Ramneek Gupta, managing director at Citi Ventures, thinks companies will need to safeguard against unforeseen consequences of AI adoption.
In financial services, technological change tends to move slow. And that is certain to be the case when it comes to AI adoption, according to Ramneek Gupta, managing director at Citi Ventures.
In this highly regulated industry, which holds some of the most sensitive consumer data, it’s paramount to safeguard against potentially unforeseen consequences of rapidly advancing technology, Gupta said, speaking at CB Insights’ Innovation Summit.
“Slow [adoption of AI] is a not a bad thing, it’s absolutely important to be slow, especially in financial services.”
Gupta was joined by Bharath Kadaba, Chief Innovation Officer at Intuit and John Fawcett, CEO and founder of Quantopian, in a panel discussion moderated by Robin Wigglesworth of the Financial Times.
Gupta underlined the sentiment that although progress in AI is inevitable, slow progress is best because it safeguards against potential unforeseen consequences of rapidly advancing technology.
“It’s about being extremely careful about the data you have. We are stewarding our consumer’s data, and they own it,” Kadaba said.
Kadaba highlighted how important it was for the financial services industry to carefully integrate AI into its operations at a scale that safeguards against unintended consequences that could affect individuals as well as the global economy.
Gupta was confident though that AI would find its way into financial services. “It has to happen, no two ways about it, given cost structures and productivity needs.”
But it likely will first be implemented for technology like automated investing, as in Quantopian’s development of algorithmic investing. Developing user-friendly interfaces with AI was seen as being a bit further in the distance. “Areas where issues become hard, are when humans ask questions back, and we need intelligent responses,” said Kadaba.
Robin Wigglesworth, Financial Times: Hello everybody. Thanks for having a bit of patience with me. So we’ve got another cracking panel here. So I had a chat with all of them already, so I think it’s gonna be quite interesting, and this happens to be also sort of more home turf for me as well. I started getting obsessed with this a few years ago, and actually had a really interesting lunch with Fawcett two years ago just before my second kid was born and I started getting more interested in that than the incoming birth…
John Fawcett, Founder and CEO, Quantopian: Glad you’re on the record with that now.
Robin: Yeah, no, no, my wife knows. It was actually when I was on paternity leave… That’s when she realized that this was…
John: True love.
Robin: Yeah, and there’s obviously a lots of interesting things going on in financial services, artificial intelligence in general, and financial services industry is not immune from that. It’d be interesting to hear whether you think this is just an incremental new hack or whether this is actually genuinely transformational. To what extent, how excited should we be about this? I don’t know. Bharath, do you wanna kick us off?
Bharath Kadaba, Chief Innovation Officer, Intuit: Absolutely. So as you know AI and machine learning has been around for a very long time, and so the recent interest I think is really at a different level than it has been in the past, and the fundamental reason of course is around the compute power that’s there and the amount of data that’s available that makes machine learning to get to that next level, and you’re seeing the evidence of that of course in a lot of the applications that we’re seeing coming out in the market. So we think that in financial services, all of that technology capability will in fact take us to that next level, and I think it’s a breakthrough as opposed to an incremental change.
Robin: All right. Ramneek.
Ramneek Gupta, venture investing, Citi: I’ll add to that. I sort of agree, but sort of disagree as well. There are certain parts of financial services where I think we will see that happen, and certain others, I think it’s a long way off. So depends on your time frame. Consumer facing stuff, I think it’s a little way off. I think about defense, intelligence data, health care data, and financial services data. They’re very carefully guarded, kept, handled, and bad things happen when you let things loose on them, right? And so getting those things in front of consumers is very…it’s gonna be a long way. Whereas backend stuff like fraud, cybersecurity, anti-money laundering, all these sort of things, I feel like in a human assisted fashion, not complete AI, there’s gonna be a lot of application. It’s already happening.
John: I think it’s very uneven, but also it’s uneven in different markets. So to me, the most exciting stuff is happening in China, partially because it’s a different regulatory environment, but if you look at payments, the two biggest payment processors in China are Tencent and Alibaba, and they’re able to extend credit and they’re extending credit not on the basis of some standardized credit score, but really on an AI that’s studying the behavior of all these people. Which I think is remarkable.
