Auren Hoffman (00:01):
Welcome to World of DaaS, a show for data enthusiasts. I'm your host Auren Hoffman, CEO of Safegraph. For more conversations, videos and transcripts, visit safegraph.com/podcasts.
Auren Hoffman (00:15):
All right, hello fellow data nerds. My guest today is Alex Dalyac. Alex is co-founder & CEO of Tractable AI. Tractable has raised a 120 million dollars of capital, and is the world's first computer vision unicorn for financial services. So, Alex, welcome to World of DaaS.
Alex Dalyac (00:32):
Thank you very much for having me.
Auren Hoffman (00:34):
You have all these images of car crashes, and things like that. How did you go about initially sourcing all of these images?
Alex Dalyac (00:42):
Well, one of the things that attracted us to this application of helping people recover from accidents faster with computer vision was first of all just how much data there actually is on this, and how much labeled data there is, which is key for supervised learning. For a couple of decades now, anytime you get into an accident with your car, the body shop is gonna take photos of the damage, write-up an estimate and submit it to your insurance company.
Auren Hoffman (01:05):
So, not on the crash site, but when you send it over to the body shop, they have to take a bunch of photos? Okay, got it.
Alex Dalyac (01:11):
Correct, that's a workload that's existed for a couple of decades. The key is that the people paying for the work, and the people doing the work are different, so you need a digitized audit trail, and that is actually a goldmine for artificial intelligence that was building itself up without people really knowing about it. For example, there are literally billions of images of damaged cars being produced every year, globally.
Auren Hoffman (01:30):
Alex Dalyac (01:30):
Auren Hoffman (01:31):
And that's owned by the insurance company, like who owns that data?
Alex Dalyac (01:35):
The images are always, by default, the property of the individual who took them, but then of course, based on the agreements that you have, there are various rights over who can use it, and for all intents and purposes, yes, insurers have the rights to be able to use those images, and send them to third-parties for help, and so that's our case.
Auren Hoffman (01:52):
Initially, you had an idea that you could do this, but maybe you didn't have the data. The data's owned by the insurance companies, how did you convince them to share that data with you, so that you could help them?
Alex Dalyac (02:04):
Yeah, it's always that cold start problem, right? It all started with a partnership with a software company that actually provides this software for body shops to write estimates and send them to their insurer. So that company's called Mitchell International, it's now more recently, uh, rebranded into a company called Enlyte, a very great partner of ours, but they basically for 50 years or more have been doing this, so they are sitting on terabytes upon terabytes of- of this data, and of course granted that you get the permissions from your customers, we were able to make a partnership happen where Tractable would provide the artificial intelligence know-how, which especially back then in 2015 was in very short demand.
Alex Dalyac (02:42):
There were just five or six labs that's practiced deep learning around the world, and so we brought that to the table, and they brought that data to the table, and so that- that really kick-started a dataset, uh, that was really, really large, in the nine figure of images, and then since then, once you have that, you're able to go to each insurer and say, "Look, you know, however much data you're- you're going to bring, it's gonna be small compared to what we already have, so if you want to benefit from the pool that we're bringing, you need to contribute to it as well," and by doing so we've been able to diversify that.
Auren Hoffman (03:11):
And do you think of it like a data co-op, everybody who contributes, your product gets better for all your customers?
Alex Dalyac (03:18):
Correct, of course. Every time it's a discussion with the customer. Some customers may decide, "You know what, I don't wanna contribute at all, and benefit from."
Auren Hoffman (03:26):
So there is a way, essentially to be a freeloader or-
Alex Dalyac (03:28):
There isn't, they'll say, "Look, I don't want others to benefit, and so I won't benefit, it'll just be my data," or there'll be cases where they'll say, "Look, I'm happy to contribute to a pool, and benefit from it, as long as it's not my competitors, or my top competitors," and so that's why the fact that we now operate in 15 countries helps, because actually the solution we offer in Japan, all of those sedans and pickup trucks that we've seen in America do help train the algorithm for the Japanese market.
Auren Hoffman (03:52):
... but then, I guess, at one point, like why would you even care, if it's gonna be a common good for everybody, why would, why- why would it really matter, are you really helping your competitor, or is it point 01 percent better, or something, over time?
Alex Dalyac (04:05):
Yes, you're right that over time it becomes less and less of a question.
Auren Hoffman (04:09):
Okay, interesting. You, one of the, one of the other companies in the insurance space obviously that's been around for a longtime, is Verisk. They're well-known for having a data co-op, and Scott Stephenson is their CEO was, talked about it on World of DaaS, do you think about your data co-op in a similar way that Verisk thinks about their data co-op?
