Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
The Science of "Where": Decoding India for Modern Retail | Devashish Fuloria (GeoIQ)
330 Plays2 years ago
"India is a data-poor country."
Most founders would see this as an insurmountable hurdle. In this episode, Devashish Fuloria explains how he saw it as the ultimate business opportunity—building a formidable competitive moat not despite, but because of the chaos of Indian data.
Devashish Fuloria is the Co-founder and CEO of GeoIQ, a location intelligence startup recently acquired by Lenskart in a deal valued between $15-20 million. In a masterclass on capital efficiency, he achieved this strategic exit after raising just $3.35 million in funding. With a lean team of around 40 employees, Devashish built a powerful AI engine that synthesizes over 600 disparate data sources to provide street-level intelligence, saving clients like Lenskart and GIVA over ₹300 Cr in potential costs. His journey is truly unconventional, beginning with a PhD in Materials Science from Imperial College London before pivoting to sports journalism at ESPNcricinfo and then to entrepreneurship.
In his conversation with host Akshay Datt, Devashish revealed several key insights:
- The "Data-Poor" Moat: GeoIQ's core advantage comes from its specialized ability to clean and structure messy, unstructured data unique to the Indian market—a feat global competitors struggle to replicate.
- The COVID-19 Pivot: A free API built to track containment zones during the pandemic became a turning point. It proved that businesses could consume location data at a user-level via API, shifting GeoIQ's model from selling maps to a high-volume, scalable SaaS product for the fintech sector.
- The Lenskart Playbook: The journey with Lenskart—from a key customer to lead investor to acquirer—serves as the ultimate B2B growth strategy. Lenskart validated the product by opening over 1,400 stores with a 5x increase in speed, which directly led to the strategic acquisition.
- Capital-Efficient Combat: GeoIQ successfully competed against global giants like Foursquare (with $448M in funding) by being ruthlessly focused on solving a single, high-value problem: de-risking offline retail expansion.
Chapters
(00:00) - Introduction
(01:37) - From the Himalayas to IIT: The Path to Engineering & a Love for Geology
(03:51) - A PhD in Materials Science & The Power of First-Principles Thinking
(07:02) - The Unconventional Detour: From Engineering to Sportswriting at ESPNcricinfo
(12:09) - The Genesis of GeoIQ: Finding Opportunity in "Data-Poor" India
(16:03) - Under the Hood: How GeoIQ Turns 600+ Messy Data Sources into Intelligence
(25:17) - The Pandemic Pivot: How a Free COVID API Unlocked a New Business Model
(29:12) - Finding Product-Market Fit: How Fintech's Need for User Data Drove Scale
(41:12) - The Lenskart Story: From Critical Vendor to a $20M Acquisition
(46:34) - The Future: Expanding to the US & the Next Frontier of Location AI
Hashtags
#FounderThesis #DevashishFuloria #GeoIQ #Lenskart #AkshayDatt #StartupIndia #SaaS #LocationIntelligence #B2B #DataScience #AI #MachineLearning #Fintech #RetailTech #CapitalEfficiency #Entrepreneurship #VentureCapital #ProductMarketFit #StartupPivot #Acquisition
Recommended
Transcript
Introduction to Devashish Falloria and GUIQ
00:00:00
Speaker
Hello everyone, I'm Devashish Falloria. I'm the CEO of GUIQ.
00:00:17
Speaker
Everything is data. This podcast that you are listening to right now is nothing more than bits and bytes streaming to your device via the internet. And while you may think that this is one-way data, we at the podium get back data from you, the listener, about where you are located, what device you are using, what is your OS, and which app you are using right now. In such a data-rich world, the companies that can leverage data for better decision-making are really at the top of the value chain.
00:00:45
Speaker
Think of the large tech giants like Google and Facebook. They are in the business of leveraging user data to help advertisers serve their ads more effectively.
Devashish's Early Life and Education
00:00:54
Speaker
In the niche of location data is a company called GOIQ which helps businesses make smarter customer decisions by using location data.
00:01:02
Speaker
In this episode of the Founder Thesis Podcast, your host Akshay Dutt talks to serial entrepreneur and co-founder and CEO of G.O.I.Q. Devashish Filoria. Devashish talks about the power of location intelligence and how we built an API business that leverages big data and machine learning to help consumer internet companies make smarter decisions. Stay tuned and subscribe to the Founder Thesis Podcast and any audio streaming platform to learn about the leveraging data for business modes.
00:01:37
Speaker
I was born in the mountains. I'm from Uttarakhand. It was a town called Pori Garhwal. Beautiful place surrounded by jungles. So we grew up with a lot of outdoors. I used to do well in school. It was quite clear that there were two options. Either you go and become an engineer or you become a doctor. I did not want to become a doctor because there was a doctor at home and I see what kind of a that means. So by default, I ended up working towards
00:02:05
Speaker
engineering and one of my uncles was a professor at IIT Kanpur, so I had been visiting IIT Kanpur. So there was some sort of aspiration value associated with that, but nothing especially which was just focused specifically on an IIT. Even today if I look back, I feel if somebody would have guided me, I would have been a geologist, not an engineer, but that's how life takes you.
