Host: Blair Zhu, Mint Ventures
Special Guest: Nick Emmons, Co-founder and CEO of Upshot
Youtube: WEB3 Founders Real Talk with Upshot
Podcast: WEB3 Founders Real Talk with Upshot
Blair: Hi everyone, welcome back to Web3 Founders Real Talk. We’re here to bring you captivating conversations with some movers and shakers of Web3 industries. Today, I’m really excited to take everyone on this journey of discovery and inspiration with the CEO and Co-founder of Upshot, Nick. Welcome to the show, Nick.
Nick: Hi, thanks for having me.
Contents
Background Introduction of Upshot
Blair: Would you mind telling us a little bit about yourself and your project？
Nick: Sure, so I’m the co-founder and CEO of Upshot. In the line, we build financial infrastructure for NFTs. A lot of this is built on top of the ML-powered NFT pricing models we’ve built, and other ML-powered financial tools or insights that we’ve built for creating more expressive or might mature NFT financial markets.
Blair: Would you mind telling us a little bit about your background？ How did you get in (crypto)？
Nick: Yeah, so prior to Upshot, I was leading blockchain development at one of the largest insurance companies and asset managers in the U.S. and Canada called John Hancock, and another company called Manulife. There, we were doing a lot of research around building systems for decentralized insurance for non-parametric long-tailed risk. So, how can we accurately price risk in these really long-tail scenarios? How could we pool risk in kind of risk-mitigated ways? And how can we verify the outcomes of different insurance claims? It was through that research that I became interested in the broader problem of bringing liquidity to illiquid markets. So, it led nicely into some of the interests that have led to Upshot, I guess.
Blair: Yeah, well, it sounds like you’re a perfect fit for the crypto industry because I would say there are a lot of risk elements. So, definitely, we need some procedures for risk mitigation, and that’s something I can gather from your background. It sounds like Upshot is dedicated to becoming a financial infrastructure to facilitate the collision of NFT and DeFi. So, we would just like to know what made your team focus on this niche in particular.
Nick: Yeah, I think it was around mid-2020 when we had been exploring some general, more academic research around subjective consensus mechanisms and incentives for enabling decentralized communities of actors to reach agreement on resolutions to subjective questions. And as we were thinking about the practical go-to-market for these practical use cases, we started thinking about this problem of NFT price discovery. This was still very early in the NFT days. This was before the initial NFT bull markets, but we anticipated this problem of NFTs being low-velocity, non-fungible assets that don’t change hands very often, which means that these open market interactions, these transfers of ownership, are not a sufficient means of pricing them as they are with more fungible or liquid assets. And when you look to, I guess, analogues in the real world, you see things like real estate, things like traditional art, these other kinds of non-fungible, relatively low-velocity assets using appraisals heavily and sort of supplementing their price discovery. We thought there was an interesting opportunity to build an optimized solution around appraisals for NFTs to power a more real-time, reliable price source for them. Because once you have a reliable source of valuation for assets, you can start to build a diverse suite of really interesting DeFi primitives or DeFi protocols on top of them. So that’s what kicked off a lot of our early interest and focused on NFT pricing.
Challenges and Resistance
Blair: Yeah, NFT appraisal I would say definitely is one of the substantial pain points in this sector. Well, since you mentioned that it’s still very nascent, it’s still in its early stage, have you ever encountered any sort of pushback, resistance, or challenges that made you think, “Wow, it’s really struggling during the journey”? And also, can you share some insights on how to overcome these blockers along the way?
Nick: Yeah, lots. I’ll share a technical blocker we reached that we had to push through, and then I think more of a market-based blocker on the technical side. When we initially started building infrastructure for NFT pricing, we were focused on some of the research we had been working on in the very early days of Upshot, which was around this crowdsourced consensus mechanism. We asked subjective questions to the crowd, had these human actors respond to these questions, and set up incentives such that they maximized their reward when they responded with those questions honestly. Then, we used the result of those responses to power the NFT pricing problem. And so we had these human appraisers, just human collectors appraising specific NFTs. The problem with this is it’s not very scalable and not very accurate. When you’re asking human appraisers to appraise NFTs, maybe they can get through 100 in a day, maybe less, maybe a little more. But you are significantly hindered in terms of the number of assets that can be valued via that mechanism, and then accuracy suffers considerably as well. You know, we are humans. We are sub-optimal cognition machines compared to more powerful computational engines. So, you have to ask these highly approximative questions to humans when you’re asking them to appraise things. You say, “Is this NFT more valuable than this one?” or “Given this NFT in these four price ranges, where does the value fall?” or “What price range does the value of this NFT fall within?” So we had to come up with a new solution after having built out this human-crowdsourced appraisal engine. And so, in the face of that, it was pretty obvious that a more ML-based pricing mechanism was going to be more viable, allowing us to scale significantly larger. Right now, we’re appraising over 100 million NFTs, as opposed to 100 a day. We’re doing that every hour or even sub-hourly, and we’re able to do it with much tighter bounds of accuracy. So, by moving from this initially crowdsourced mechanism to an ML-powered one, we’re able to overcome the initial hurdle of a lack of scalability and accuracy in that type of appraisal.