I actually, I think that there’s a lot that been embedded in the consumer services that are out there, whether it be payment or life insurance, that’s another very large one. I think it’s actually starting to really affect businesses. I still think it’s early, but I think we’re seeing very real effects on every single area of business, and it’s a little bit uneven.
Robin: But isn’t there an element of, I mean, look, I love this subject, it’s fascinating, but just today, I got a press release from a company that is doing some really cool stuff in a really interesting area that really needs some tech help, but I noticed now they’ve just jammed in machine learning there rather arbitrarily, and as far as I understand what they’re doing is not machine learning.
John: Right, that’s like the funding cycle coming through in that.
Robin: Yeah, exactly, and it just feels a little bit like we’re trying to assign machine learning and whatever it’s gonna be on almost anything to do with financial services. And do you guys worry that people are getting sold some genuine medicine and some snake oil as well, and that the snake oil is gonna come back to haunt the financial services industry?
Bharath: Well, I think the way we think about it is the following, right. So how many of you hate doing taxes? Only a few. I am surprised by that by the way. I hate doing taxes. So we have set a vision in the company that says that taxes are done. So we want to make sure that when you are ready to do your taxes, it’s actually already done for you and all you need to do is go verify what we need to do and get that done. So it needs to take a few seconds or few minutes to go do that. Today, as you know, it takes several hours for you to actually go do that, and how do you take that several hours to a few minutes is a vision that we have set.
To Ramneek’s point, it’s not an instantaneous thing, we can we wish that and it’s gonna happen instantaneously, because the problem is very hard to solve. However, we think that with the huge amount of data that we have, we process 34 million tax returns, and we have multiple years worth of data. So we can actually look at that and figure out what are the patterns, and how do we pre-fill, and then we’re also gathering documents from every digital footprint place that we can possibly get. So gather all the documents, figure out what kind of category you belong to, because I can categorize you into category of this is you’re a nurse in San Francisco and therefore this is how it’s gonna look like, and try to do most of the work for you. So we’re executing against that vision. We know it’s gonna take a few years, but we’re gonna get there. And that’s why I think, because if the technology is there, we can apply it, we have to do it methodically, but we’re gonna get there.
John: I think the key thing that you just described that in my mind I used to separate reality from hype is, the companies that really has significant advantages or doing really breakthrough work have a compounding data set. So like the tax returns data set that you have is a naturally compounding data set, the methods that you used to analyze are gonna improve as the data set grows.
So companies have a core asset of that nature, like classic example being the self-driving car data sets accumulated by driving fleets. So if you are really building a core data asset like that then I think it enables very innovative work and application of these techniques that are just taking off because they’re all extremely data hungry.
Robin: Yeah, I mean, just on Intuit, I mean, having done taxes in Norway, the UK, France, United Arab Emirates, and the U.S…
John: Which was your least favorite?
Robin: It’s the U.S.
Bharath: The U.S., has to be.
Robin: I swear to God, all your human rights are being violated in the U.S. That is torture. I mean, if you can sold that I’ll give you the Nobel Prize right here. I’ll put the Norwegians…
John: Put a word in.
Robin: Well, yeah, put a word in. What role does machine and artificial intelligence in general play in the Intuit’s…first of all, the actual processes and what you do, and how do you scale that one?
Bharath: So the best way to do that is to give you an example. So last year what we did was we did two things. Using machine learning, we were able to say that okay if you’re a…again I take the example of a nurse in San Francisco, suppose the nurse is starting to enter the W-2 data and enters $600,000 as her income. What we do is based on the fact that we understand nurses in San Francisco and what the average income is, we flag that and say, “Hey, maybe you wanna double check the data that you just entered just to make sure that you have not gotten that $600,000 wrong.” So trying to detect anomaly automatically and help suggest the ways to make sure that you actually get that corrected is one example of how we could do that.
The second one is lot of people actually go through standard deduction versus itemized deduction. So you go throughout the whole process of filling it out and only to discover that actually it’s not a good idea to go do that and do something else. You may have spent 40 minutes to actually go do that.