Alex Dalyac (04:26):
In the sense of the data co-operative brought together creates a very valuable good that the whole customer base can benefit from, yes. In terms of the exact agreements that they have in place, I can't comment as to whether they're the same as ours? It's certain that the founding story of Verisk is remarkable, right? It, well, it started off with insurers pooling their data together, so that they had a single area to choose from, and then eventually that became a private company, which was Verisk. In our case, we didn't start off with that. We kinda built it as we went along.
Auren Hoffman (04:54):
If you look at these different data co-ops, there's many different ways to do it, one of them is the Verisk Visa way, where they're essentially, the co-op is owned by the companies, and then maybe eventually they become their own company, or something like that, but there're of course, there are many other, the Tractable way is also a way of building a co-op, there's many different ways of building a valuable data co-op. Your product's built on machine learning. The interesting things about all these companies that have machine learning is that there often has to some sort of human-in-the-loop to get it going from the get-go, like you can't get perfect right away.
Auren Hoffman (05:27):
How did that work, and how did the economics of that change over time too?
Alex Dalyac (05:30):
It's just like autonomous driving, right? You go from- from level zero to level five automation, and you go, uh, gradually every step of the way. You can't jump straight into level five, because actually those intermediary steps are really important to generate really important feedback data from humans, in order to further improve the algorithm. If you even think of Google Search, that's like a giant human-in-the-loop system, where if they've ranked the pages wrong, well humans will click and go, "Hang on, that fifth link is more relevant, and that's what I'm clicking on," Google says, "Thank you, human-in-the-loop," and- and updates its machine on the algorithm.
Alex Dalyac (06:01):
It's very important to have it. In our early days, that's one of the things we found out is, "Wait a second, yes, we're an AI company, it doesn't mean that right from the get-go, everything needs to be done by AI." On the contrary, we want to, and the customer wants to have that extra level of quality safety, and to include a human QA service on top, to make sure that the accuracy's always there. In your early stages, that's pretty significant. You know, you may even have human QA almost all the time, but as you keep going, and the years rack up, and your dataset racks up, then you're able to tone that down, and take it down to close to zero percent levels.
Auren Hoffman (06:36):
What type of humans do you need, do you need to like train these humans, or can they be mechanical turkers, or how do you like scale-up all those humans?
Alex Dalyac (06:44):
So, in our case, these are specialized tasks. You need to be an expert at appraising vehicles. These are people that work in-house with us, that we train to become experts at appraising vehicles. I just wanna kinda finish on a journey that we've been through in this case, today we are analyzing about two and a half billion dollars worth of vehicle repairs and purchases, that's how much we enable, well upward of- of 10,000 vehicles scanned a day. That would be completely impossible to manage with the team that we have. So, on that service, to give you a sense, I think 85 percent of our revenue comes from, uh, artificial intelligence that is 100 percent automated, without any human QA.
Alex Dalyac (07:22):
However, every time there's a brand new product, a brand new AI model, that's when the- the human QA starts off a bit high, and to your point, it does change the unit economics.
Auren Hoffman (07:31):
And before Tractable, an insurance company had to have a person look at the file to make some sort of determination on it, or was there some sort of heuristics that they used, based on if it hit all the heuristics they could put it through, and if it was unclear, maybe then they had a human look at it, or how'd it work beforehand?
Alex Dalyac (07:50):
You've got a few steps. The first step of, "Hey, is Tractable confident in this AI result, to send it right through without any QA," but then there's also, "Is the customer comfortable acting on that output, with any human QA on their end either," that's why we have kinda different product offerings. We have one product offering, which is the one that customers like to get started with, which says, rather, "Use your AI to tell us which cases we need to be looking at," and then we'll have humans look at those, so it's more of a prioritization tool, and then we have another one which is, "Okay, we can offer an incredible customer experience, if AI just- just does the whole thing itself in real time."
Auren Hoffman (08:23):
Because you can immediately give the customer, uh, a reimbursement or whatever it might be, that's coming, right?
Alex Dalyac (08:29):
Auren Hoffman (08:30):
Drop the time pretty quickly?
Alex Dalyac (08:31):
That's correct. One example application that we're quite delighted by at the moment, is yeah, allowing people to get their- their auto claim payout, or even a- actually their property, their home claim paid out immediately, uh, literally while they're on the phone, on that first phone call-
Auren Hoffman (08:47):
Yeah, it's amazing.
