Two Kites Dancing In A Hurricane

Why companies find themselves chasing the wrong prize & can we fix prediction markets before it's too late

“Success is very intoxicating. It is very difficult to handle all the fame and adulation. It corrupts you. You start to believe that everybody around you is in awe of you, that everybody wants you, and that everybody is thinking of you all the time.” - Ajith Kumar

“The roar of the crowd has always been the sweetest music” - Vin Scully

---

Early success is intoxicating. That’s especially true when your success happens in the face of people telling you all the reasons it won’t. Fuck the haters, you were right, they were wrong!

But a unique danger of early success is that you’ve won the wrong prize. We joke a lot about “play stupid games, win stupid prizes” but oftentimes the games we play evolve in real-time. And so what may have led to you winning the first stage is exactly what stands in the way of you winning the larger prize as the game matures.

One of the ways this type of outcome manifests itself is when companies reach a local maxima by mistake and fail to realize it. It feels good to be winning, so much so that you can lose sight of the fact that you’ve insulated yourself from recognizing where you truly are.

In many cases it can be a mirage, propped up by external factors (let’s say an economic boom flooding customers with disposable income perhaps). It might also be the case that the product or service you’ve built works great but only within a limited scope or under specific conditions that don’t scale to broader markets.

The core tension here is that chasing the real big boy prize (i.e. the global maxima) requires stepping down from your current peak. This is humbling. It means making tough decisions, abandoning a core feature, overhauling the technology stack or cannibalizing what you thought was working. What makes this even more challenging…most of the time you need to be making this decision right as people¹ are telling you how great you are! A lot of the people who told you you were wrong before are now climbing over each other to validate your success. It’s such a dangerous position to be in because it breeds complacency just when you need to be making radical change.

This is the position prediction markets find themselves in today. In their current form they are never going to reach mass market adoption. I’m not going to spend any words here arguing about whether they’ve already achieved this status². Perhaps you disagree with the premise and will now either close the tab or hate-read the rest of this piece. That is your prerogative. But I will reiterate why the model is broken today and what I believe these types of platforms should look like.

---

At risk of sounding too tech-bro’y I won’t rehash the innovator’s dilemma but suffice it to say that the canonical examples here are Kodak and Blockbuster. These companies (and many others) found lucrative success which created an inertia to avoid change. We know how those stories end but it’s not constructive to just throw our hands up and say, be better. So what specifically leads to these outcomes and do we see any of this percolating within prediction markets today…

Sometimes the barrier is technical. Startups build products in opinionated ways that sometimes work for the instantiation phase but can quickly ossify into architectural shackles for the future. Scaling beyond this initial traction or pivoting the product design means threatening some core component that seemed to work³. The natural tendency is to address this with incremental patches but that can quickly lead to a kind of Frankenstein product. Plus it only delays accepting the hard truth that what’s really needed is a fundamental rebuild or reimagining of the product.

Early social networks saw this happen when they hit performance ceilings. Friendster was a pioneer of social networking in 2002 with millions of users connecting friends-of-friends online. But it ran into trouble when one particular feature (the ability to see friends up to “three degrees” removed) caused the platform to buckle under the computational load of mapping these exponential connections. The team refused to scale it back, focusing instead on new ideas and flashy partnerships even as their existing users threatened to leave for MySpace. Friendster reached this local peak of popularity but couldn’t move past it because its core architecture was flawed and the team refused to acknowledge, tear down and fix it⁴.

It’s not necessarily surprising these types of behaviors persist. We are all human. Achieving some semblance of success – especially as a startup where the failure rate is so high – can naturally inflate egos. Founders and investors begin to believe their own hype and double-down on the formula that has led them to where they are today, even as warning signals flash brighter. It’s easy to dismiss new information or even refuse to confront the reality of the environment today versus what it was in the past. The human mind is funny this way. We can rationalize a lot of things if there’s enough incentive to do so.

Research Has No Motion

Pre-iPhone, Research In Motion’s (RIM) Blackberry was the king of smartphones with >40% US smartphone market share. It was built on a specific concept of what a smartphone was: a better PDA for enterprise users specifically optimized for email, battery life and that sweet, sweet tactile keyboard. And yet…

life comes at you fast

What’s probably under-appreciated today is that Blackberry was exceptional at serving its customers. So much so that when the world shifted around them, RIM couldn’t change with it.