Career Beginnings and Realizations
00:02:32
Speaker
So IIT was absolutely great for learning. I joined Tata Motors and got an early job. It was into sales and marketing. One year, I think it was based out of Bombay, but traveling north, southeast, west across the country. And the interesting thing was I was a salesman and I was selling these axles under a trailer truck. The axles.
00:02:58
Speaker
And I was somewhere in Rajkot, and one mechanic asked me a question about the quality of the waste, and I had no answer to it. I said, OK, so what really did we do with the degree? I mean, we did not have this sort of practical knowledge about loads and real world voting. So at that point, I started thinking about higher studies.
00:03:21
Speaker
So this opportunity came from Imperial College, London. I went ahead and it was a straight admission to PhD, no masters. Quite cool. The entire PhD was set up based on that. It was an eye opener for me. I said, you can solve complex problems, but you have to keep going back to the basics. And what was your micro-specialization? Typically in PhD, there's like very micro area, very small niche in which people become the best in the world in that small niche.
00:03:51
Speaker
So it is very interesting because now I was looking at deformation of semi-solid alloys. So when you are making or creating some sort of an alloy, obviously you have to heat it up, it's liquid and then it starts solidifying and then you define the properties based on the rate of that solidification.
00:04:13
Speaker
My entire Ph.D. was in that zone of semi-solid. When it was not fully cast, it was not fully liquid, but a lot of property changes were to be controlled in that phase. And I was focused into that specific 50 degree temperature zone where a lot of things used to happen.
00:04:33
Speaker
So that's, that was very interesting.
Transition to Sports and Data Ventures
00:04:36
Speaker
So spend some time with the UK government as well, as part of a policy group, caught a great view of how the policy makers think about things. That six months I still established my views even today, which remained true on electric vehicles, how geopolitics comes into the picture, what the government would be thinking about before applying a policy.
00:04:58
Speaker
But post that I went into an engineering consulting where I was serious about this job because there was a massive cricket pitch just outside the office. Beautiful ground. I said, I would like to work at this place because this is a cricket ground. I used to play a lot of cricket. And the director who was interested in my work
00:05:20
Speaker
was primarily interested because he saw somewhere in my CV that I'm an off-spinner. And he said, hey, we want an off-spinner in our club. So apart from all the work, this cricket thing was also somewhere speeding into the whole decision making. But then I was again working in the structural integrity of
00:05:39
Speaker
oil refineries, airbus structures, ship structures. It's primarily in Europe. So it was a consulting company which used to tell these bigger oil companies what sort of monitoring mechanisms to employ to avoid certain problems. And it's all based on technology. So now from image, one of the key transformations that I did was the algorithms that I was writing on images.
00:06:06
Speaker
on converting 2D images to 3D structures. It's called tomography. I applied the same thing on sound now. So now in all these structural integrity, it was a sound signal which was being used as a measuring tool. So now how do you generate
00:06:21
Speaker
a 2D image from one-dimensional signal. So that's where same principles were applied in real structures. There were images being generated, neural nets which were the name of the game back then, no AI back then. A lot of these algorithms were based on how do you identify any changes in structures based on these signals.
00:06:43
Speaker
The best part was spending a couple of days inside A380, which a skeleton of A380 had to lose. And then you realize how flimsy a blame is. I guess after this is when you came back to India and you had six years engaged back with sports. Tell me about how that happened.
00:07:02
Speaker
So, I just walked into the offices of Krikimpu in Bangalore just to ask questions, just being curious. How do they work? How do they operate? I went there once, I went there. Second time, I went to Bombay and the editor, Sambit, he said, come over again. And I was just being curious. I had no plans of working in cricket. But then he said,
00:07:23
Speaker
that, hey, why are you wasting my time? If you want to work, just come and work and spend 15 days with us and we'll see. And at that point of time, I said, oh, okay. So this is also a possibility. So I spent 15 days at cricket for really love the place. And that 15 days became three and a half years very, very quickly. So it's a writing slash editorial job. So partly the team is out.
00:07:49
Speaker
So you are a writer. If you are on the desk, then you are the editor. Fantastic job working with an international team, working 24-7. I don't remember ever taking a holiday for Independence Day or Diwali or anything. I mean, we are just working through and through. It's a really fun period of my life. But then I was also picking up signals like, OK, Triginfo is Triginfo because it has a lot of data.
00:08:17
Speaker
And I used to love playing with the data myself. So I thought, hey, data brings in the engagement. And something similar could be done for other sports as well.
The Birth of GeoIQ
00:08:29
Speaker
And that's when I started thinking about that, OK, now I need to go back to building things. And the first thing I thought of was that, OK, let's apply the same philosophy to amateur sports. And that's when we created me and other co-founders. We got together. We discussed this idea that, hey,
00:08:47
Speaker
Can we build communities around sport? But with data is a central element to it. So we created this app which would allow people to challenge each other, record each other's score and it used to automatically build their ratings in the background. So imagine if you're in Bangalore and let's say there are 5000 badminton players in Bangalore. Even if you have not played a player, you would know
00:09:13
Speaker
how good or how bad the player is based on few interactions from other matches, et cetera. So there were some algorithms involved in the backend, but the whole idea was to bring in engagement with some data. So that's what we did. We did it for one and a half years, couldn't scale beyond Bangalore. So at that point of time, and there were co-founders, people had some other pressing needs and so that had to be shut down and wrapped up.