What we do is again, with machine learning, we can analyze your profile and predict what are the chance of you being in an itemized deduction or a standardized. So we just suggest that to you so you avoid that 40 minutes of tax prep time. Our two examples of how we’re starting to bring this…and the idea there ultimately of course is every entry that you either do you figure out how to enter the data automatically or you figure out how to help them do that fast. Make sense?
Robin: I’m actually Norwegian despite the silly English name and silly English accent. We get pre-filled out tax reforms, basically tax reforms that the government scrapes every financial data, all the banking system, all the real estate registries, they scrape that, put it all together and give you your tax returns and say, “If you see anything wrong then you let us know.” Otherwise, it’s all done and dusted. And you sometimes have to fix things. I used to think that was a nightmare. Compared to the system here it’s magic, but is that something you could do in the U.S. with a system like that that we could pre-populate the forms and use, for example, if you’re a nurse in San Francisco…
Bharath: So we actually don’t think in terms of forms already. We have an interview process, right. We never show you the forms, we just ask you a bunch of questions. The idea is to drive the number of questions down to zero. That would be the way to think about it. So you just don’t need to even know that there are these forms sitting behind what it is that you’re trying to do. The government actually figures it out. Now there’s one other interesting thing in the U.S. which is really around, I own my tax prep, and I don’t want the government to actually do my taxes is another sentiment we have in the U.S. which may not…
John: Sounds profoundly un-American to have the government send you your tax returns.
Bharath: Exactly, “Hey, don’t get into my thing,” right?
Robin: I know that headache they will save you once a year. I bet Ramneek, I mean you mentioned that you think this will affect different industries at different paces. Some are more advanced and some are far behind than others. If we dwell a little bit, I mean, how does for example something like wealth management, how does that fit in here? Which is consumer facing, but can be fairly data intensive, it’s also had these compounding data sets that presumably the more people you have on platform the better you’ll work.
Ramneek: I think wealth management is a unique situation in the sense that you can be a little bit off and yet nobody will know for the long time. So it’s a little bit more shall we say open an area to test things out than a few other areas in financial services. So I do think it’s more fertile that others.
Now, how do you bring artificial intelligence in there? I think a lot of the robo advisors are taking the first step there. That’s very heartening to see. It brings down cost, it brings down hassles, and it hopefully in the long run improves returns for the majority of the population. So we do see some movement in that direction, and also because it is a little bit easier to do. There are certain things you can’t. You absolutely have to settle then there and if there is a discrepancy, it is very difficult to take it to market that way. Of course, you know a lot more about this, John.
John: Yeah, so for me, that’s my area, so I think of it as a little more striated I guess, but the trading infrastructure, is highly automated and there’s quite a bit of artificial intelligences deployed to analyze the order book. So the way that the exchanges are working and the way the market makers operate, it’s close to saturation I would say. That’s like a 10 year old arms race that’s been running for a long time.
Robin: When I’m talking to a high frequency trader about do they use machine learning, they look at me like I’m blabbering moron.
John: And we also use computers.
Robin: Yeah, exactly, yeah.
John: And I think the other end of the spectrum like advisory, really the problem there is a combination of marketing and making products match your customers, which is a fertile area for AI, but it’s a very different application. And then in the middle, sort of the trend in the industry today is really strongly toward in particular for retail, toward passive indexing, and that’s a problem when there’s a lot of automation but not a ton of artificial intelligence. That’s sort of a different type of problem. So I think what is really untouched and pretty interesting is that middle ground about creating the product. So Quantopian’s creating an institutional product, but applying artificial intelligence to the construction of our portfolio of investments.
Robin: So you use artificial intelligence to pluck and identify the promising algorithms that would work in the hedge fund side of it?
John: Exactly. So we run free platform and people can come in and create algos. Those algos individually can run the gamut, I mean they can be traditional hypothesis driven investments or they can be more machine learning style investment styles, but they all operate independently. Our job is to go through a very large database of these strategies and choose the ones that we wanna have in our products. So the ones that we think are producing alpha for our hedge fund product.
That process, the selection process is really the place that’s where we’re applying AI directly. So we have an algo to select the algos, and that’s the algo.
Robin: One algo to rule them all.