The leadership team very famously dismissed the iPhone at first.

“It wasn’t secure. It had rapid battery drain and a lousy digital keyboard” - Larry Conlee (COO at RIM)

And then quickly became defensive.

RIM’s arrogance that this new phone would never appeal to its enterprise customer base wasn’t necessarily unfounded. But this completely missed the transcendence of smartphones moving beyond just email machines to everything-devices for everyone. The company suffered from crippling technical and platform debt, a common symptom of companies who find early success. Their operating system and infrastructure was optimized for secure messaging and battery efficiency. By the time they accepted the reality of the situation it was way too late.

There’s an argument that companies who find themselves in these types of positions⁵ should operate in an almost schizo mindset with one team working to exploit the current success and another working to disrupt it. Apple is perhaps a model example of this as they let the iPhone cannibalize the iPod and then let the iPad cannibalize the Mac. But if it was easy everyone would do it.

Yahoo

Probably a Mount Rushmore bag fumble. For a time Yahoo was the homepage of the internet for millions of people. It was a portal to the internet⁶ – news, email, finance, games. It viewed search as just one of many features, so much so that Yahoo didn’t even use its own search technology in the early 2000s⁷. The leadership team now (in)famously passed up multiple opportunities to deepen their search capabilities, most notably its chance to buy Google in 2002 for $5 billion. It seems obvious in hindsight but Yahoo failed to understand what Google knew: search was the foundation of the digital experience. Whoever owned search was going to own internet traffic and by extension ad revenue. Yahoo overindexed on the strength of its brand and display advertising while catastrophically underestimating the seismic shift to search-centric navigation and later social networks with personalized content feeds.

---

remember this guy?

Excuse the cliche but in frothy markets a rising tide lifts all boats. Crypto knows this all too well⁸. It’s difficult to know if your startup has genuine traction or if it’s riding an unsustainable momentum wave. Making matters murkier, these periods tend to coincide with surges in venture funding and speculative consumer behavior which can mask underlying fundamental issues. WeWork’s hilariously rapid rise & fall illustrates this well: easy capital led to massive expansion that masked a completely broken model.

Underneath all the branding and high-brow language, WeWork’s core business model was straightforward:

Lease office space long-term → spend money to fit it out → sublease short-term at a markup

If you’re unfamiliar with the story you may be wondering, hmm this sounds a lot like a short-term landlord. This is exactly what it was. A real-estate carry trade dressed up as a software platform.

But WeWork wasn’t necessarily interested in building an enduring business, they were optimizing for something very different: explosive growth and narrative valuation. The reason this worked for a short period of time is because Adam Neumann was incredibly charismatic and could sell a vision. Investors ate this shit up and fueled a specific kind of growth that was completely divorced from reality⁹. Plenty of outsiders (i.e. analysts) saw it for what it was: a real estate company with an upside-down risk profile, flaky customers and structural losses built into the business.

Most of this is retrospective analysis of now-failed companies. It’s low-hanging fruit in some sense. But it reflects 3 different examples of failed insights. Companies fail because of an inability to progress technologically, an inability to recognize & respond to competition or an inability to adapt the business model.

I believe we’re seeing the same unfolding right now with prediction markets.

The theoretical promise of prediction markets is alluring:

• Harness the wisdom of crowds

• Better information

• Turn speculation into collective insight

• Infinite markets

But today’s leading platforms have crested a local peak. They’ve unearthed a model that yields some traction and volume but it isn’t the design that will achieve the true vision of ubiquitous, liquid prediction markets on everything.

Superficially both have shown signs of success, nobody doubts that. Kalshi reported that the industry collectively would see ~$30 billion in annualized volume this year¹⁰. The sector has seen another renewed surge of interest in 2024-25, especially as the narratives of on-chain finance plus the gamification of trading push deeper into the cultural zeitgeist. The excessive marketing push from both Polymarket & Kalshi also probably has something to do with this¹¹.

But if we peel back the onion a bit and look deeper, there are red flags suggesting the growth and PMF may not be what it appears. The elephant in the room is liquidity.

For these markets to be useful they need deep liquidity, i.e. lots of people willing to bet on either side of a market so that prices are meaningful and reveal real price discovery.