00:09:42
Speaker
building an advertising-led business like this would be going to pay itself through advertising.
00:09:49
Speaker
See, at that point of time, building the first product, you probably are not thinking that far. And at that point of time, the only thing you wanted to do was to increase the engagement. Now, there could have been many business models. I think we were a bit immature back then. This model we thought of would have been around getting all the tournaments into the system and asking the tournament organizers
00:10:13
Speaker
to pay for this platform because they would get all the players into the system. So create an engagement system from there on. We also did make some money out of it. But again, it wasn't looking scalable at that point of time. Okay. So then what now? You are probably out of your personal funds also. You must have put in your savings.
00:10:33
Speaker
out of a job also, like you must have been in that fairly lost type of a state. Yeah. We had raised a bit of money even for Zilladder, but yeah, money runs out. But I think what, so I wasn't really keen on getting back onto a job straight away. One thing which I done in 2011 and I really cherish it is taking that three, four months off to just
00:10:59
Speaker
slow down a bit and think without any unnecessary rush. And so I took some time off and I've really loved these breaks in between because I've spent a lot of time back in the hills. So I came back and there were a few thoughts again in mind. I started speaking to some of the mentors here that this is what we want to do. There are three or four possibilities, something in the B2C space.
00:11:25
Speaker
something in the B2B space. At that point of time, my current co-founder, Tushit, he was also going through that zone of flux. And I've known Tushit over the last seven, eight years. He's a junior from IIT Kanpur. So we were spending a lot of time brainstorming on what works, what doesn't work, getting feedback from the mentors. And it was
00:11:50
Speaker
Basically that exercise of iterating through multiple ideas and seeing that, okay, this seems like a good possibility. And that's when we entered into working into building GUI cube. And what was the possibility you saw? What was the gap or the problem statement that you wanted to solve?
00:12:08
Speaker
Around that time, the sheet was in a freelance capacity working on a problem. It was one of the big-time retailers in India who was trying to spread across tier-2, tier-3 cities and open 5,000 stores, 6,000 stores. They made these big announcements in media. Small, five-act stores.
00:12:28
Speaker
Small format stores, yes. But their early success rate had been really poor. So only 20%. Which sector was it in? What was the product? A retail. A retail. I mean, with a retail like food, clothing. A general retail. So Reliance, fresh kind of thing. So 25% of their stores were surviving after two years.
00:12:50
Speaker
And they would say that we know the metrocities, so we know where to open these stores. But how do we decide where to go in India? And so, the question that was coming was, where is the data? How do we take data bank decisions? We are left to the mercy of...
00:13:09
Speaker
what the ground beam is telling us, and they would push their own agenda, so there's no centralized innovation. So we started looking into what exists externally, what is not part of the company system, but what exists externally. So we looked at government data, we looked at satellite imagery, we looked at map data, and just started layering everything up on a map. The question was about location, so we started stacking
00:13:36
Speaker
all the information that we could get on a spatial database. So that, let's say if you want to take a call on merit on some street and you want to understand what's happening around that location, you start getting certain kind of ideas. It was quite powerful because this sort of external data systems
00:13:55
Speaker
haven't existed in India till now. And India is a data-poor country, so people collect their own data. But outside that, they don't really know much. So this was the holy grail that we were after, that if we can make this real-world data, real-world information easily available in a manner in which people can extract intelligence out of it, I think we would add a lot of value. So that's what we started with, broadly, that the entire country's data
00:14:25
Speaker
in a single place, available through maps, available to everybody.
GeoIQ's Technology and Challenges
00:14:30
Speaker
That's how we started. What were the data points? You said like government data. What exactly was this? Like the kind of data that you were putting onto a
00:14:39
Speaker
So if you look at government data, it is very rich. There are questions about its vintage census. Something as simple as census is 2011. It hasn't happened. But if you take the other argument, there is no other survey which is as comprehensive as census.
00:14:57
Speaker
and it forms the basis of everything. So, no census just doesn't measure the population, it measures maybe 200 things about each and every locality. Government calls them what? Each and every household basically. For every household they will ask you how many children do you have, age of children, what do you earn, do you have a car, do you own the house or are you living in the rented house?
00:15:21
Speaker
Yes, rented house, what is the call it in? Very subtle questions, but it's a big question bank. Until it publishes it in credit form. Says that in this ward, in this village, these are the broad parameters. Now, government talks in wards.
00:15:37
Speaker
But the world talks in pin codes and localities. So there is no match. So therefore, there was first value was that, can we help people speak to this database first. Government also publishes a lot of growth numbers. So now these are more recent numbers.
00:15:53
Speaker
what is the growth rate of population in this little village, that village, etc. So now you apply growth indicator, so you get updated numbers across it. Then we also, now this is broad information aggregated, we set to
00:16:08
Speaker
combine it in a form of a pin code, we first need to break it down into smaller unit. So we use satellite imagery to identify where buildings are. And then we broke all these numbers because these were all population-based numbers onto buildings. So now we could go down to 100 meter by 100 meter.