John: And in the darkness bind them. Yeah, so I think the construction of the products, the investment process is stubbornly manual. It’s shocking how little automation there is the construction of portfolios, and so I think that’s where the really interesting disruptive stuff will happen for AI within investment management. Trading, if you talk to professional traders, I mean, it’s happened, right? They’re all basically CS PhDs now already. And I think the robo trend’s really, really interesting for advisory, but it hasn’t happened yet really for investment.
Robin: How’s the hedge fund doing?
John: Can’t tell you. You know I can’t tell you.
Robin: What areas is artificial intelligence…so obviously, yeah. So high frequency trading, you sort of add analysis to the order, but what’s gonna happen to the next wave? We’ve seen a lot of artificial intelligence predictive technology put to use. What other areas have we come furthest in where we’re using this? I mean, Bharath, it sounds like you guys use quite a lot of this. But are there any other areas that you know that I might be aware of how advanced we’ve become?
Bharath: Earlier today on the panel when we talked about chatbots and AI, I think the area where the problems get really hard is when humans ask question and we now have to come back with intelligent responses. The way software is build today is largely built in a very rigid way, and it’s hardly configurable and now you have to think about what the questions that people will ask, and how do we navigate through the software and actually produce that intelligent answer.
For example, today if you say, “Do I have enough money to go out to go to a restaurant?” If you are using Mint which is one of our products, you have to go through multiple screens to actually look at your budget and figure out where you are etc, etc. It’s a process that many actually people want to avoid, and they just go out to the restaurant. Now, if it’s an intelligent system, you just ask the question, “Hey, can I afford to go out to this restaurant tonight?” And it says, “That restaurant, usually you’re gonna end up with a $300 bill because you’re taking your girlfriend etc, etc. And I don’t think you should be doing it.”
John: The wine list is a little too good for you now.
Bharath: Exactly. “Why don’t you stick to the leftovers in the refrigerator” might be the answer it might provide. To get to that level of sophistication I think we are far off.
John: The thing I always wonder about with this, so I always think of the automatic elevators where you tell you tell the floor and tells you the elevator to get on. I don’t know if anybody else has this, but I have the phantom limb problem where I really just wanna press the floor button. So I always wonder what the user interface is, are people gonna like them. So like eventually, if you extrapolate this, you get in a car and it just won’t take you there. You’re like, “No, I really want to go to the restaurant,” it’s like, “You really shouldn’t. I think it’s better if you stay home.” So I think the other thing I always think of is the two Google home devices talking to each other, which is magnificent.
I think the AI UX problem is fascinating and there hasn’t…to your point, behind the scenes creating the product, creating the services have really come a long way, but who knows how to really…
Ramneek: Yeah, that’s why we’re seeing more adoption in the behind the scenes part. The UX is the most difficult part of sulfur and will take a long time.
John: There’s also like an uncanny valley.
Ramneek: Uncanny valley, exactly, and in the background and in the backend, massive unsolved problems that I think if we let a bunch of Google PhDs at it, I think we’ll solve a lot better than… Anti-money laundering, take that for example. At 20,000, 30,000 transactions that are processed on the credit card and debit card infrastructure, you have to do so many things to figure out if something is a bad transaction. And people are getting so good at hiding that you have to go across multiple systems, you have to find sophisticated patterns, these sort of things which, it’s impossible for humans to analyze, right? And so we are just forced, I mean it’s not even whether we want to. It is the only way to solve for some of these challenges and fortunately we don’t have UI UX situation there, and we do that in a human assisted fashion obviously. The first line of defense is these sort of systems and then they throw out a bunch of exceptions and then you handle those with humans.
John: The same is true in investment. The really most sophisticated investment strategies that exist are AI driven and couldn’t be created by a human or a team of humans, there’s just massive amounts of data and it’s extremely dynamic.
Ramneek: So it’s not an option anymore. That’s where we’ve taken huge strides in creating in all the rec tech stuff, compliance stuff, and the backend anti-money laundering, fraud detection, all these sort of things. But the front end, ask how many questions, “Take me there. Can I do this? Can I do that?” That’s a very difficult problem.
Bharath: So I think to also underscore things like fraud, we miss a lot in Intuit because that’s a problem that we must solve and we’re solving it, and that’s something that we can do. Similarly, tax prep, making it easy. a lot of the backend heavy lifting is happening in the backend, we can do.
