Sharks Eat Fish: What 67% Profit Concentration on Polymarket Means for iGaming Operators

TL;DR

On May 3, 2026 the Wall Street Journal published an analysis of 1.6 million Polymarket accounts showing that 67% of all profits accrue to just 0.1% of accounts — fewer than 2,000 traders splitting roughly $500 million in winnings — and that more than 70% of Polymarket users lose money. The typical user is down between $1 and $100; the worst-performing decile loses an average of $4,000 each. On Kalshi, the company itself disclosed that as of April 2026 every profitable user was matched by 2.9 unprofitable ones. Independent data put aggregate prediction-market trading volume at $29.8 billion with 68.8% of Polymarket users losing money and the top 1% taking 76.5% of total profits. The distribution is not a regulatory failure or a fraud problem. It is the predictable mathematical signature of a venue that uses LMSR-style automated market making, lets professional firms in alongside retail, and markets itself as a "skill" product to an audience that is mostly playing on intuition. The same dynamic eats poker rooms, equity options retail, and sports exchanges. For regulated iGaming operators, the concentration data is genuinely useful: it tells you exactly which segment prediction markets are actually serving (a small population of algorithmic professionals) and which segment they are failing (the mid-stake, skill-curious recreational player you also want). That gap is the strategic opening. Below is the structural read of why the concentration is so extreme, what it means for operator retention, and the four product moves that turn the WSJ data into a competitive position.

---

What the WSJ Data Actually Shows

The Journal analyzed Polymarket's full trading record from November 2022 through April 2026 — 1.6 million of the roughly 2.3 million accounts ever registered on the venue. The headline numbers are unambiguous:

• 67% of all profits went to 0.1% of accounts — under 2,000 traders.
• The top tier collectively earned roughly $500 million in net winnings over the analysis window.
• More than 70% of Polymarket users lost money.
• The typical losing user was down $1 to $100; the bottom decile lost an average of $4,000 each.
• On Kalshi, company spokesperson Elizabeth Diana confirmed that in April 2026 the venue had 2.9 unprofitable users for every profitable one.
• A separate aggregate analysis covering the broader prediction-market category at $29.8 billion in trading volume found 68.8% of Polymarket users losing money and the top 1% capturing 76.5% of profits. Bloomberg has separately reported that more than 100,000 Polymarket accounts have lost at least $1,000 since January 2025.

The reporting also surfaced specific examples of who the 0.1% actually are. A former Princeton student raised $500,000 from crypto incubator Alliance Capital and now runs a five-person trading firm placing $500K–$1M in flow across Polymarket, Kalshi and smaller venues. Another firm — co-founded by a current student, employing about a dozen people, all students — turned $1,000 into a seven-figure sum trading crypto-price contracts on Kalshi and is now in the platform's top five. They spend more than $200,000 a year on real-time data feeds, AI agents and servers and execute tens of thousands of trades per day. A former professional poker player, Michael Boss, has earned more than $668,000 on Kalshi since starting three months ago, mostly on sports — placing 60 trades per minute and adjusting orders 30 times per second.

The losing side gets named too. John Pederson, a 33-year-old Detroit chef who lost his job after a car accident, took out a loan, started betting on Kalshi, built $41,000 in winnings on snowfall and sports markets, then lost the entire balance in January 2026 betting on whether A$AP Rocky would say the word "rapper" during a Jimmy Fallon appearance. (He did say it — but only in the YouTube cut, not the on-air broadcast that Kalshi's resolution rules required. The contract resolved no.)

That last detail matters more than it looks. The WSJ separately analyzed 35,000 completed "mention" contracts on Kalshi — markets on whether a public figure will say a specific word — and found that bets quoted at a 50% probability of winning resolved profitably only about 40% of the time. In efficient-market terms, the recreational side of the order book is systematically overpaying for the underlying probabilities by roughly 10 percentage points in that category. That is a structural edge sitting in the venue, free for any participant disciplined enough to measure it.

Why the Concentration Is Structural, Not a Scandal

Reactions to the WSJ piece have split predictably between "prediction markets are gambling pretending not to be" and "this is what every market looks like." Both miss the operator point. The 67%/0.1% distribution is not a moral failing of either platform — it is the mathematical signature of three design choices stacked on top of each other.