00:16:25
Speaker
So the way to break it down into buildings is just say if one ward has 100 buildings and so you just divide that by 100. You break it on the rooftop. 100 or basically on the rooftop area. Okay. Proportionate to area. Proportionate to area.
00:16:41
Speaker
That also allows you really good separation in terms of, hey, this is an army land, agricultural land. This is where the population is concentrated. You start getting a really, really fine-grained picture of how the population is spread out and what sort of socioeconomic backgrounds they're coming from. Now you start, and the technology exists for satellite image to be tagged as residential farmland commercial.
00:17:06
Speaker
That is where algorithms come into the picture. Very generic algorithms exist, but we had to tune some of these algorithms to our needs. Then recombine it with all the other data points. There's a lot of cross-stitching happening here between the print database.
00:17:22
Speaker
and then break it down to these really small units now if you once and you had the you were able to map a ward like say south Delhi ward for example so you were able to map that on a map that this is south Delhi ward yeah this is what the government is talking about
00:17:40
Speaker
So that is something you have to manually do for each word or is that like available, like the word location. Government has those, these are special files, geospatial files. They're called shape files. These are polygons. You convert them, fill in the data, then break it down further. It's surprising that people don't really know
00:18:04
Speaker
what are the extent you might call south delhi as including large but another but somebody else might say that south delhi for me and here so there is but don't use these standard methodologies even pin codes it's a postal department shape so we were breaking it down all this data points
00:18:23
Speaker
Once you've broken it down, now you can recombine it and initiate. So, the one thing which, you know, very early thing that we released was, we released population on PIN codes. Because we had broken it down, now we recombined it. It would be surprised to know that every company works around PIN codes as some sort of base. But nobody knows in India what the population of a PIN code is, because these are different shapes.
00:18:49
Speaker
They might be overlapping. So now the only place you will find population of a pin code is UIQ's database, because nobody had done it before. So there's little things which start creating little impact. But I think it was in the second year that we started getting some sort of a thesis on how this data has to be used in a more productized manner. That's how we started creating. Even that step has through a couple of pivots.
00:19:18
Speaker
Second year is which year? This is 2019. So, what was your thesis in 2019 that you formed about? In 2019, we said, if we are giving people location data, the most important manner in which location data makes sense is in maps.
Pivot During COVID-19 and New Opportunities
00:19:39
Speaker
So you basically gave them a client's platform where they could just see everything layer by layer. Population as a layer, this as a layer. So it was a tool that we created for business analysts to quickly identify where their kind of hotspots or dark spots were. It was primarily data as a service.
00:20:00
Speaker
but on maps. There were parallels across the globe. One of these American companies, Esri, has been doing something similar for 50 years. It's more than a couple of billion-dollar revenue company, a fairly legacy-based player. They have their own file formats. They have their own business analysts, et cetera. But we thought that we can just simplify it and just preload the maps with data points. So you just send it layers and access this information.
00:20:29
Speaker
But data point is obviously population. What else? Then there would be obviously a lot of socio-economic parameters. But then there would also be what kind of like car ownership, house ownership like that. Car ownership, house ownership. If you're looking at a countrywide thing or countrywide data spread, you also look at material of house, cement or hedged roof. Is it connected to sewage or not?
00:20:58
Speaker
these sort of parameters give certain indicators about the kind of population that you're looking at, then there would also be business location. So we had started crawling all business websites to see where their stores are.
00:21:14
Speaker
So that would also give us an idea of where certain kind of business clusters were. Is this this shop or this particular street is only apparel heavy or what other kind of businesses are existing here? Then you start layering it with real estate information.
00:21:30
Speaker
Okay, there are apartments here, lots of apartments everywhere and the real estate value. So now real estate value becomes a very direct indicator of on this particular street, how rich the people are. So nobody tells you their income, but you have to use these secondary indicators. So you think of a different C1 street and the real estate value shooting up and suddenly you see a lot of brands coming in the vicinity.
00:21:54
Speaker
So now you have not only figured out that these are rich people, but these are rich people who are going out shopping in these kind of shops and in these kind of restaurants. So that sort of graph starts building about places. So even without knowing somebody in that stream,
00:22:11
Speaker
were able to scrape this from like listing sites like we listed listing or business listing, yellow pages, those kind of sites. So you just basically you find anything which has an address in it and you start putting it in this database and make sure that you're making those connections so that it becomes a very strong graphical database.
00:22:32
Speaker
And so once you fought this thesis that you want to build this, did you raise funds to build it? Because this would need more investment. You would need a team. We raised a small seed round from a bunch of angels in 2019. Again, things were not very neat and established at that point of time, but we needed some bit of focus. And at that point of time, we also realized that data is not the only thing which will make you make a business out of this because
00:23:02
Speaker
Hey, I have created thousands of data points. Now, for a client, it becomes difficult to choose what is important, what is not. What they need is an answer specific to their question. And what we were saying is that people would choose to take the easy options out. They will only choose population, some income statistic, but nothing else.
