you know, basically we are making elixir more sandbox. And so it borrowed a little bit from this product called Lego Mindstorm. I don't know if you've seen that Minecraft.
00:47:47
Speaker
Lego Mindstorms. Yeah, Mindstorms. Like M-I-N-B-C-O-R-M-S. Lego Mindstorms. So one Lego Mindstorms, I bought this for my kid, right? So it's a box that comes with a bunch of generic sensors and a little cube, which is the processor.
00:48:04
Speaker
And you can just like any other Lego thing, you can put these things together. You can make a car, you can make a walking robot. My son, what he built was something that can sketch by itself. So he stuck like a sketch pen to it and it can, it'll do its own. Obviously they have more geometric kind of, they're not going to do like Picasso type of stuff. It'll do like its own geometric kind of sketches.
00:48:26
Speaker
With a bunch of little things, you can build a variety of solutions. So we are taking Alexa down that path. So essentially a platform approach, like Salesforce, you can buy it for any kind of business and then you can either customize it yourself or work with a partner who will customize it for you.
00:48:47
Speaker
Yeah, that is our Mt. Soji. Why Mt. Fuji? My project names came with names like Orca, Barracuda, things that are aggressive. So I wanted to go with something a bit more positive in a mountain positivity, scaling the mountain, getting to the top type of...
00:49:07
Speaker
How does the communication happen? How does it send data to the cloud? Right now we are posting all the data on Azure and AWS using the two biggest cloud platforms so the data is posted there.
00:49:22
Speaker
Is it like a 3G sim inside Alexa? How does it talk? How does the data get? Yeah, there is a sim card. In India especially, our timing was just about perfect. So GEO launching in India and when we launched at Alexa was almost at the same time. So there were a lot of
00:49:43
Speaker
places in India where you could not get a cell phone signal, especially where these plans were looking. And when we were doing some of our early prototypes, that was one of the challenges. So we had to rely on the plans network to send the data and receive the data back. So GEO helped us a little bit there.
00:50:03
Speaker
like Alexa has a 4G, it has an internet gateway, it has a 4G SIM card and that like the factory will figure out whether they want to do a SIM or because each country would have a separate system like that's not something you can do centrally like bundling it to the SIM.
00:50:18
Speaker
Yeah. So in UA, for example, you've got to go through it. So your provider changes, your pricing changes a little bit, right? There are certain compliances required in some geographies around that stuff and byte listing of IPs and whatnot. And you ship it ready to communicate or you ship it and the customer puts in a statement and makes it ready to
00:50:40
Speaker
So the subscription for communication is also through us, right? So the customer is not taking care of that. So we are taking care of that. So we put a SIM in there and it's ready to go. So when it lands at a customer site, they just need to hook on power, hook on the communication cables, hook on the plumbing, and then grout it, meaning like anchor it to the floor because it's sufficient force. You can topple over it, right? So you just grout it to the floor and it's ready to go.
Manufacturing and Assembly Insights
00:51:08
Speaker
That process today used to take us two to three weeks in the past. Today it takes us three, four days. So that's how long the installation process takes. How did you crash it so much, like from three, four weeks to three, four days? By just looking at the different permutations and combinations of problems that we were facing, most of which had to do with just planning.
00:51:28
Speaker
and customer readiness. So when we sign a contract, customer says, okay, blah, blah, blah. This is what they're going to do in terms of data, in terms of power, in terms of communication and whatnot. And getting there and having all of those things ready to go. It is the most important thing in terms of how fast we go live. If customer readiness is there, we can literally go live in a few hours. It's just very hard to get 100% customer readiness, right? And you have like a plant in India, which builds the Alexa.
00:51:58
Speaker
Yeah, so we have a plant is a big word. So we have a plant just outside of Pune, about an hour drive from our office here. And what we do there is we do some wiring, soldering, right? And assembling of components and a whole lot of testing, right? Yep. That's more like a lab, I guess then.
00:52:14
Speaker
We have an assembly line where we are just putting together the Alexa. We pre-built them, but we might adjust the number of IOs, the types of sensors based on what we're shipping out. The top section, which is the brain and the touchscreen. So it's our design, but then we have...
00:52:33
Speaker
order them in bulk and we keep because you can't order five votes at a time. You got to order hundreds of these things at a time. And then we, because that's when the cost comes significantly lower. And then, yeah, so we do some of our own sensors and some of the sensors we buy. So even that is kind of stopped a minimum inventory. We keep that. And then we just put this stuff together and a bunch of cables running between
Haber’s Revenue Model
00:52:55
Speaker
these things.
00:52:55
Speaker
Tell me about how your revenue works. What is the one-time cost you charge? What is the subscription? Help me understand the revenue model. The impact that Alexa makes is substantial. We're talking about at least hundreds of thousands of... Because these plants are producing almost millions of dollars of product every day. They're making hundreds of tons of steel or whatever they're making. They're quite high revenue that they're producing.