And I think a problem that even we still face today, to a degree, is more market-based. In that, as an approximation for the value of NFTs, people have sort of gravitated towards the floor price as the canonical value for an NFT. So, creating a compelling argument for why individualized pricing is useful has become a bit of an uphill battle for us. I think it’s a battle we’re winning. It’s a battle that we care very deeply about, and we’re going to continue to educate the market around us. But NFTs obviously are non-fungible assets. You can’t summate the entire value of a collection by saying, “This is the value of its least valuable items,” and that means that’s reflective of the entirety of the collection and the kind of diverse composition evaluations across a collection. That’s why individualized prices are important. And we’re starting to see some of the payings of relying so heavily on floor price for so long as we try to build more interesting financial primitives like NFT lending for Grails of NFTs or perpetual that track different parts of the collection outside of the floor and things like that. And so that’s something we’re solving via education, via sort of solving these problems in the wild by building out lending infrastructure that specializes towards Grails or not-on-floor-asset perpetual for the same indexes for the same, etc. But yeah, those would be the two problems I would say that we faced and have been fighting through.
Competitive Edge of Appraisal Products
Blair: Yeah, wow, it sounds like a lot of work. It’s actually really impressive, and I think that’s definitely the beauty of the whole Web3 entrepreneurship thing because you get to face different challenges, but you also get to be creative and find solutions that can solve the issues better or take things to the next level. Since you mentioned that you actually came from the manual effort, definitely not scalable, that’s why you picked a machine learning-powered NFT appraisal system. So, could you tell us a little bit more about it? How does this appraisal system distinguish itself from other tools on the market? Because there are a lot of players working in the same sectors. Also, how does it contribute to unlocking the full potential of the NFT market, since we mentioned that pricing and appraisal would be a really substantial pain point? Additionally, what strategy does Upshot employ to measure this near real-time NFT pricing?
Nick: Sure, yeah. So at a high level, I think the core of our ML model is not much different from how ML models work for other providers, or anyone pricing any financial asset. We’re taking in as much applicable data as we can. We’re taking in real market data, sales, mints, and softer market data which is like order book data. We’re looking pretty deeply at various networks of wallets, communities, and collections, and how various clusters of nodes or wallets in those networks interact with each other. That influence those clusters have on future price movement, we look at things like sentiment analysis and all these different types of data. And I think one thing that was important for us early on was a structured models system that was highly scalable, both horizontally and vertically. And what I mean by that is a model that is mostly generally applicable to a large swath of the NFT space. So not have to create these customized prioritizations or customized models as we want to add new support for new collections as collections start getting indexed by us, they also just enter our appraisal workflow. It’s a highly generalizable model that works across the NFT space and without significant degradations of accuracy, so that was a really important problem for us to solve. And it was a slightly difficult problem for us to solve. But doing so has allowed us to create coverage that exceeds any other provider producing appraisals by large orders of magnitude.