What my group usually does is focus on some of these harder ones. So we are experimenting with chatbots and those kinds of things where we don’t quite know how it’s gonna play out. So we build a ton of experiments, we try them out with customers, continue to learn and see where the industry is going to make sure that we can take advantage of the latest and the greatest breakthroughs that happen with respect to building these things. That’s the approach that we’re taking to make sure that we’re covering the field, know what problems we can solve today, and which ones will have to rely on the industry itself to evolve over time to make sure that we can take advantage of it.
John: Fraud’s a great example because it makes me think of the regulatory values that regulate the investment industry and there is very uneven development between the industry and then the regulators that are supposed to be overseeing. I’m talking about an area that would be ripe for AI applications.
Robin: I talked to the SEC and they’re using quite a bit like this. At the SEC, I would say RoboCop programmed their fraud detection.
John: Is that what we want?
Robin: But it works. They’ve actually spotted quite a few…it’s about flagging certain things rather identifying every single piece of a fraud, but it’s a good sort of an early warning system.
Ramneek: Add one more thing, this is a wonderful area for big incumbents to partner with startups. As I think we heard earlier, a lot of areas and especially in financial services, 2015 more than 2014 was we just take out the incumbents. And now it’s more about how do we partner with the right incumbents because when you’re applying machine learning, when you’re applying AI, it’s just not the algorithms. It’s the data, it’s the systems, it’s in the insights, and all these things are never found in one place with one organization and with one set of people. So you by definition have to partner and create an ecosystem and everyone’s learning into it, and we are…everyone’s very open to it now from both sides, which is great to see from my vantage point.
John: I actually think that’s the big disadvantage for the U.S. versus Chinese based companies because there it’s very, very integrated. There was no payment infrastructure, and Tencent and Alibaba created it. And now they’re offering… There’s no incumbent investment advisory business there, so there’s a huge opportunity for them in that area too. So I think there’s the competitive dynamic is favoring the integrated model.
Robin: So China is gonna rule us all basically in AI?
John: I just think it’s a big advantage. They’re under one roof, they have all the data, they have access to hundreds of millions of users. It’s interesting to see it unfolding.
Robin: I was talking to somebody at MIT about this and she’s done lots of work on recidivism and machine learning to predict the chance of recidivism. This is fantastically useful and helpful, but essentially we are reducing a person’s propensity towards crime to numbers, and there are some issues around that, and there’s almost inevitably going to be a bit of a backlash against the reign of the algos, and we can already see some effects of it, especially in financial services because it affects people in very severe way, and same with health. But interesting to hear you thoughts around that whether we should be more…because one thing is a AI-powered fraud detection system, but then there’s algos that say this is how you should invest, and sometimes they might screw up, and then have quite severe repercussions for somebody’s life.
John: In the markets, this is the continual headliners like flash crash, but actually the very scary event for the institutional set of investors was in the summer of ’07 when there was the quant… Yeah, exactly, melt down. So in the summer of ’07 there were several quant hedge funds that had very similar portfolios, but they also had very similar risk controls and so there was a fund that unwound and that created this cascade of selling pressure, like knife edge return profile in the space of a day. So that really put the fear of AI into institutional investors, and what I’ve noticed recently as we’re meeting investors is a lot of questions about what is the composition of the algos that we’re providing, are they hypothesis driven? Are they true like blackbox ML? And when I think about it, what’s interesting about it is they are most interested in ML and AI style algorithms right now. So I feel like within investment we’ve been through that cycle once. Probably not the last time, but…
Robin: Certainly not.
John: Yeah, you know, it only took 10 years to recover from ’07.
Robin: It’s interesting people still talk global alpha and there’s a bit of a, something to be wary of.
John: I think it’s just what scares people about AI, like you don’t know what it’s doing if it’s truly black box AI or…
Robin: So your approach or your algorithms are just finding exactly the same patterns and everybody else and this is hidden crowding and then we’re just doing that again. The approaches aren’t radically different from one firm to the other.
John: There’s that. I think to me the most interesting thing is that the risk controls are the same and that really accelerated all of the selling pressure. But yeah, I think it’s institutional investors are allocators, they’re charged with the allocating the money to managers, and part of the history of that profession is I need to understand your investment process so that I can be comfortable with it, then fulfill my fiduciary responsibility to give you this money to invest and you can’t….we’re talking about investment strategies that by definition are more complex than a human should be able to understand, and also if you’re also using like a neural network like well, do we understand why it’s doing what it’s doing? Not really.