• Both Kalshi & Polymarket struggle mightily with this except for a few very high-profile markets

• Huge concentration of trading volume is around large events (US elections, highly anticipated Fed decisions) but most markets exhibit crazy wide bid-ask spreads and almost no activity. In many cases, market-makers do not even want to get filled¹².

What this suggests is that these platforms have yet to crack the nut of scaling the breadth and depth of markets. They’ve plateaued at a level where they’re able to do reasonably well within a few dozen popular markets but the longer-tail “markets for everything” vision is not happening.

To mask some of these issues both companies have resorted to incentives and unsustainable behavior (sound familiar?), a telltale sign of reaching a local maxima where organic growth isn’t enough¹³.

Polymarket introduced their liquidity rewards program in an attempt to tighten spreads (i.e. if you rest orders near the current price you’re rewarded theoretically). This helps make order books appear tighter and does provide a better experience for traders by minimizing slippage to some degree. But it’s still a subsidy. Similarly, Kalshi rolled out a volume incentive program which literally offers cash-back rewards to users based on how much they trade. They’re paying people to use the product.

Now I can feel some of you shouting “Uber subsidized for a long time!!!”. Yes, incentives aren’t inherently bad. That doesn’t mean they’re good either!¹⁴ Especially given the current prediction market dynamic, these can quickly become a hamster wheel that’s impossible to get off of before it’s too late.

Another fact we know is that a not-insignificant share of trading volume is wash-trading. I don’t think it’s productive to spend time on debating the exact share but obviously wash trading can make markets appear more liquid when in reality it’s a few actors churning to either earn rewards or create hype. It just means that organic demand is directionally weaker than it seems.

He-Who-Traded-Last-Sets-The-Price

In a healthy, functioning market you should be able to place a bet near the current market odds without the price moving too much. That is just not the reality today on these platforms. Even moderately sized orders move odds significantly, a clear sign of shallow books. Too often these markets don’t reflect anything other than whoever-traded-last and this is the core top-of-book liquidity issue I’ve spoken about before. This reality suggests there is a small community of hardcore users who keep some markets alive but which are, for the most part, not broadly reliable or liquid.

Ok but why is this so?

The binary-only market structure can’t compete with perps. It’s a cumbersome approach that leads to fragmented liquidity and even though these teams are trying to bolt on workarounds to solve this, they’re clunky at best. You also have this bizarre structure in many of these markets where there’s a stand-in for unknowns called “Other”, but that introduces the issue of unpacking emerging contenders from that basket into their own individual markets. The binary nature also means you’re not offering true leverage in the way that users want it, which in turn means you can’t actually generate valuable volume in the way that perps exchanges do. I’ve seen people argue about this on twitter quite a bit and it still shocks me that they can’t recognize how betting $100 on a 1c probability outcome on a prediction market is not the same as opening a $100 100x leverage position on a perps exchange¹⁵.

The dirty secret here is that in order to solve this fundamental problem you need to redesign the underlying protocol to allow for generalizing and treating dynamic events as first-class citizens. You have to create a perp-like experience which means you have to address the jump risk that exists in binary outcome markets. This is quite obvious to anyone who actively uses both perps exchanges and prediction markets, which perhaps unbeknownst to the teams building today, are the users you need. Addressing jump risk means redesigning the system to ensure asset prices move continuously, aka they don’t just arbitrarily jump from say a 45% probability to 100%¹⁶. Without solving for this core limitation, you’ll never be able to introduce the type of leverage needed to make the product attractive to the users who will drive real value to your platform. Leverage relies on continuous price movement to safely liquidate positions before those losses exceed the collateral so that sudden moves (i.e. moving from 45% to 100% immediately) don’t wipe out one side of the order book. If you don’t have this, then you can’t perform timely margin calls or liquidations and inevitably the platform will become insolvent some day.

Another core reason these markets don’t work in the current structure is that you have no native multi-outcome hedging. For one, there’s no natural way to hedge as-is since these markets resolve YES/NO and the “underlying” is the outcome itself. In contrast, if I’m long a BTC perp, I can short BTC elsewhere to hedge. This concept doesn’t exist today in prediction market structure and so it’s extremely difficult to provide deep liquidity (or leverage) if market-makers are forced to take direct event risk. This is again why I believe the argument “prediction markets are new, we are in the hypergrowth phase” is naive.