Alex Dalyac (08:48):
... or potentially even without any phone call at all, if they've, if they've registered their loss electronically. In order to do that, well, you've basically got to put yourself in the claim handler's shoes. You've got, this is the first time the customer's telling you about a loss, we say, "Are you at you home," or, "Are you by your car?" "Can you take imagery of the damage?" Those algorithms need to run in real time, you can't have a human-in-the-loop at that point, because you need real time to deliver results to be able to just wire that bank transfer.
Auren Hoffman (09:14):
So, imagine if there's like a fire or something like that, how important that would be for the customer to get that money as quickly as possible?
Alex Dalyac (09:22):
Exactly, so actually in October we went live in Japan, during Typhoon Mindulle, which was a- a pretty high category, uh, typhoon. When those hit, it basically literally leaves people without a proper roof over their heads, and the problem is it impacts an entire province of Japan, for example, and so you have a huge shortage of appraisers, and of repairers, at a time where you don't have a proper home anymore. In those cases, having to wait 50 days to get some financial relief from your insurer is not fun at all, and so yeah, in those moments, being able to come in and say, "Hey, as soon as you call us, today, we will have that financial relief sent to you," it's a truly delightful moment.
Auren Hoffman (10:00):
You mention Japan, I know you guys are in, obviously the US, I think you're in France, Poland, UK, you're in a bunch of different countries. I assume the algorithms have to be tweaked to different rules? There are some things that are both cultural, but there's also things that are legal in each different jurisdiction, is that right?
Alex Dalyac (10:17):
Yes, absolutely, absolutely. There's a lot of points you're touching on there. So, certainly every customer's gonna be different. You need to develop ways to be able to make your algorithms fit the data of- of your customer, or, you know, in a more, (laughs), business way, fit the standards, right, um, kinda replicate the appraisal standards desired by that insurer, in that country, on those kinds of vehicles or homes, and that's a data domain shift, and so you wanna minimize how much machine learning, applied machine learning effort is required in order to do that? I think that's actually potentially a very big, uh, space for applied machine learning, which is kinda developing tools to be able to do that really fast.
Alex Dalyac (10:55):
So, there's that piece, but then you're right, there's also legalities. So, for example, in some American states, a estimate of the repair cost has to be signed, authored by a licensed appraiser. That means however fast you make the- the customer experience with artificial intelligence, at some point, someone licensed has to validate that that's correct. You can still make this- this experience lightning fast, right, "I'm on the phone, here's the amount, we all agree," you have a backend process where just before actually wiring the funds, you have a licensed appraiser confirm, "This is all good," and, you know, you still have a delightful customer experience, but, yes, legalities do change and matter.
Auren Hoffman (11:33):
Most insurance companies only operate in one jurisdiction. There's not a ton of them that are massively cross-border, or something like that. It's not like Toyota which sells cars everywhere in the world, am I right about that, and if so, is there like a hard go-to-market problem? It's not like Toyota can bring you into all these different countries?
Alex Dalyac (11:52):
One thing that I find fascinating is why the largest American insurers are just in America, and it's actually the Europeans often that have gone worldwide.
Auren Hoffman (12:03):
So there are in multiple-
Alex Dalyac (12:04):
Auren Hoffman (12:04):
... country, okay, I got it, okay. Their home market isn't big enough or something?
Alex Dalyac (12:08):
It could very well be that. The other thing that I find fascinating is how, you don't really have general insurance in America as much as you do in Europe. So, State Farm is the largest property and casualty insurer, and so they'll do auto, home, life, but, you know, they won't do health, whereas in Europe, the largest insurer in the whole world is Axa, and that's a French insurer. They basically operate in every single line of insurance there is, whether that's medical, or that's business, commercial, personal, P&C, you name it. I think that actually makes it fascinating, because not only... and of course, they are global, and so not only can a relationship with Axa take you global, but it can also help you as an AI company expand from one domain to the next.
Alex Dalyac (12:48):
When we have a- a relationship with the CEO of Axa, UK, this is the person that is in the end, the buck stops with this person, in terms of making 15 percent of all the car repairs in the country, 15 percent of all of the home repairs in the country, 15 percent of all of the insured medical bills in the country, all of that rolls up to this person, and so if you have an artificial intelligence that can help understand damage to cars, or understand damage to homes, or understand a medical scan, a huge adoption moment is that same individual. I think, that's what makes it so fascinating to work with insurers, especially in Europe, because they really are this protective backbone of many sectors of the economy.