00:23:23
Speaker
So, at that point of time, we were working with one of the scooter-sharing companies here, which had a lot of their own location and permission. They had scooters spread across. They knew where there were more bookings happening, where the scooters were getting stolen, and they wanted to understand
00:23:42
Speaker
what makes a location bad for them. And our team was, the DNA of the team was very data-sense oriented, very machine learning oriented. We said that, hey, if we have to now tell these guys what is good and what is bad for you, we ask them, tell us what you know already.
00:24:01
Speaker
Bangalore, you know, where more bookings are happening, where theft is happening. Let's start bias our choices here. Let the model figure out, let the model interact with everything that we have and everything that you have and this will figure the exact location, locational parameters. So we did that and we figured out certain very interesting insights out of that, that there's scooters getting stolen within roads were very narrow. That was one parameter out of these houses which get picked up.
00:24:30
Speaker
Their scooter was getting sold at where the population density was extremely high. Their scooter was getting stolen where the density of auto repair shops was high. So now this whole equation was getting made based on what they knew and based on this data.
00:24:48
Speaker
So that was when, again, the next stage in our development started happening.
GeoIQ's Role in Financial Services
00:24:53
Speaker
We said data is fine, but some bit of AI has to come into the picture because we as humans won't be able to figure out what is important out of these thousand parameters. So we said, well, let's start with what you know. You must have been providing this as a consulting rather than as a SaaS at that stage because it was so complex that a regular user would not be able to make sense of it.
00:25:17
Speaker
And so we were in the middle of this and then COVID happened. And then all these projects went on a standstill. And at that point of time, we had seven member team, smart people, fast working people. And people were quite passionate about, did you have enough money in the bank to survive? We had some money and
00:25:41
Speaker
So the question at that point of time was that, what do we do about COVID? And people were quite passionate about all these lockdowns and what was happening. So we realized that people were asking when the lockdowns happened, people wanted to know where are these containment zones and where they are not. And government was releasing this information.
00:26:01
Speaker
But the containment zones were where the government was putting these blockages, where there was a high intensity of cases. So, they were saying that nobody goes in and out of this zone. So, they were blocking out certain parts of the city, which would impact e-commerce players because they would not be able to do their deliveries and so on.
00:26:22
Speaker
Absolutely, e-commerce, food delivery, everything was dependent on that. Now, government was every day releasing this information in PDFs and they were telling that this area is blocked out. Now, what is the extent of that area? Nobody really understand. We said...
00:26:39
Speaker
We understand this. It's not that same problem that it's not a pin code that they're sharing. So it's not usable by a business because businesses ask customers for pin code. They would block out a building. Now, where is that building? What is the extent of this building? And they come up with a definition like this building and 20 meters around it.
00:26:56
Speaker
Now, from our perspective, we know where the building is. 20 meters around it is a circle that we have to draw. So, we were able to very quickly read all the government information across different cities, across different states. And within two days, we built this very simple API, opened it up to the world, saying that
00:27:18
Speaker
you give me the geo-coordinate, the latitude, longitude, I'll tell you if it is inside or outside, that's what you are doing. Within a week, 2000 companies had signed up for that API, including Amazon, including Flipkart, Inmobi, all sorts of companies.
00:27:36
Speaker
Okay. What was Inmobi's interest here? Inmobi wanted to give separate advertising channels. They wanted to target based on geos as well. So they wanted to understand a different messaging might work. So a lot of people signed up for this, many people used. And if you remember people, there was offices started opening in June and July. So then businesses wanted to know if their employees were coming from containment zone or not. So they, they started using DFL. A lot of people started using these.
00:28:06
Speaker
Yes. For an e-commerce company, which is taking my address, how do they get the latitude longitude? There is a like a solution for converting an address to latitude longitude. They have the lat long from your device. And they also have the lat long from having done past deliveries at that address. And if there was just an address and we would convert it using one of the existing geocoders. So that's what happened, but we knew that it was a, and this was a free to use thing. You were not monetizing it.
00:28:36
Speaker
It's a treat to use. We tried to, but then we said that it is ephemeral. This won't go beyond three, four months. But the question is, what insights is this giving us? Initially we had thought that people do not have lat-longs and therefore we were giving information on maps. Now when we realize that people have lat-longs of the users,
00:28:59
Speaker
So we can give answer on lat-long through an API. Now that became the very interesting start of what GeoIQ is today is because some of the financial companies had also used that API. And then they started asking us that, Hey, but you also have this other data. Can you supply this other data in the form of an API as well? Because for us.
00:29:23
Speaker
user personalization, understanding the user better, understanding the risk better. We need as much data as we can. Since you're talking about lending focused companies, before they lend money, they need to do a risk underwriting and this would help them in risk underwriting. Yes. So now you think about
00:29:41
Speaker
where people were using it before was primarily for expansion, for doing pin code level analysis. Now they started asking questions on a specific user. If this user lives on the street, what do I know about this user? I know these thousand new things about this user. That's where we started experimenting with a couple of fintech companies. We started seeing that this actually starts predicting risk. This starts predicting incomes.