Prediction markets are eventually settled (i.e. they actually close at resolution), which obviously perpetual futures are not. They are open-ended. A perp-like design could alleviate some of the common behavior that makes prediction markets unappealing¹⁷ by incentivizing active trading to shift the market to function more continuously. The oracle problem is even more pronounced here as well since it’s a one-time discrete outcome versus price feeds, which though they have problems are at least updated continuously.

Downstream of these design issues is the capital efficiency problem but that feels well understood at this point. I personally don’t believe “earning stablecoin yield” on your deployed capital moves the needle¹⁸. Especially since I expect exchanges to provide this anyway. So what is the tradeoff being made here? If every bet is prepaid in full, that’s great for eliminating counterparty risk! And there is ~some~ group of users who you will attract. But it is horrendous for the wider user base you need, it’s extremely unproductive from a capital perspective and all it does is dramatically increase the cost of participation. Which is especially damning when you need heterogenous user types for these markets to work at scale, because these choices mean a worse experience for every cohort. Market-makers need huge amounts of capital to provide liquidity and casual traders face massive opportunity costs.

There is certainly more to unpack here, specifically around how to attempt to solve some of these fundamental challenges. More sophisticated and dynamic margining would be required, specifically to account for factors like time-to-event (risk is highest when the event resolution is close and odds are near 50/50). Introducing concepts like leverage decay as resolution nears would also be necessary and banding liquidation levels early-on would help. Borrowing a sort of prime brokerage concept from traditional finance by enabling JIT collateralizing is another step in the right direction. This would free capital up to be used more efficiently and allow for multiple orders across markets simultaneously, with the book updating on fills. Introducing these types of mechanisms first with scalar markets and then moving into binary ones seems like the most logical sequencing.

The point is that there’s a ton of design space here that isn’t being explored, in part because there’s a belief that the model today is the final form. I’m just not really seeing enough people willing to address that limitations exist in the first place. Maybe unsurprisingly, the ones who recognize these are often the exact user-type these platforms should want to attract (read: perp traders). But instead I see their critiques mostly hand-waved away and told to look at the top-line volume & growth numbers¹⁹. I want prediction markets to evolve, and I want them to be mass-adopted and I personally think markets-for-everything is a good thing. Most of my frustration stems from what feels like an acceptance that today’s versions are the optimal ones and I just don’t think this is true. Alas, if you are interested in helping to build a more perp-like exchange version of these platforms please slide into my dm’s or hmu on tg.

hny everyone!

1. mostly investors & the media

2. There is a wide gap between awareness that a thing exists and demand to actually use it

3. against all odds as a startup no less!

4. Incidentally, MySpace then became a victim of its own sort of local maxima trap: it was built on a unique user experience with customizable user pages and a focus on music/pop culture groups. The platform was super ad-driven and ultimately leaned into its ad portal model while Facebook came around with a much cleaner, faster and more “real” identity based network. Facebook appealed to some of the early MySpace users but certainly appealed much more to the next & much larger cohort of social media users.

5. The bigger the initial success, the harder it becomes to evolve. This is also one of the reasons Zuck is the goat.

6. the original everything app, perhaps

7. it outsourced search to third-party engines and even used Google for a while

8. see Opensea, among many others

9. In the case of WeWork that meant blitzscaling locations (i.e. opening as many buildings in as many cities as possible, irrespective of profitability), locking in long-term leases at scale and poo-pooing the idea that unit economics mattered (“we can grow our way out of these losses”).

10. More on how organic this volume is later

11. brute-forcing in some cases can work

12. One of Kalshi’s founders admitted recently that its internal market-maker isn’t even profitable

13. A small aside here is that, in this particular market dynamic I get the sense that most people believe these are the only two major players competing. I don’t believe that necessarily matters at this stage but if these two teams believe it does, then it becomes an existential threat to their company if the other is seen as “taking the lead” in this presumed 2 horse race. It’s a particularly precarious place to be and in my view is based on a false assumption.

14. i also find it amusing how everyone likes to point to the exception to the rule rather than all the dead bodies

15. if you don’t believe me or don’t get it, i don’t have time to try to convince you, sorry

16. We’ve seen how frequently and blatantly these events have been manipulated/insider traded but that’s another can of worms i’m not particularly interested in atm. Please stop committing crime.

17. Many participants just hold until resolution instead of actively trading the probability

18. It’s borderline crazy it’s taken this long to even get that

19. definitely-real-and-organic