Auren Hoffman (13:28):
In the US, every single state regulates insurance differently. Many of them have actual elected officials. California has an insurance commissioner whose job it is to regulate insurance, and I believe Texas has a similar type of thing, and which is somewhat crazy, er, you move 20 minutes from one border to the other, and all of a sudden, you're in a completely different... as a start-up, I imagine this is very, very difficult. Are there companies to help you navigate all the different insurance rules and stuff like that, or do you need to build this in-house system to get you smart enough to- to make sure adhering to each specific state?
Alex Dalyac (14:08):
Certainly, especially when it comes to underwriting. If you have a new model of underwriting, and you wanna use a new characteristic, and you wanna incorporate that into your underwriting models, that's when it becomes super complex in the United States, because you need to get an okay from every department of insurance, and it's one person. One thing that we find really, really pleasing about our approach, is we don't actually discriminate by humans. We don't care about any personal or private information. We just want to know, "Show us your car," we don't need to know your name, your address, your face, none of that, and so as a result, we don't discriminate, that helps a huge amount.
Alex Dalyac (14:45):
We're also on the claim side, right, it's about, "Let's figure out how much it costs to get this repaired, and let's get it sent to you fast," not so much about, "How much I should charge you as a premium," and so for that purpose actually, one thing that is truly incredible about working in the United States, is the scale that you can attain, because that insurance company will have one Chief Claims Officer, and if they decide your solution is terrific, off you go, at full-scale, whereas in Europe, for example, it's gonna be tightly, uh, federated. You're gonna have a Chief Claims Officer for every one of those 44 European countries.
Auren Hoffman (15:15):
Now, when I was reading about Tractable, I heard a story that you initially had this customer that was, produced plastic pipe wells, or something, and it was like your only customer when you started, and it was the whole thing you're trying to find product-market fit, and then they end up leaving you, (laughs), while you were fundraising. What'd you learn from that experience, and how did that shape the company?
Alex Dalyac (15:37):
I find it really hopefully useful to the entrepreneur community to share stories like that, now that we are, you know, one of the 20 computer vision unicorns, because it shows that the path is never rosy, and there are extremely difficult moments that you go through, and that basically you just need to not give up, (laughs). And so, yeah, one of those rather unusual moments, (laughs), was we're in the middle of our fundraising process for our first round ever. We only have one customer, and they drop us, (laughs).
Alex Dalyac (16:02):
You would think you don't have a company left at that point, well, you know, thank goodness we had identified auto insurance, and- and investors didn't think it was so important to- to make a huge fuss about pipe well space, but yeah, I- I think the learning on that was, we should've definitely kinda managed that relationship a little better, maybe also that company thought they were just working with one dude, doing his postgraduate degree in university, and all of a sudden it was becoming this ambitious start-up, so maybe they didn't expect things to go in that direction, and they choose to find their next university student that didn't have these goals in mind.
Auren Hoffman (16:34):
Alex Dalyac (16:34):
But I really think the biggest lesson is- is to not give up.
Auren Hoffman (16:37):
Winning customers in insurance especially is like a really slow grind. Insurance companies are not known as companies that move very, very quickly. They're moving very, very deliberately, as we've mentioned before. They're very regulated. What tactic did you use to convince them to make a bet on you in your early growth stage?
Alex Dalyac (16:56):
It is grueling. There's one in particular, I love joking to this- this individual whose become a bit of a friendly acquaintance. I love referring to the first email that we ever exchanged, which was in 2015, and so, uh, I keep saying, "Yeah, it's- it's been, it's been six years." Actually we're- we're in the process of running our first ever proper proof of concept with them, so I- I keep, I keep joking to him-
Auren Hoffman (17:15):
Alex Dalyac (17:16):
... "We only get to run a POC with you once every six years," so-
Auren Hoffman (17:18):
(laughs), it better be good, yeah.
Alex Dalyac (17:21):
... it better be good. Good jokes around it, he's a terrific guy. Yes, it's, uh, it takes a long, long, long time, so how do you do it?
Alex Dalyac (17:27):
There is certainly one aspect which is wonderful, I think, is once you've signed a good number of them up, they're there to stay, so that- that's probably amazing, and I think, and that- that's part of why I hope Tractable will be a really, a long story that will, that will keep growing. We'll have this incredible base, customer base to support us, as we keep expanding. In the short run, getting the first one, one thing was certainly technological superiority. You know, you've gotta have a technology that nobody else has, that blows peoples minds, and so what we had was... there's actually a video available online, but what we would say is, "Look, we have these algorithms that can assess damage on a car, and you can test it yourself right now."