First, the LMSR pricing rule subsidizes liquidity by design. The Logarithmic Market Scoring Rule that underlies most of the academic and commercial prediction-market plumbing is, by Robin Hanson's original construction, a subsidized market maker. The platform takes a bounded, predictable loss in exchange for guaranteeing infinite liquidity at a quoted price. Liquidity-sensitive variants (LS-LMSR) widen the b-parameter as volume grows, mimicking equity-style depth scaling. Either way, there is a quoted price the moment the market opens, even on contracts where almost no informed flow has arrived yet. That quoted price is, by construction, only as accurate as the current order book — and on a thin, newly-listed market, it is mispriced relative to the eventual fundamental probability. Sophisticated participants who can model that mispricing pick the early money off the curve. The platform's b-subsidy and the recreational order flow that arrives afterwards together pay the bill.

Second, the venue lets professional firms participate on the same order book as retail, with no segmentation. Unlike a regulated sportsbook, where the book sets odds and chooses who to take action from, a CFTC-registered prediction market is a multilateral exchange. Anyone with capital can post liquidity, take liquidity, and run algorithms against the book. There is no equivalent of a "limited" account. There is no minimum lot for retail and a different minimum for institutional. There is no segmentation by participant sophistication. The venue's job is to be neutral — that is what makes it a designated contract market under the Commodity Exchange Act rather than a gambling product. But the consequence of that neutrality is that the recreational user with a $500 deposit is sitting in the same pool as a five-person Princeton-funded firm running 10,000 algorithmic trades a day on six-figure data feeds. In poker terms, every table has at least one shark.

Third, the marketing positions the product as a skill game. Both Polymarket and Kalshi have leaned hard into the framing that prediction markets are information aggregators and skill exercises — better than sports betting because the price reflects collective wisdom, better than the casino because the user has agency. The framing is true at the venue level — prediction markets are genuinely better information aggregators than betting books for many event categories — and false at the individual-participant level for almost every retail user. The retail population, by definition, does not have access to the third-party data feeds, the algorithmic execution stack, or the trading discipline that the actual winning population uses. The skill framing creates the audience the venue needs to be liquid; the order-book mechanics ensure that audience is the source of profit for the small population at the top.

Stack those three design choices and you get a market where the headline outcome distribution — 0.1% takes 67% — is the expected result, not an aberration. It is the same mathematical signature you see in retail equity options after Robinhood opened the floodgates, in online poker once data tools and HUDs became standard, and in sports exchanges like Betfair where a small population of value-bettors and market-makers historically captured most of the cross-flow margin. Prediction markets are not anomalous. They are running the exact playbook every neutral exchange runs when it lets retail meet professionals on a level order book.

What This Means for Sportsbook Operators

The instinct of most operator commercial teams reading the WSJ piece will be to feel quietly vindicated: "this is why our regulated, book-set, RG-protected product is the better outcome for the player." That is partly true and entirely insufficient as a strategic response. The data tells you something more specific and more useful, which is that prediction markets are winning the cohort they are worst at serving.

Look at who actually loses on Polymarket and Kalshi. The bottom-decile $4,000-loser is not a casual one-time bettor — that user types in a stake and never comes back. The bottom decile is the mid-stake, recurring, skill-curious recreational player. The $200–$2,000-per-week sports bettor who reads betting Twitter, watches probability content, knows what implied odds are, treats this as a hobby with intent. That cohort is exactly the same cohort I described as the highest-LTV defection risk in the CFTC five-state war post — the segment every U.S. sportsbook is watching migrate to prediction markets because the prediction-market interface and pricing feel more sophisticated.

The WSJ data tells us what happens to that cohort after they migrate. They lose. Systematically. By design. To algorithmic firms running infrastructure they have no access to and no realistic way to compete with. The 0.1%/67% distribution is not a retention story for prediction markets — it is a churn timer. The venues offset it with constant new-user acquisition (sponsored content deals with journalists and influencers, partnership pushes, sports-event coverage), but the underlying pool of recreational deposit dollars is being burned faster than most observers realize. Bloomberg's "100,000 Polymarket accounts down at least $1,000 since January 2025" is exactly the long-tail cohort that should, in a healthy retention funnel, still be playing 18 months in.

The strategic question is whether your operator product can be the place that cohort lands when it stops losing on prediction markets. The honest answer for most regulated sportsbooks today is: not without changes. The migration was driven by interface, pricing transparency, futures depth, and freedom from state-by-state friction. If the bettor comes back from a $4,000 loss on Kalshi, they will not be satisfied with a parlay-only product, opaque hold, sluggish odds movement, or RG protocols that feel like surveillance. They will want some of the prediction-market UX without the shark population.

That is a buildable product. It is also the product that the operators winning the prediction-market era will offer.