00:53:25
Speaker
And even if Alexa has a very minor impact, like even a 1% change or even half a percent point change, that translates to a big impact for the customer. So at a minimum we're talking about
00:53:40
Speaker
six-figure savings, right? In a lot of cases, seven-figure savings to our customers, meaning hundreds of thousands of dollars to millions of dollars of savings, recurring savings every year. So it's substantial savings. And we take a very small fraction of that, but we end up
00:53:56
Speaker
monetizing tens to hundreds of thousands of dollars per year. So typically we assign a multi-year contract. So we don't do any kind of one-time. We don't charge our customers an installation fee or we don't sell them the hardware because the way Elixir works is a combination of what's in the device plus what's on the cloud.
00:54:17
Speaker
So the cloud is continuously communicating, there's a dashboard, there's data visualization, but there's also the AI getting updated from the cloud. So the product is not just the physical product. And then SMEs also. Yeah. So that's why we have built a commercial model in a recurring fashion and the savings are also recurring.
00:54:38
Speaker
depending on the size of the customer and pricing is something that we haven't perfected. I know we're leaving a lot of money on the table. We're giving a visit, which is great. It's a great thing to do to create a lot of value for your customers. So yeah, so we end up monetizing something like 50 to a hundred thousand dollars a year.
00:54:56
Speaker
And for a customer, what is the ROI? If they're spending 100,000 hours, do they typically save? At least 5 to 10x at a minimum, if not more. So we have some kind of saving sharing models where we bill our customer a fixed monthly, we build them a fixed amount of money every month, and then we calculate the impact
00:55:20
Speaker
maybe once in a quarter and then share, we'd get anywhere between 5% to 15% to 30% of the savings. That's a pretty innovative pricing model. So you have skin in the game basically to make the savings happen. Yeah, that's brilliant. Yeah, and customers are extremely happy to do that. And we are happy to do that as well because we're confident in what the product can do and also the sharing of savings is commercially better for us. It has always been.
00:55:46
Speaker
Yeah, so the customers get substantial savings and what they pay for Alexa is really a drop in the bucket for them compared to the savings. So yeah, so they are, like I said, going back to our first installations, those customers are still customers and we continue to monitor the Alexa with them.
00:56:05
Speaker
So we rarely lose a customer because the product is so sticky. I hate comparing it to like a pacemaker, but it's like a pacemaker, right? Once you implant the pacemaker, there's no, you're not going to take it out. So similarly, once an Alexa is in the plan, it's very hard for the customer to take it or remove it because they'd immediately lose the impact that Alexa is making. The only way that the customer could replicate it was to kind of replicate a product, which is not something that they would want to do.
Competitive Landscape
00:56:32
Speaker
Are there replacements out there like another company which does the same thing? Yes, so everybody's now trying to get there. So if you think about us, so there's the
00:56:42
Speaker
kind of the data generation layer, right? Then you have the data acquisition layer, right? And then let's say the next one is the data kind of homogenization, data wrangling kind of layer, data visualization layer. And finally you have the insights and even after the insights, you have the intervention, right?
00:57:03
Speaker
So there are different people playing in different parts. So if you think of Alexa, it is comprehensive. It's doing most of these. It's not a piece of plant machinery, but it's doing everything else. It's generating data. It's homogenizing the data. It's analyzing it, visualizing it, generating insights of the data out of the data, then using those insights to intervene. So there are people coming from, so for example, if you're a German pump manufacturer, like run force,
00:57:31
Speaker
very large German, they are making their pumps smarter and doing things like that. So the things that we are controlling are finally the machinery in the plant. So those guys are making their equipment smarter. But the replacement cycles for these are like 20 years, 30 years, because if you buy a piece of equipment for millions of dollars, you're going to run it for multiple decades. So the replacement cycles are very high. So this allows us this opportunity to kind of plug the gap
00:58:00
Speaker
and use Alex to make the plant smarter. And even today, if you were to go and buy equipment from one of these equipment vendors, they would have two versions. They would have the smart version and they would have the not so smart version. So it's like a rear view cameras in cars, right? Earlier, you would only get them in like a Mercedes S-Class.
00:58:19
Speaker
Now they're kind of common, but even now they are an additional accessory. Not every car gives it to you as a standard feature. Now there are car companies that give it to you as standard feature. But it takes time. It takes many decades for this to happen. And this gives somebody like a haber the opportunity to go in and deliver this value because that gap exists. But everybody is coming into this space. So the software guys
00:58:43
Speaker
are now thinking about edge computing devices, so on and so forth. Even somebody like an Amazon, they're starting to provide a vibration analysis layer. Now, you had startups that came up with vibration analysis, which means you put a vibration sensor, which is translating noise or vibrations into an electric signal, and then you use that to predict different things. When is it going to fail? Do I need to change the lubricant? Is it time for maintenance? So on and so forth.