Alex Dalyac (18:05):
We would pull up the demo in the meeting room, and we would say, "Tell us what to type into Google, we'll type anything you like, tell us what to scroll down to, we'll take that image and we'll drop it," and that just created such a show. It was interactive. They'd say, "Oh, try that funny one with the bumper sticker, and try this one," and, "Oh, that one's got flood," and we'd drop it in and in the very early days it doesn't work absolutely every time, but it still truly impresses. Of course now it's, uh, extremely robust, so I think, that was important. They say, "Wow," and nobody else can produce that kind of- of technology-
Auren Hoffman (18:33):
It's unique, yeah.
Alex Dalyac (18:34):
... that's right, so there's that technological superiority that's proven in a kinda really striking way. Then, that plus in the end, across all the countries we work with, people like to be sold to by someone like that. And so if you're in France, you've gotta find someone French, if you're in America, you've gotta find someone American, even if you're in the Midwest, best to have someone whose Midwestern, if you're in, uh, in Texas, best to have a Texan, and so forth. So I even made, for example, the mistake of, um, me being European, of, I'd settled down in New York City, a bunch of New Yorkers would come over and we'd realize there's a big culture clash between New Yorkers and Midwesterners, so we eventually realized, "Okay, (laughs), you know, we gotta diversity the- the America team."
Alex Dalyac (19:11):
That's one piece that's very important, and then the third one, to be honest, is just utter relentlessness, because yes, it's very, very hard to make such large companies move. You've just gotta throw so much energy to try and start creating a bit of motion.
Auren Hoffman (19:26):
You know, Safegraph we sell data to insurance companies, and it's just really good advice to keep on, and- and really kinda push it as well. You're headed in also car insurance and doing car imagery, it's kinda like your core bread and butter. You're now headed into property insurance space as well. As you enter into like these other verticals, you know, how do you see these insurers combining with your technology and- and maybe also other alternative data to improve their claims management?
Alex Dalyac (19:54):
Auren Hoffman (19:54):
You know, for instance, with property insurance, maybe you really need really good weather data to really help you. Maybe you really need to understand what's in the soil, or something. Okay, well the roof's not there, but you need some sort of data about what the roof was made out of, and so you need old satellite imagery. I mean, how do you think about these other data sources? The roof isn't there anymore, how do you, how do you figure some of these things out?
Alex Dalyac (20:17):
(laughs), wow, you've, you've- you're just spot on. Uh, (laughs), you've basically named a few perfect data partnerships. So, yes, you need a complete solution. There's loads of other data that can be helpful. If you're looking at a car, we're looking outside in, it can help a lot to have connected vehicle data. So that's why one of our big forays at the moment is working with vehicle manufacturers. We actually hope to- to sign and announce a couple of them this year, but we think it'll be incredibly powerful to be complimenting connected vehicle data with computer vision from a smart phone, to have a complete assessment of that car, whether the car needs to be repaired, or just sold, just changed hands, without any kind of human assessment, or towing to any location needed.
Alex Dalyac (20:53):
So there's one there, which is connected vehicle data. The same thing with homes. Home are getting more and more connected, and that can be extremely helpful, for example, to identify, you know, water damage situations. Whe- when we're talking about- about hurricane damage recovery, then absolutely knowing the dimensions of that roof is so important, and that comes through aerial imagery.
Auren Hoffman (21:11):
Going back to the company, you founded Tractable, I think, right out of college, which is amazing. In terms of recruiting talent, what advice would you give other young founders who are trying to recruit people, who are likely more experienced than themself?
Alex Dalyac (21:24):
Yep, that was definitely one of the really tough things, and I was on a podcast with Reid Hoffman where I told him this was one of the things I'd found really hard. You feel really nervous when you're straight-out of college, thinking, "I can't hire someone that's basically, who I will be 10 years from now if I go into industry, that's crazy, that a person's like me in 10 years," surely they don't wanna work for me, and so yeah, you tend to just hire very junior people. I would say you've just gotta kinda force yourself to hire people that are older and more experienced. Maybe have an adviser, that's on your team, or an investor, kind of help to do that, and it's awkward the first time, but the more you do it, the easier it gets.
Alex Dalyac (21:57):
That's kind of what I found. One thing that Reid said, which I found fascinating is, "Look, if the mission of your company is one that you really believe it as being one of the most fulfilling ones in a lifetime, then you've got one of the best mission statements on the planet, regardless of how senior the other person that you're interviewing is, this is literally one of the best missions in the world for them to work on, so it doesn't matter how old you are," I thought that was terrific, and to me that's- that's really what start-ups around the world can learn from Silicon Valley.