The Four Operator Moves the Concentration Data Points To

1. Build a "fair pricing" surface in your sportsbook

The WSJ "mention contract" finding — 50%-quoted markets resolving profitably only 40% of the time — is the kind of structural mispricing recreational users have no way to detect. Your sportsbook does. Surface it. Build a feature in your product that compares your book's implied probability to the live prediction-market price, flags categories where the prediction-market price is structurally biased against retail flow, and explains in the UI what the user is seeing. This is content marketing as product. It tells the skill-curious cohort that the operator has a perspective on price quality, not just a margin to take. The intelligence-layer infrastructure to run this is straightforward — ingest the event-contract feeds, run the rolling expected-value analysis, surface the delta — and it gives your sportsbook a credibility surface that no acquisition ad can buy.

2. Score "ex-fish" cohort re-acquisition as a distinct funnel

The cohort returning from prediction markets after a meaningful loss is a different acquisition shape than a brand-new sportsbook user. They have mental models. They speak in implied probabilities. They are skeptical of generic free-bet offers because they know the math. Define them as a distinct funnel: recent history of crypto-exchange linkages, deposit/withdrawal patterns consistent with prediction-market activity, returning-after-dormancy signals, content engagement on probability and exchange topics. Build a re-acquisition flow specifically for that cohort that leads with pricing transparency, not bonus dollars. Most operators today are running a single onboarding funnel and missing the structural difference between "first-time bettor" and "experienced bettor returning from a venue that ate them."

3. Make your responsible-gaming layer the credible layer

The single most important asymmetry between a regulated sportsbook and a CFTC-registered prediction market is that the sportsbook has affordability protocols, deposit caps, time-on-device limits, and self-exclusion. Prediction markets, operating under derivatives law, do not. The Pederson story — laid-off chef, took a loan, lost $41,000 on a single mention contract — is exactly the user a real RG protocol catches before the loan-to-loss spiral happens. Operators have spent two years framing RG as a compliance cost; the WSJ data turns it into a credible product differentiator for the cohort being burned by the unregulated alternative. Lead with it. Put the affordability check in your acquisition messaging. Make the contrast visible. The "we won't let you lose your rent on a single market" story becomes more powerful, not less, every time a Pederson piece runs in a major outlet.

4. Pick a posture on offering exchange-style pricing inside a regulated wrapper

The deeper product question the data forces is whether your product can offer some of what prediction markets offer — exchange-style pricing transparency, futures depth, peer-to-peer style markets — inside a state-licensed wrapper with RG and customer-protection layers. Tier-1 operators are already exploring this through partnerships with CFTC-registered venues (the distribution-model decision I outlined in the CFTC piece). Tier-2 operators have a window in the next 12 to 18 months to integrate event-contract content as a display and pricing layer without taking on the legal exposure of redistributing it as a tradeable product. Either way, doing nothing locks you into the parlay-and-hold sportsbook model exactly as the most engaged cohort moves on. The 0.1%/67% data is, paradoxically, the strongest operator argument for getting moving — because it tells you the destination cohort is going to be looking for somewhere else to go, fast.

What To Watch in the Next 90 Days

A few specific signals will tell you how the dynamic evolves.

1. Volume by user-cohort disclosures. Both Kalshi and Polymarket have so far been opaque about cohort-level retention numbers. If the WSJ piece pressures either venue into disclosing their loss distributions on a recurring basis, that data becomes infrastructure for every operator's competitive playbook.
2. Regulatory framing shifts. State AGs already arguing the policy gap on prediction-market RG protocols (the 39-AG coalition in the Nevada appeal) now have direct quantitative ammunition. Watch for the data showing up in state filings.
3. Venue product responses. Expect both Polymarket and Kalshi to introduce some form of recreational-trader protection — deposit caps, mention-market warnings, position-size limits — as a defensive PR move. Whichever moves first sets the template the other has to copy.
4. Tier-1 operator pricing-transparency features. First major sportsbook to ship a prediction-market price-comparison surface inside their app will set the competitive bar for the rest of the market.
5. Bloomberg / WSJ follow-up data. Both outlets clearly have pipelines into the underlying trading data. Expect quarterly updates that operators should track as if they were a market data feed.

The Bottom Line

The 67%/0.1% distribution on Polymarket and the 2.9-to-1 loss ratio on Kalshi are not breaking news in any moral sense — they are the structural fingerprint of a neutral exchange that lets professional capital meet retail flow without segmentation, and they look the way every neutral exchange has always looked when the same conditions hold. What is genuinely new is that the data is now public, sourced from the venues' own records, and pegged to specific dollar amounts that put the recreational losing population in the hundreds of thousands of accounts and the cumulative recreational losses in the high nine figures.