And then vertically, I mean a lot of models tend to degrade as you look toward the long tail of assets within a collection when you look toward the Grail, when you look toward the highly valuable assets within a collection. That’s because these things don’t change hands frequently. NFTs in general don’t change hands very frequently. And when they do change hands, it’s largely happening on the floor. A Grail may change hands once a quarter, once a year, or sometimes once every three years, sometimes it never changes hands. So, figuring out how to take this highly generalized model and also make it applicable at the tails of collections was an especially important problem for us to solve. Because, as I mentioned earlier, appraisals aren’t as useful near the floor where the market is already interacting with those assets somewhat frequently. The market does a decent job of pricing floors, the market does a really bad job of pricing the mid-tier NFTs and Grail NFTs, they don’t change hands frequently. And so we’ve made some useful insights into how to borrow the data richness or relative data abundance from these lower-value assets that change hands more frequently to infer the value of the mid-tier and Grail assets that don’t change hands frequently. That has unlocked a considerable amount of accuracy benefits when pricing those more long-tail assets. So that’s a bit of how our models work and the things we’ve prioritized. Just to touch on the second part of your question regarding how we measure accuracy, it’s not like reinventing the wheel here. This is a common statistical practice. Essentially, we take the timeline or time series of actual market sales that NFTs have realized. We overlay our appraisals onto them, and for every sale, we look at the most recent appraisal that happened before it. And we say, how far off we are in predicting the next sale that is occurring, and then across several percentiles, sort of at the 5th percentile, 90th percentile, and 95th percentile, we can start to get a sense of what percentiles are our error bound relative to those data points. And we’ve been able to exhibit industry-leading accuracy across all three of those percentiles quite consistently across space, something we are pretty proud of.
Practical Scenarios of API Tools
Blair: Well, that’s very amazing, I would say because I can tell that it’s like your guys’ unique differentiators. But it’s really good to know some backstage stories behind it. Thank you for sharing. Well, I do notice that you guys have a really comprehensive scope of products, and as I mentioned before, your mission is to build up the NFT and DeFi infrastructure thing. And also, you released API tools to provide really sophisticated data and insights for developers. We are just wondering what kind of use cases they are trying to build with the API tools that you made.
Nick: Sure, there are a number of use cases. I think at the side of the DeFi X NFT space, which is maybe where we’re focused more on, just quickly covering some of the other use cases, they’re quite applicable across marketplaces, portfolio trackers, wallets, and analytics platforms. They’re used a lot by individual traders or market makers, deploying capital themselves into the space, just to inform some of those decisions. There’s a quite broad top-of-funnel number of sectors that benefit from having some notion of accurate pricing for NFTs and other kinds of quant/ML-based metrics, and I think where a lot of our focus is, both in partnership and in tools that we’re building ourselves, exist at that intersection of DeFi and NFT. So using these appraisals to create more capital-efficient NFT lending markets, we’re actively running lending strategies that we’ve built on top of our NFT appraisals to manage external capital and deploy it across NFT lending markets.
We’re doing similar types of things shortly in the spot market space via NFT AMM, powered by our NFT appraisals. We’re starting to get more into a sector within that space that I think is going to be quite large, which is NFT Perpetual. Using these NFT price feeds means creating synthetic exposure to different parts of the NFT space and building NFT indexes as an input to these NFT Perpetuals. This allows people to have synthetic exposure to baskets of collections or indexes of NFTs for a more passively constructed exposure to the space. So there are a lot of things happening, and NFT prices unlock various sources of value and different ways of engaging with NFTs in a more interesting financial sense. That’s where a lot of our focus has been.
Lending Strategy of Upshot
Blair: Yeah, looking forward to seeing more innovations in NFT and NFT-fi sectors as well, especially since we’re being really bullish on this sector. So, looking forward to seeing more innovations and creativities. While you mention lending strategy a couple of times, and we have noticed that you guys are stepping into this lending strategy market as well, can you give us more information? Or, can you just elaborate on what’s your lending strategy in particular and how does Upshot plan to improve the overall liquidity with this new product? Because I know there are a lot of players there just aiming to have these liquidity solutions in the market, but everyone probably would have their own recipe differently.
Nick: For sure, yeah. So, there are a couple of problems in the lending space today, and one of these is that we have these two different parts of NFT lending which are the peer-to-pool model and the peer-to-peer model. The peer-to-pool model works by pulling capital and just deploying it into the NFT lending space, based on singular oracle price feeds per collection, which means you’re extending loans that are perpetual to the floors of collections. The downside here obviously is that much of the collection, as you start to get to the mid-tier Grail NFTs within a collection, are largely incompatible with that because if they did want to take a loan against their NFT, they would be taken out alone as if their Grail was priced at the floor, leaving significant money on the table and taking on more risk that would likely make sense for that type of engagement. And then on the peer-to-peer side, it offers a theoretically more capital-efficient sort of relationship with lenders and borrowers, and that you have negotiated terms for the LTV, the interest rate, the duration, etc, of these loans, meaning that if a lender and a borrower have a good sense of what the value of a Grail or moving to your NFT is, they can construct terms that better reflect the appropriate LTV interest rate, etc, for undertaking a loan there, and the problem here is obviously that it’s peer-to-peer, it’s not super scalable.