Ramneek: And in lending is the same thing. I mean, consumer lending fortunately I think you can’t put these things into practice at least without proving a lot of things… There are unintended consequences, things like that. So there is…that is why I say some of these things will take a long time to get into practice and that’s not a bad thing. A lot of things have to be figured out about the unintended situations, the chronic cases where everything goes in one direction, there’s a crash and this and that.
John: Social implications.
Rmaneek: Social implications. So slow is not a bad thing is what I wanna underscore here. It is absolutely important to be slow, especially in financial services.
Robin: Do you think we are slow? As species we’re not very good about moving cautiously forward with new stuff. We tend to sort of, “Oh my God this is the coolest thing I’ve ever seen, let’s all do it. And oh, oops, that blew off in our faces,” especially in financial services. Even good ideas tend to be taken to the extreme.
Ramneek: Little less I think because you get excited but you can’t put it in practice at scale. Yes, a few startups here and there would come up with a new product that may or may not be completely compliant with everything out there, and yes a few early adopters will run to it, but at scale you have to get to at least compliance with all the regulatory infrastructure. So I think it’s….good or bad, we’re a little bit protected and insulated there.
Robin: Bharath, then do you worry at all about some of your systems? Even a small mistake replicated across a huge client base and compounded over time could lead to some pretty bad headlines or bad situation.
Bharath: I think that, so going slowly to that point, it’s about being extremely careful about the data that you have because we believe it is consumers’ data that they own and we are only stewarding that. And what we do with the data and then the care that you take where you don’t want to get your taxes wrong, for example. So the amount of work that we do to test whatever software that we build is huge and then the fraud piece, which is we got to make sure that we protect our consumers. So there’s a ton of conservatism that goes within, for example, in Intuit with respect to how we approach these things. So that’s why I’m not predicting that hey, one or two years we’re gonna get all of these done. It’s really more about being very cautious about how we do that because we got to get it right.
John: So our investment process is fully automated. So we have to confront this every day. Do we feel comfortable with the algos going out in trading. I think it does drive a lot of process, the whole selection of these algos is driven to meet that need. But at the end we turn them on and run them. So we are running just a fully automated process in that regard.
Bharath: Again, and I think the panel in the morning talked a little bit about when you do this predict analytics, what is the risk that you are taking? Is it a life or death prediction?
John: And what’s the amount of leverage? Technology brings leverage. How much…
Robin: I mean, humans screw up all the time. I mean, there was an interesting paper that came out a couple of years ago about algorithm aversion. So when humans saw an algorithm screw up once they trusted it far less even when they saw the human screw up twice as much. We just have immense skepticism once an algo makes one tiny mistake, it’s like yeah, it’s done. But humans, oh well, bad weather or whatever. We blame something, which I thought was interesting. Do you worry there will be some sort of backlash? There will be some sort of problem? Or is it gonna be the bit like, somebody talked about privacy earlier. We say privacy a big thing and worried about that, and there’ll be occasional backlashes, but then fundamentally we’re just moving into a world where nobody really cares. We give away tons of… Some people care, but most people on average do give away far more themselves than any other generation has before.
Ramneek: I think it’s gonna be a lot of, meaning it is gonna find its way and there will be an occasional things where you drop the ball and everyone gets freaked out about it, but it has to happen. There is just no two ways about it, you cannot… The cost structures, the productivity needs. Everything else that’s going in that direction then ask for convenience, this, that and the other, you will have slight bumps but it has to happen. It’ll be slower than in photosharing for example.
Robin: I thought actually two minutes, we’ll end ahead of time so that we have more time for your questions. Do we have any question from the back there?
Cameron McCurdy, CB Insights, Host: Yeah, we do. So the first question is for John. What’s the competitive advantage for Quantopian and what stops competitors from matching payouts? How do you attract the best quants to your platform?