Future Focus on Platform and Sensor Technology
00:59:10
Speaker
Now startups did that. Then you have the people like SKF, the guys who make the bearings. Now they are doing it. So, okay. Now you even have somebody like an Amazon who's coming from a different tangential angle. So there's many people getting into space and then they have the consultants, like the big four consultants are trying to do stuff. So yeah, it's a, it's noisy and chaotic. It's a bit confusing for customers, right? There are, there's no clear cut competition, right? So it's not like.
00:59:37
Speaker
No, I want to purchase a laptop and these are the options and figure out the operating system and then you figure out the brand. So there's a clear buying process. So it's a bit confusing for the customers right now, which is why we don't have, because our product is so good. In theory, we should have people lining up outside waiting for Alexa, but we don't have it. We have some inbound.
00:59:59
Speaker
I would say right now 20 to 30% of our sales is inbound and interest, but still 70% is kind of push sales and we're going out there and pushing Alexas. And that is something that needs to be, the cloud has to be cleared a little bit there so that customers understand and can compare and buy. So like, what is the kind of, what's on your radar in terms of growth? See right now, so let me like use a very simplified analogy. Like you had these.
01:00:28
Speaker
desktops earlier which could not get Wi-Fi signal and so you had a dongle which you could plug into your desktop and make it Wi-Fi compatible. So right now Haber is somewhere there but obviously not that simple but like there's a lot more that you can plug in an Alexa and get smart capabilities and then like you said yourself all of these manufacturers are
01:00:51
Speaker
eventually, maybe 20 years down the line, we'll have these as features. So where do you see Haber 20 years down the line?
01:00:59
Speaker
See, we will turn into a platform, but more importantly, we're going to solve for sensors in a big way. So if you think about this, all this technology that's getting built, but there's a lot of interesting things happening between data and using that data to do different things. For example, even we are doing some of this work, which is, can you use vibration data to predict the shape
01:01:26
Speaker
and size of an object and things like that. So there's a lot of interesting work happening there but the creation of this data itself has to start from the sensors. I don't see a lot of disruption happening there and again the barrier to entry in that space is also very high because you've got to understand the problem and bring together a different group of scientists from different disciplines to solve the problem.
01:01:48
Speaker
So I see as having more kind of data generation capability, meaning we have a lot of different advanced sensing capabilities, things like that. So it could eventually also have a play like you have like Intel inside when you buy a laptop. So then they could be like a Haber inside, which let's say Siemens is selling a machine with Haber inside. Like that could be like a long term. Exactly. That Haber inside thing is something that we have even discussed internally.
01:02:16
Speaker
We're yet to partner, so we have gotten a bunch of partnership offers, but we're yet to partner and test that out. So today we don't have any go-to-market partners. Everything is direct to customer. We test the customer installation, everything we do ourselves. We sell ourselves, we install ourselves. We don't combine Alexa with any other product, it's standalone. So we haven't
01:02:38
Speaker
There are multiple discussions happening around partnerships that I named some time ago. We're talking to that company, a very large Swedish company that does parts of the plant and other equipment manufacturers to kind of do this thing you're saying, Haber inside. But we did the right kind of leverage, like this one conversation I had. I was like, okay, we're going to put Haber inside of our product.
01:03:01
Speaker
But we are going to call it ABC product, right? No, the customers are going to know Haber's inside of it. So that's not something I'm very keen on. So we need to get sufficient leverage like an Intel has on the PC industry or the laptop industry. Amazing. Okay. So essentially like a two, three decade roadmap would probably have very less hardware happening at Haber and most of it being analytics, AI, and like more of that intelligence layers being done by Haber. Yes. Yes. Yep.
01:03:31
Speaker
Like I was saying earlier, how do we take all the data coming from different elixirs and to have it all talk to each other, share insights, and then work in tandem, working together, right? Not sharing data, but they're not working together. So if we can get there, then we have taken this to a whole new level.
01:03:50
Speaker
Do you also want to do like vertical integration and actually go and acquire a company which is making these equipments themselves? The only kind of company I would be interested in acquiring would be like a unique sensor company that has solved a very unique way of measuring something that earlier could only be done in a lab and now can be done. Like backward integration side of it.
01:04:15
Speaker
Because think about a smartphone, right? Now, a lot of things a smartphone can do is because of the sensors that exist in you. And either that ecosystem has to develop, in my opinion, that it's not moving us fast. So that is definitely something that we have started looking at ourselves, right? To solve for some of those sensor-related problems. Meaning, how do you measure something in real time that can only be measured in a lab or such a place?
Real-Time Process Measurement Innovations
01:04:43
Speaker
For example, going back to that example of brightness of pulp. How can you measure brightness of pulp in real time versus taking a sample and taking it to a lab and then understanding what the brightness is? If you like the Found A Thesis podcast, then do check out our other shows on subjects like marketing, technology, career advice, books, and drama. Visit the podium.in, that is, T-H-E-P-O-D-I-U-N.I-N for a complete list of all our shows.