In order for capital to enter that system, you need lenders that not only have capital but are quite sophisticated and understanding the evaluation of many different assets, and so what we’ve done is we’ve taken our appraisals, and we’ve used them as an important input into developing a more sophisticated generally applicable lending strategy across a larger swath of the space, because when you have appraisals now, all of sudden you have a sense of valuation, and then you can start to create more interesting metrics that are I guess more specific to lending things like recovery rates, which are essentially look ahead of appraisals saying we have an appraisal today, which is fine, but what does that mean about the valuation of this asset 15 days from now or 30 days from now. A loan may reach its stage of maturation, and in doing that, you need to develop some perspective on how prices are going to move in the future. So we poured a bunch of work into making predictive estimates around valuation and then packaging those up in a way that doesn’t require capital providers seeking to get financial exposure to lending markets to need to know anything about lending markets. They deposit capital, and we do all of the hard stuff of figuring out loan terms, constructing a few diversified portfolios, managing risk, etc. And so that’s a way that we’ve been working to take this pricing of the structure that we’ve worked really hard to build and onboard significantly more capital than was available to get into the NFT lending space before, and so we’re hoping that will help spur significantly more liquidity into that space.
Blair: Well, the product is coming soon right？ Is it about to launch？
Nick: So these are live via a launch partner called Astaria today. Astaria is doing some real audits of their contract, so we’re going to be pushing an update to those strategies as they push an update to their system, likely in the next four weeks, but these strategies are expanding as well and will be applied to a much larger part of the NFT space shortly too, but yeah, these are live.
Blair: Wow, so it’s coming our way, and I cannot wait to test it out because I do see that you guys value scalability very much. You have been branching out to different areas like API, lending, appraisals, and more. Could you please share more information about the expanding product scope and the infrastructure or ecosystem you aim to build? Additionally, what is your main focus? It’s a lot, and it’s all super amazing, but we want to know what your main strategy will be.
Nick: I think our main focus is, it has always been and will always be to be seen as a source of truth and reliable pricing for NFTs. The reason we’re starting to move up this DeFi X NFT stack is to bootstrap this nascent ecosystem that we anticipate will be quite large. It also allows us to build more optimized solutions that integrate efficiently with our pricing systems. At our core, we will always be a price oracle for NFTs, supporting the creation of liquid markets for anything based on that price oracle. We’re building these additional things to help bootstrap that process and develop more optimized solutions according to our vision of what is optimal or preferable.
Potential of NFTFi Boom
Blair: Got it. Yeah, makes sense to me. I mean, I think we mentioned a couple of times that you guys value scalability, and it seems like this is a core strategy in your whole product scope, in addition to addressing NFT pricing and liquidity concerns. I can see you guys are actively working towards unlocking further innovative applications within this sector. Also, since we’ve mentioned that we value scalability, which means you guys are probably thinking about, oh, are we gonna have another explosive growth of DeFi, like DeFi summer? What kind of factors do you believe could serve as a really significant catalyst for maybe an NFT summer or NFTFi summer? Do you see NFTFi playing a pivotal role as one of the drivers for the next cycle of growth? I know this question could be really tricky, but I think everyone’s starting to, you know, discuss the next cycle recently, so what’s your take on this?
Nick: Yeah, absolutely. I think NFTFi will be one of the leading narratives of the next bull market. I’m quite bullish right now. Personally, I think things are starting to pick up again, and I think there are going to be a couple of things that spur some of that growth. Specifically within NFTFi, there are a couple of narratives that are going to experience fairly explosive growth in this next cycle. But specific to that category, I think these financial primitives that are being stood up and starting to see more capital inflows are going to spur a lot of that growth. To date, NFTs have been somewhat restricted to spot market interactions. People can go loan only, people can’t take out leverage. The introduction of new vehicles, both for leveraging in the NFT markets as well as hedging positions in the NFT markets, will allow for significantly more capital to enter that space and likely spur significant growth.