Auren Hoffman (22:26):
You did have a co-founder Adrien Cohen who previously was the co-founder, I think, of Lazada, which was like a big Southeast Asian e-commerce giant acquired for Alibaba, for well over a billion dollars, how did you, how did you guys, how'd you convince him to work with you, because he presumably could've been like set for life, and living on a private island, or something, so how'd you get someone like that to held co-found the company with you?
Alex Dalyac (22:49):
Yeah, co- comment on Adrien's finances. I think, in this case, what really, really appealed to Adrien was the technology. Adrien is a- a businessman by background, and so when you work in- in e-commerce i- in 2012, '13, '14, '15, it's a huge moment for e-commerce, but it's a technology that been, er, with the internet, right, which had been around now for a bit of time, you're moving goods around which have been around for, uh, some bit of time, artificial intelligence, that is a very deep technology, a very complex technology that promises to revolutionize mankind this century.
Alex Dalyac (23:22):
And so, I think, for him it was an incredible way to- to get involved there, at a point in time where you could be co-founder, but right after our seed round, so it's a company with serious investors, and kind of a- a- a proven opportunity, so I think that's what did it.
Auren Hoffman (23:33):
Interesting, that's cool. You know, another story I heard is that you tried to get acquired right after you raised your Series A, and then it didn't pan out, which obviously I'm sure you're extremely thankful for now, what was kind of the motivation, how did you wo- work with your investors, and your board? What did you learn from that experience?
Alex Dalyac (23:52):
Again, I think, this is one of the other interesting stories to share to other entrepreneurs out there, because entrepreneurs out there might be thinking, "Oh, I've got this opportunity to exit, should I do it," you might just be passing on something enormous that you could be building if you kept going, and yeah, back then we thought we had nothing. This was... leading up to our Series A, the best contract and insurer we had was a 50,000 dollar pilot, that was it. That's what we had-
Alex Dalyac (24:16):
... a two million dollar round to get, plus that very important partnership with Mitchell. We thought, "This is terrible, like we can't generate revenue for this business, there's nothing, let's go sell to this tech giant." Yeah, thank God it didn't workout, (laughs). I think, um-
Auren Hoffman (24:29):
Alex Dalyac (24:29):
... I think, the lesson there is, it's really astounding to what extent, if you want something to become really big, it'll become really big if you really, really want it to, basically, and- and if you don't believe in it, then there's no way it'll ever get there. I think the other thing to be mindful of is getting advice from people who are a bit more opportunistic might bias you to basically taking like easy way out, as opposed to those who are truly passionate and have really peak ambition.
Auren Hoffman (24:53):
I know that you were initially rejected, the first time you applied for Entrepreneur First, which is a start-up incubator that you eventually attended, and by the way, full disclosure, I'm an investor in Entrepreneur First, and that's kinda how we met, what advice would you give others who may have been rejected for the first time, how do you keep trying, how do you know when to stop what you're doing, pivot, et cetera?
Alex Dalyac (25:14):
Certainly, Entrepreneur First, incredible incubator program. It changed my life. That first rejection is what changed my life, because, and maybe this'll help answer your question. I- I knew nothing about tech the first time I applied. I was an economics and mathematics undergrad. I'd done this, um, this work in e-commerce for a little bit, and then I had friends in a fashion school in London, and I thought, "Oh, let's help them crowdfund for, so that they could directly commercialize their fashion design, instead of joining a- a fashion house," so I knew nothing about tech, and so, EF said, "Well, we're only here to build tech companies, so sorry, but no thank you," (laughs).
Alex Dalyac (25:46):
Then there's one thing that Matt Clifford, the founder, CEO of EF said, which really stuck with me, which is, "Look, never before in time has it been possible for one person in their bedroom to build a product, and serve millions of people around the world," and that's the power of the programmable computer combined with the internet, that's the scale," and I thought, "Okay, that's, he makes a really good point now, this is the stuff of magic." I always had, up until then I'd seen software as frustratingly intangible, and say, "Is there such a thing as a software product, I can't touch it, I can't, you know, break it apart," it's, I didn't like that, but the moment Matt said that, it just transformed how I saw things.