00:30:11
Speaker
This starts predicting where are you going to face problems in collections tomorrow. But it is all happening at a user level rather than a broad level. Now this started making a massive sense for us because if you're looking at a user level,
00:30:27
Speaker
The volumes are higher because everybody is dealing with thousands of users. So for me, rather than selling maps, now we started selling that. Hey, I'll give you these data points per user for X rupees. That X rupee is a small number, which makes sense to the business because it has a massive value for them. And it made sense for me because that X rupee into 100,000 users is a massive bill for. So now everything started making sense from that point onwards.
00:30:56
Speaker
So therefore, in post-COVID, our entire focus shifted to financial services, where now all this location information that we are building gets delivered around the user. Imagine, let's say you go to a bank and you're asking for a loan and the bank asks you, hey, you're Akshay, give me a parent.
00:31:18
Speaker
give me some of your bank statements, is that, etc., other card. These signals, they're trying to understand you. They would ping bureau in the background. They also ask your address. But because they ask your address, because of that address now, there are no thousand more things about you now. And what happened is, in the fintech boom that was happening in the last couple of years, everybody is going out, reach out to the new to credit user.
00:31:49
Speaker
New to credit user base have very weak information bases and that's why you're going out to there. These companies also need some bit of information. So this data suddenly starts becoming really powerful. Let's say people have downloaded my app. One person is a Chhattisgarh in some village. One person is sitting in Ahmedabad. How do I compare these two users in real time? And that's where it means
00:32:15
Speaker
locational information started becoming extremely important in the financial systems. User information based on lat-long is like one of course would be maybe that for that specific geography you would be giving some sort of socio-economic indicators like this is a upmarket area or it's a downmarket area or high density, low density. Give me an example, what kind of a picture can you paint through just one lateral?
00:32:40
Speaker
So, for one lat long, imagine you are in Bombay, and in Bombay, we know Bandra West and Bandra East are two different areas. So, if I know somebody is living in Bandra West, even without going there, I can define that this person is slightly affluent, and the one in Bandra East is maybe mid-income.
00:33:00
Speaker
And if I go towards Dharavi, I know it is lower income. This is our common knowledge coming into the mix. Now, even if I go to Bandra West, I know Hill Road is a very bizarre kind of area. It is slightly lower income compared to Kata Road, which is extremely high income. So, these...
00:33:18
Speaker
There are these minute separation between these places. How this data comes into the picture is like I explained earlier. Density of buildings around. So if I know that this is a lack law, this is where the user is. Within 100 meters, what is the average rent rate? What is the average money people are paying in the restaurants in this area? Is this a rich neighborhood surrounded by poor neighborhood surrounded by rich neighborhood? So you can do all these things.
00:33:47
Speaker
do all these things with the data that is coming. Do you give this as like numbers or ratings scale or because they go out as ratings? Earlier I spoke about the machine learning bit, which you rightly pointed out was going out as a consulting thing.
00:34:02
Speaker
We productized that thing because now we have around 3000 features about each location. I don't know how to choose out of that. Even a client don't know what to choose out of that. We said, hey, there is a machine learning tool. All you need to do is throw in the pattern that you definitely know. These are my 1 lakh users out of which I definitely know these are high income groups and these are low income group users.
00:34:28
Speaker
I want to create a model. Firstly, I need to understand what makes these rich people rich from a location perspective. What makes these poor people poor? And if I understand that, then I can repeat that throughout India. So they create a model. It takes 30 minutes to create a model. You basically upload a data, press a few buttons. It does all the analysis in the background. It generates a report that the locational features are predicting
00:34:56
Speaker
to such and such accuracy limits. Now you want to use it. So for a lending fintech would upload all the customers who have taken loan, what is the value of the loan? What is the repayment history? And therefore, which is a good customer, which is a bad customer, which is a high value customer, which is a low value customer. And that data is read by your system. And then your system is able to in future give some sort of a recommendation that this is high likelihood of repayment or high likelihood of default.
00:35:26
Speaker
See, one caveat of working with financial institutions is that data privacy is an important item. You can't steal information. So what banks and financial players share with us or upload on our system is just a bunch of flat loans or addresses, and they're the condition of good and bad. I don't need to know. It's a one and zero
Expanding GeoIQ's Applications
00:35:46
Speaker
for me.
00:35:46
Speaker
Now that one and these definitions could be completely different from company to company. For me, it's a one and zero. Is it literally like just good, bad, or is it also like shades between good and bad? Like very good, average, bad, very bad. It can be shades. It can be shades as well. Very simply put it is something like one and zero. Now that question.
00:36:06
Speaker
1 and 0 is just like it could be 0.5 and it could be 0.7 also. Got it. Okay. And now we start the model gets trained on our system and then you can use this model at the scale. So you're frame the model the same because the data is coming to you with this rating of between 0 to 1. So next time you get a new lat-long, you are able to give back a rating between 0 to 1.