We’ve seen this in previous cycles as more derivative instruments come online for a vanilla DeFi, the growth that those kinds of things enabled. I think as the derivative of instruments, particularly around leverage, starts to come online in the NFT space, we’ll see similar levels of growth. One of the things that’s particularly compelling about NFTs as well, which is another catalyst for growth there in my opinion, is that they’re this incredibly expressive data structure. We don’t need to just be representing PFPs or generative art or gaming items as NFTs. We can use them to represent essentially anything on the chain. And so, as more interesting financial instruments or financial assets start to be represented as NFTs, as real-world assets start to come online as NFTs, we’ll see this significantly expanded size of the NFT market. In combination with these derivative markets, these different financial primitives are being stood up for NFTs, leading to significant growth of the entire space. So, it’s both the sort of execution and bringing online of derivative markets for NFTs as well as a significantly expanded scope of what NFTs represent that I believe will be the two main catalysts of the NFT space, NFTFi space, and lead to a lot of growth in crypto more broadly.
Outlook on NFT utility
Blair: Well, thank you for sharing. On top of your answer, I was trying to figure out what do you think of NFT utility. We are all aware of what’s going on with Azuki, and PFP was the main driver of mainstream adoption. That’s why there are a lot of luxury brands and even Starbucks stepping into this world to somehow have their own attempt in the NFT world. So, what do you think of that? Do you think PFP is still going to be the major utility in the NFT space?
Nick: I think PFPs have been an interesting experiment in sort of online community building to a degree – online identity. And there have been a small number of communities built around PFPs that will sustain and continue to grow in the next cycle. I think Azuki is a great example, I think the Yuga ecosystems are a great example, I think pudgy penguins, CryptoDickbutts – these different collections have shown quite a proclivity for becoming the foundation for strong communities online. I think as what happens with any sort of explosive growth of a new asset, there was a lot of noise, there was a lot of snake oil, or a lot of grifts in the last cycle, and I think a lot of the PFPs that were launched over the last cycle will fade to irrelevancy over time, um, and there will be pockets that are valuable but they represent a much smaller percentage of the overall NFT market. I think NFT is a real utility. NFT is that tie to real-world assets. Gaming will represent a larger percentage of the NFT market. The NFTs that just represent more natively financial assets like insurance policies, maybe real estate, maybe different types of options or bonds or liquidity positions will become a much larger percentage of the NFT market over the coming months and years.
The Vision of NFT and AI Intersection
Blair: Yeah, well, I’m actually 100% siding with you on this. But yeah, other than real-world assets, NFT utility, there’s actually one word that has been, wow, I would say a buzzword, like ChatGPT. AI is all actually about ChatGPT. I sometimes just feel like AI is taking over our life. It’s all over the place, everyone is talking about AI, and it’s actually becoming a new headline for all web3 entrepreneurs. It’s not actually only for web3 but for everyone. So, what’s your take on the intersection of NFTs and AI-driven trends? Considering you guys have been implementing machine learning as a fundamental aspect of your business model, what do you think, what kind of innovation we could expect in this?
Nick: I think there’s a lot. I think AI is going to touch everything we do as a society and as people, so there’s a lot to cover there. I think specifically in terms of pricing and NFT financial markets, which is our focus, I think AI has sort of reached a level of maturity and I guess production readiness over the past year has shown itself as likely the most deflationary technology we’ve ever seen and has significantly increased the capabilities for us to interact in complicated ways, with especially long-tail financial markets. So, this is something that we’re doing. I think we’ll see a lot more work around using AI more directly in these financial markets to bring more efficiency and liquidity into them. Using AI to more directly price assets and in types of assets we may not even consider assets today or that we’ve seen as being too opaque or long-tailed to consider seriously for financial use cases. But I also think, and this is something we’re working on as well as part of this, AI is starting to come on-chain more and AI models or infrastructure around AI are being built in ways that are more trustless and verifiable on-chain. In building larger financial markets as these NFT financial markets mature, it’s going to be increasingly important that the models that are influencing them are interacting with them are increasingly trustless and the way you can do this is by bringing them on-chain and building verifiable inference or related infrastructure around that. So, that’s what I’m particularly excited about. We’re doing some work with some people in this space right now that will be announced thing shortly. But I think we’ll see more AI touch the chain in increasingly interesting ways.
Blair: Yeah, looking forward to seeing more narratives and crossover collaborations. Well, thank you so much for sharing everything in terms of your product, and also the industry, and other insights. Again, thank you so much for your time.
Nick: Thank you for having me.