Alex Dalyac (26:20):
I then went ahead and did a- a conversion course in computer science at Imperial, kind of really got into deep learning and image recognition, if the EF hadn't worked out, I would have gone to do a PhD, so that transition into science, it wouldn't have happened without that rejection, and so, I guess, the lesson to people is probably when something fails, it's- it's always tempting to kind of place the blame on something outside your control, "Oh, that incubator program, they don't get it, let me go find another one or something." I think you can improve yourself so much more by... well actually taking that blame and saying, "Okay, why was I not good enough, what do I need to do to improve, why did they say, 'No,' what would make them say, 'Yes.'"
Alex Dalyac (26:59):
I'm gonna go work on it, and so, "Okay, I'm gonna go learn computer science and be- be a machine learning scientist."
Auren Hoffman (27:03):
Okay, cool, I have heard rumors that you are actually a funk musician on the side, and you both sing and rap. What is the secret to kind of nurturing your creative side while you're building this unicorn, (laughs), and do you think one helps the other?
Alex Dalyac (27:18):
Yeah, yeah, yeah, that's really funny, your- your, you must've done some amusing research, because-
Auren Hoffman (27:23):
Alex Dalyac (27:23):
... during that Series A, where all we had was that 50K contract, we were staying at home in- in San Francisco, and yeah, just to kind of let off some steam, I just went and like, uh, I guess, rapped in front of the laptop, recorded it and put it on YouTube, and hoped nobody would find it, but, uh-
Auren Hoffman (27:37):
Alex Dalyac (27:37):
... there you go.
Auren Hoffman (27:39):
Do you ever like rap in your sales pitches, or anything, or, uh?
Alex Dalyac (27:42):
Okay, so- so, we had a- a really difficult moment in the United States for a bit, and then there were these enormous breakthroughs by one of the most famous American brands, you can look it up online, partnered with us. There were kind of these sudden massive breakthroughs in the US that we experienced last year. We actually, (laughs), myself and- and- and Julie Kheyfets, who- who runs North America, we rewrote Still D.R.E. from- from Snoop Dogg and Dr. Dre, to talk about the Tractable story. We have that whole script written out, and, uh, we have yet to perform it, so if you want-
Auren Hoffman (28:10):
Alex Dalyac (28:12):
... afterwards, we- we could put a little 10, 15-second clip where we actually do that, and incorporate it in this podcast.
Auren Hoffman (28:19):
Oh, that'd be awe- could you lay down, can you lay down a couple of beats now, or no?
Alex Dalyac (28:22):
Let me see, so I hope, let me see what have I got, (laughs)?
Auren Hoffman (28:25):
Alex Dalyac (28:25):
Okay, let's see.
Auren Hoffman (28:27):
All right here we go, first on World of DaaS, we're laying down some beats here.
Alex Dalyac (28:31):
I hope you've got Still D.R.E. on your mind, hopefully we'll- we'll add it in- in postproduction. And there's a moment in the beginning when Dr. Dre comes in and goes, "Oh for sure, check me out, okay." So, that bit, okay, we're gonna, we're gonna layer in some- some Tractable lyrics here, so he goes-
Auren Hoffman (28:46):
All right, let's do it.
Alex Dalyac (28:47):
... he goes, "Oh for sure, check me out, it's still AI brother, car claims brother, though we've grown a lot, now the US a lot, because when we flew the spots that we used to rock, we'd hear legal departments barge in and block, Japan, they pay homage, but America say we fell off, how brother, our last client was the farmers, they wanna know if we still got it? They say US a chain, they wanna know how we feel about it."
Auren Hoffman (29:14):
Oh man, this is good. All right. Well if, you know, if Tractable doesn't workout, (laughs), I think we got another career for you. This is pretty exciting. You can be on American Idol here, or- or the European version of that.
Alex Dalyac (29:26):
Auren Hoffman (29:26):
Alex Dalyac (29:27):
I, um, I'm interested in artificial intelligence, but it's fun-
Auren Hoffman (29:30):
Alex Dalyac (29:30):
... I think it, it's a, you know, we have, people often ask like, "What's the culture you wanna build," and to me the- the three values that come out are results, feedback and fun? So results is the ambition, feedback is exactly what Matt Clifford said, you know, "Go learn computer science," but when you place yourself super ambitious goals, and when you screw up, you say, "It's my fault," and you tell everybody it's your fault, you gotta have some fun, (laughs).
Auren Hoffman (29:52):
Alex Dalyac (29:53):
You gotta make people lay down some rap tunes when- when they want to.
Auren Hoffman (29:56):
We're gonna do some tunes for World of DaaS in the future. Last question we ask all of our guests, what is the conventional wisdom or advice that you think is generally bad advice?