John: Sure. There’s a set of things but I think the first thing is our philosophy. So our philosophy is that the shortage within investment management, quantitative investment management is the talent. There’s a huge amount of work to be done taking all the data that’s coming from the real economy and using that to discern price and there is simply not enough professionals doing it today. And if you look at all the incumbents, they send every large research departments but they top out at 150 or 200 people, and so the point of Quantopian is to coordinate the effort of a much larger research group. So we have 100,000 people in our community using the platform to do research and create algorithms, and I think there is a couple of reasons that we’re able to attract the best and convince them to put their best ideas on Quantopian and then hold on to them.
Probably the most important is that we let them own their own IP. So Quantopian doesn’t lay claim to the IP. We license it from them, so the author owns their IP the whole way through. That one fact is completely different from every other financial institution or investment manager. So the fact that the quant is independent, owns the IP and has controls over it, I think differentiates us and is a structural difference that’s difficult to replicate. And then secondly we…the structure of our agreements with the authors, they’re paid based on the performance of their algorithm alone. So in the industry there’s notion of netting risk. So you’re inside of a fund with a lot of other managers, the other managers do poorly, you do well, you still may not get paid because you have to net all these things together. So we’ve structured the fund that the people are paid independently.
I think those two things are really to scaling, and then we have a compounding data set. The database of authors and algorithms that we have accumulated. We have more information and as a result better techniques for selection and that’s really the key to our competitive advantage in terms of the fund products.
Cameron: Great, thanks. So the next question is for across the panel. There’s been a lot of talk about the use of alternative data in investing. What is an interesting alternative data set that you use and are there any untapped opportunities or lower utilized data sets that are out there?
John: Did he say that’s to me again?
Robin: No, to across the panel, [crosstalk].
Robin: I’ll tell you what I found really cool is email receipts. Cookies that are in our email inboxes see if you get an email from Amazon. They scan for the receipt into a real time database of spending patterns.
John: That’s also a bit of a holy grail because getting like SKU level data in retail is very, very difficult. I think…this is a topic that I think about a lot. So we have lots of interesting data sets that our users use to generate signals, and it’s everything from sentiment gleaned from online sources, transaction data, credit card panels. That’s the type of data that is really, really signal rich.
I think the thing that’s most interesting about this topic to me is what should the rules be for using this data, because we’re all walking around with devices in our pockets that are spewing out data about how habits and our activities and everything like that. So from an investment perspective, one of the most exciting for investment industry right now is the emergence of all this data directly from the economy. So like location based data. The retail companies themselves wanna buy that data in order to understand their marketing spend and the demand that they’ll experience from every locations. And then investors also wanting to buy that data in order to make predictions about revenue. So it’s incredibly predictive data sets. And I would say we’re probably in adolescent phase in terms of not so much generating usable results, because I think that’s not well trod ground, but people are doing it, but understanding the implications for the markets and for individuals by doing it.
Bharath: I almost think that’s not alternative data though.
John: Core data now?
Bharath: That is unavailable data that is becoming available. There’s nothing alternative about the the spend button, correlating that to what would that particular company, do and don’ts of its revenue and sales. I do see alternative data being brought up in other areas, where are people starting to correlate completely uncorrelated or unthought-of correlated data. For example, we think about that a lot in security, cybersecurity. There are a lot of uncorrelated datas being used to make predictions.
Robin: Like the S&P 500 is closely correlated to butter production in Bangladesh. Or Nic Cage films is correlated to swimming pool deaths in the U.S.
Bharath: That’s right. That’s always a challenge, and yeah, you can go crazy on finding correlations that don’t mean anything. But to me, in lending for example and underwriting, people are starting to look at a lot of alternative data that was not talked about at all, even in the neighborhood of lending. So what kind of car you drive and things… They’re starting to go really far out.
It’s unproven if that’s actually predictive at all. So in lending, you’re not seeing any real impact of using the color of the leather seats in your car to whether you’ll pay back your…
Ramneek: The most interesting piece of work along those lines is the work at MIT Media Lab. So it’s around…this is a work done by Professor Sandy Pentland over the last 10 years. So this is about tracking your movement data, cellphone movement data, and then use that data to predict a bunch of stuff that you would think is not at all correlated with anything else. There’s actually a startup that spun out of these guys in Israel that’s actually trying to market this idea. I think they’re called Endor. So it may be a useful company to check out on alternative data.