00:36:27
Speaker
Absolutely. We tell them that this is 0.9, 0.1, 0.2. On day one, they know what quality of user. Not on day one. Within half a second, they know what kind of user this is without really asking any questions. This is like one use case very specifically for lending companies. Give me examples of other use cases beyond lending. Even in financial institution, there are multiple use cases. Now, this one and zero question is
00:36:55
Speaker
quite vast. It is right at the entry point the same company says I want to know one and zero in terms of income because based on that I will do my product recommendation. The second question that they ask is which is a different question another one and zero it is hey if I've given a loan to this person
00:37:12
Speaker
how risky or how not risky it is. There is another question then they would ask that, hey, I've given a loan to this person, will I be able to make a collection if the need be? So what will be the collection propensity? So now there are two follow on questions. If you think of it, even product wise,
00:37:30
Speaker
For 10,000 rupee loan, a good and bad definition is completely different compared to a 1 lakh rupee loan. So now there are the number of use cases within these days are numerous. The good thing for us is that all these decisions are being taken in near real-time now. So people need more and more data.
00:37:48
Speaker
And so that's what's happening in lending companies, multiple use cases within lending companies. When you go to insurance companies, again, very similar, everything is very similar in terms of that. Now the question that they are interested in fraud, one or zero, how likely is fraud going to happen at this particular place? And this is like for motor insurance and stuff like that, like motor insurance, even health insurance as well.
00:38:16
Speaker
Apart from Frintech and insurance, the following use cases come in e-commerce. E-commerce has a major problem with returns and frauds because that's a massive cost for them. In e-commerce, when you go to the checkout, right at that point, when you type in your address, a model is predicting in the background how likely a fraud candidate are you. Only if the model says that you are highly likely, they say no cash on deliveries for you.
00:38:44
Speaker
They'll say, okay, no return policy for you. That flow changes. In all these things, again, everybody till now has been looking at an individual level, at an individual's historical status level, and they would look at me covers, they would say, is this card about to expire or whatnot. But what we are bringing into the picture is that
00:39:06
Speaker
the real world idea around it. What does this person look like? And if there are two fraudulent transactions, is there something common between them in the real world? What is your revenue like currently? This year, what do you expect to close at? Have you raised more funds? Talk to me about the monetization of this platform and the business economics.
00:39:26
Speaker
To start with, I think one and a half years ago, our investors used to question us, hey, India is not the market for this. You need to go to the US because in India, nobody will pay you 5 lakh, 10 lakh rupees a year for this. Now, on an average, our customers on the top customer is somewhere around two groves and lifting a finger. So that has been one of the major changes that we have created in the system. It's a value.
00:39:54
Speaker
Yeah. And is it based on number of calls, like a number of calls? Yeah. The bigger users would make millions of calls per month. So it is based on how many calls? Yeah.
00:40:06
Speaker
And just to clarify for our listeners, so calls is essentially every time they ask you for data. Yes. Every question on a lat-long is one call. So that started happening. We have gone beyond a million dollars in ARR, which is an important number for any SaaS-based company. And in the next, this financial year, we are looking to cross around $3 million in ARR. The plan is fairly well-established for that.
00:40:32
Speaker
I think the learnings that we took from the last one and a half wars have been played.
00:40:38
Speaker
You raised some more funds also, right? After that initial seed you did in 2019. Yeah, yeah. So in 2020 also we took some funds from a couple of investors. And last year we got some of the marquee names to back us. So Piyush, Bansal, Kunal, Bell. These guys jumped into the mix as well. So we are yet to go for a VC round. The plan is to establish the revenues, establish all these things.
00:41:05
Speaker
How did you get such market names and tell me about 5 days? Were they challenging? What were the obstacles?
Future Prospects for GeoIQ
00:41:11
Speaker
What were the learnings? So you would have realized Akshay, because we've been selling a very new kind of product in the market. It takes time for people to understand. So last year when we were speaking to few investors, we realized that maybe now is not the good time because we still need to do a lot of work before we get into that hyper growth mechanism.
00:41:32
Speaker
What percentage of your revenue is currently from FinTech? Almost all, I think 95% is from FinTech. So that is our primary. We are still working on the FinTech piece itself because there is a large pool of customers that we still need to harness. And when you are selling something which is helping them in decision making,
00:41:52
Speaker
in real-time. The challenge is the sales cycle because you are becoming part of their real-time decision engine. It doesn't happen every day. It doesn't happen that they like it and they'll integrate it tomorrow. It goes through many sanity checks. So that is one of the challenges that we are trying to overcome. They want to do a pilot and the tech team will have to make a new algorithm.
00:42:16
Speaker
Yeah, so that part is the bureaucratic part around it takes some time. But I think once you get in, because you're now integrated, you're hard to remove. We're fairly happy with how that has progressed.
00:42:31
Speaker
work on them in pattern like unlocking e-commerce also because each of these will give you unique learnings and there'll be a certain amount of time the product needs to mature for that use case. There is work happening in the background Akshay and I think all we are trying to do is not
00:42:48
Speaker
put any revenue numbers into the mix for these different particles. At this point of time, we want to understand those problems. We are working with e-commerce players. We are working with some experiments around with guys like ShareChat as well, for example. Again, what's the use case for ShareChat? Personalize the news. Personalize the address, news, and personalized adverts because ShareChat is also an advertising platform now.
00:43:13
Speaker
Who are your competitors in this space? I've heard of this company called Nexvillian, which is also into geolocation, AI. If you think about globally, there are a few companies which are realizing the potential of location data. Also, 2023 cookies are going to go away. Then location is going to become even more important. But traditionally what companies I stated earlier, companies like Esri, companies like Carto in the US,
00:43:42
Speaker
They were taking a very map based approach, business analyst based approach. And it was like a BI tool and like more service. It is a product. It's a SaaS based product, but it is think of it as you're doing BI on maps. So they have been taking that route.