Alex Dalyac (30:06):
So there- there's one that I find, uh, yeah, really funny is, "You can't build a company with a solution that's looking for problems to solve." That's a big one, and I would say, "Well, that's what we did with Tractable,"-
Auren Hoffman (30:16):
Alex Dalyac (30:18):
... and actually you could say pretty much any deep tech, cutting edge, cutting edge tech company does that, they start with cutting edge technology, and then they say, "Okay, we're gonna try and solve something with it."
Auren Hoffman (30:26):
Where do we go, what focus, okay, here you found, okay, well car insurance could be a focus of what we're doing?
Alex Dalyac (30:32):
That's right, and so I think you can, you can do it as long as you are, you take a very rigorous scientific approach to understanding, "What can this technology do better than anything else," and then being very curious and looking at a large number of use cases, so we'd looked at- at car insurance, we'd looked at dermatology, we'd look at natural resource, uh, exploration, uh, plants, you know, the list kinda really, really went on. Yeah, taking a very rigorous scientific approach to looking at, "Okay, if I apply this technology, can it actually be the best solution out there, better than anything else," and quantifying the impact it can have, looking at that market size, then I think you can do it.
Auren Hoffman (31:10):
All right, this is awesome. Alex tell us where we can find you on the broader Interwebs, (laughs).
Alex Dalyac (31:14):
That YouTube video, (laughs).
Auren Hoffman (31:16):
Alex Dalyac (31:17):
Do the rap, no, I'm kidding, yeah, LinkedIn, LinkedIn.
Auren Hoffman (31:19):
Alex Dalyac (31:20):
Could do, could do Twitter as well, although, um, I don't know, nobody seems to care, (laughs), on that side.
Auren Hoffman (31:25):
Oh man, I love Twitter.
Alex Dalyac (31:27):
You like that stuff.
Auren Hoffman (31:28):
Twitter is I- I, er-
Alex Dalyac (31:29):
I think Twitter's amazing, but I-
Auren Hoffman (31:30):
... oh my gosh, I love Twitter so much. Right, this has been awesome. Thank you again Alex for joining us at World of DaaS. It's been really fun.
Alex Dalyac (31:37):
Auren Hoffman (31:38):
Thanks for listening. If you enjoyed the show, consider rating this podcast and leaving a review. For more World of DaaS, and DaaS is D-A-A-S, you can subscribe on Spotify or Apple Podcast, or anywhere you get your podcast, and also checkout YouTube for videos. You can find me at Twitter at @auren, that's A-U-R-E-N, Auren, and we'd love to hear from you.
Alex Dalyac, co-founder and CEO of Tractable AI, joins World of DaaS host Auren Hoffman. Tractable offers artificial intelligence to help people repair, protect & sell their vehicle & home. Auren and Alex dive into the mechanics of building an AI product in insurance, how to build a successful data co-op, and the challenges startups face when going global. They also explore the journey of building a company and how to navigate the tough moments.
Bonus nugget: listen till the end to hear Alex show off his rapping skills.
Dean Stoecker, co-founder and former CEO of Alteryx, talks with World of DaaS host Auren Hoffman. Alteryx is a $5 billion market cap data analytics automation platform. Alteryx knew that digital transformation would be driven by workers in every functional area of a business, from Sales Ops to FP&A to supply chain, and not just the PhD-trained statisticians. So they built a platform that allowed for code-free and code-friendly approaches to the entire analytics continuum.
Auren and Dean explore the rise of the citizen data scientist, how Alteryx defined the analytics space, and how they built a product that supported the entire analytics continuum. They also dive into why Alteryx invested in long-term strategies, including channel partnership and a user community, to slowly yet effectively strengthen their business.
Henry Schuck, CEO of ZoomInfo (NASDAQ: ZI), talks with World of DaaS host Auren Hoffman. ZoomInfo is a $17 billion market cap B2B data company and one of the only billion dollar data companies that have been created in the last 20 years. Auren and Henry cover ZoomInfo’s origination, how it differentiated itself in an already crowded market, best practices on building a contributory network, ZoomInfo's unique acquisition strategy, and more.
Will Marshall, co-founder and CEO of Planet, talks with World of DaaS host Auren Hoffman. Prior to Planet, Will served as a Scientist at NASA/USRA. Planet is a data business that operates hundreds of orbiting satellites that capture images of the entire world daily. Will and Auren cover Planet’s proprietary methods of building a satellite imagery dataset, how its data is used by numerous industries, Planet’s plan to go public later this year, and more.