00:43:58
Speaker
But now we feel that the kind of group that we are taking where, you know, we're not focusing on business analysts, but the data centers, because our team has a bunch of data centers, they have seen this problem day in and day out that you want to involve this data, but you don't want to spend time working on this data. So if it just came in straight away, it would be fantastic. So is Snapdeal using you like considering that Kunal Bell is an investor?
00:44:24
Speaker
some work is happening in the background. So again, and what about Let's Cut? Cut is using it as us in a very different use case. They are think of a slightly traditional use case, but in their store identification modules. And their question is that if my ground team is giving me a thousand possibilities of stores,
00:44:46
Speaker
Where should I go? Out of those 1000, which 10 are important? So Lenskart is taking again, a lat-long based decisions in understanding the potential of a store, then building another model to predict the revenue of the store, then building another model to predict what sort of merchandise mix should be there. Again, everything based on locations.
00:45:09
Speaker
I can understand revenue predictor because they would be giving you the last long data for all stores with a zero to one rating or revenue. But for product mix, how are they doing that? Again, you would say that so they have different ranges of products. Some are expensive ranges. Some are so now budget ranges. So in certain areas, you'd say is this let's identify if this area is good.
00:45:35
Speaker
Like how much does budget products sell here? How much premium products sell over here? Yes. Got it. Okay. Very interesting. Are you planning to raise more funds? Do you need funds for growth or is there enough revenue coming in to sustain? Right now we are good, but I think we are looking at that the next year's target, the 3 million target that we mentioned, and we want to raise funds post that, or at least when that pipeline is absolutely clear cut, then we'll go out to raise funds.
00:46:04
Speaker
What do you need friends for? We need to access more data, build the capacities to use even paid satellite image. Till now we have worked with only free satellite imagery. That's an expense. And give you an example, what do you get from a paid satellite imagery?
00:46:21
Speaker
Also, the resolution is the only thing. So in free satellite images, you'll get a 30 meter by 30 meter resolution. But you can go much finer than that if you start paying for satellite images. Also, we are looking to open the same platform up in the US market as well.
00:46:42
Speaker
We have released a beta, which is again being tested by a couple of customers in the US, but that will require a lot of work and investment too. I guess you must be looking at the world through a lat-long lens. Like for every information you hear, like you hear about some crime, you would want to know what's a lat-long for it, or probably you would be wishing that on FIRs come with a lat-long and things like that. Like it must have changed the way you view the world.
00:47:07
Speaker
Absolutely. The news, the moment we look at news, we see the, where is the city this is talking about? Can we pick everything and tag everything together so that we get a realistic picture and very interesting piece on that, the kind of problems that you can solve. We were doing an experiment on American data.
00:47:26
Speaker
And so we found a sample which basically said that this is where the Republicans have won, this is where the Democrats have won. Now, so we threw in whatever data we had built up in the US market into this mix and thought, okay, let's try to predict a Democrat to win using these locational parameters.
00:47:46
Speaker
So, obviously, the first few impact parameters were obvious. They were race-related, they were infrastructural city or rural, those sort of parameters came in. But somewhere down the line, there was a factor which said that in areas where the cases of diabetes are high, Democrats don't win there.
00:48:07
Speaker
So now this is basically the power of it. You throw every sort of data into the mix. It is not a causation, but it is a correlation. And let's say data teams are armed with these orthogonal interactions as well. They can design their product. In this case, they can alter their messaging completely. What do you mean by orthogonal here? Orthogonal means that you're talking about elections. You think about it and you say, what has this got to do with disease rates?
00:48:37
Speaker
So it is very far off. But when you look at the data, it holds true for a lot of cases. So now you might go ahead and dig deeper and say that maybe there is a reason for this as well. But the first step is that it gets highlighted for you and the businesses can make more better sense out of it. They can explain it better or why it is important. In this case, I think it is explainable.
00:49:03
Speaker
How is it explainable? I think we have seen that, so it's not just diabetes, diabetes is a factor, obesity is also a factor. And we see that happening in Midwest, like Emore, not in the cities. You can draw those correlation. Now, how would anybody use it as
00:49:21
Speaker
change the messaging around your candidates. The photos, the stock images you use, maybe you have slightly worse. If you get the signals, you can do something about it. Otherwise, it is very specific to race, income, city, demographics, those sort of things.
00:49:38
Speaker
Politics could be a big source of revenue for you. Helping parties to decide where to campaign. Because parties spend so much money on election campaigning. If they got this intelligence that focused more efforts on this world, this city and so on, that could be a real game changer. Yeah, but who wants to deal with politicians at this point is the question.
00:50:00
Speaker
And that brings us to the end of this conversation. I want to ask you for a favor now. Did you like listening to this show? I'd love to hear your feedback about it. Do you have your own startup ideas? I'd love to hear them. Do you have questions for any of the guests that you heard about in this show? I'd love to get your questions and pass them on to the guests. Write to me at adatthepodium.in. That's adatthepodium.in.