For regulated iGaming operators, that data is not a vindication. It is a roadmap. Prediction markets are winning the highest-LTV defection cohort and then converting that cohort into structural losses faster than the venues are admitting. The bettors who survive the experience and come back to a regulated wrapper are going to be looking for a product that gives them the parts of the prediction-market interface that worked — pricing transparency, exchange-style depth, futures access, freedom from generic-bonus condescension — without the parts that didn't, namely a shark population they had no realistic way to compete with. The operators who build that product first capture the cohort permanently. The operators who keep running the same parlay-and-hold sportsbook model from 2022 will keep shedding the same skill-curious mid-stake bettors they have been losing for two years.

The intelligence layer to build it is not exotic. Ingest the prediction-market feeds. Score the defection-and-return cohort. Surface the price-quality comparison inside the product. Lead acquisition with the responsible-gaming credibility the unregulated alternative cannot match. The WSJ analysis is the clearest piece of public evidence in two years that the strategy works because the alternative is structurally hostile to the cohort you both want.

Sharks eat fish. That is what neutral order books do. The operator product that gives the fish a place to come back to — and a UI good enough to actually want to — is the one that wins the next cycle.

---

FAQ

What did the WSJ analysis actually find?

The Journal analyzed 1.6 million of the roughly 2.3 million Polymarket accounts that have traded since November 2022. Headline findings: 67% of all profits went to 0.1% of accounts — under 2,000 traders splitting roughly $500M in winnings; more than 70% of users lose money; the typical loser is down $1–$100, the bottom decile loses an average of $4,000 each. On Kalshi, the company confirmed it had 2.9 unprofitable users for every profitable one in April 2026.

Why is the profit concentration so extreme?

Three structural reasons. First, LMSR-style market making subsidizes liquidity at quoted prices that are mispriced on thin or new markets — sophisticated participants pick that mispricing off the curve. Second, the venues are neutral exchanges that let retail and professional firms participate on the same order book with no segmentation. Third, the marketing positions the product as a skill game to an audience that overwhelmingly does not have access to the data feeds, algorithmic execution, or trading discipline the actual winning population uses. Stack those three and the distribution is mathematically expected, not anomalous.

Is this fundamentally different from sports betting?

Yes and no. Regulated sportsbooks set odds, choose who to take action from, and apply state-mandated RG protocols including affordability checks and self-exclusion. Prediction markets are CFTC-registered exchanges operating under derivatives law with no equivalent obligations. The result is that the loss distributions look more like retail equity options or unrestricted poker rooms than like a regulated sportsbook — concentrated profits, broad recreational losses, no automatic protection for over-extended users.

What's a "mention contract" and why does the 50%/40% finding matter?

Mention contracts are markets on whether a public figure will say a specific word in a public appearance. The WSJ analyzed 35,000 resolved mention contracts on Kalshi and found that bets priced at a 50% probability of winning resolved profitably only ~40% of the time. That is roughly 10 percentage points of structural overpayment by recreational flow, sitting in the venue, available to any participant disciplined enough to measure it. It is a clean example of the kind of mispricing that drives the broader concentration result.

Which operator player segment is most at risk from prediction markets?

The mid-stake, skill-curious recurring sports bettor — typically $200–$2,000 per week, futures-leaning, exchange-fluent, data-heavy. They are the highest-LTV defection-risk cohort for almost every U.S. sportsbook, they are the cohort prediction markets are most successfully acquiring, and the WSJ data shows they are also the cohort the venues are eating fastest. That is the strategic opening for operators who can build the product the cohort returns to after losing.

What's the responsible-gaming angle?

CFTC-registered exchanges have no obligation to apply gambling-style RG protocols — no deposit caps, no time-on-device limits, no affordability checks, no self-exclusion mandates. The Pederson story (laid-off chef, took a loan, lost $41,000 on a single mention contract) is exactly the user a real RG protocol catches before the spiral. Operators should be leading with RG credibility as a product differentiator, not framing it purely as compliance cost.

What should operators build first in response to this data?

Four moves: (1) a fair-pricing surface that compares your book's implied probability to live prediction-market prices and flags structural mispricings; (2) a defection-and-return cohort scoring layer feeding your CRM; (3) a re-acquisition funnel that leads with pricing transparency and RG credibility, not generic bonuses; (4) a clear posture on whether you'll integrate event-contract content as a display layer, partner with a CFTC-registered venue, or stay strictly inside the state-licensed sportsbook model.

Will this data change how Polymarket and Kalshi operate?

Likely yes, defensively. Expect both venues to introduce some recreational-trader protections — deposit caps, mention-market warnings, position-size limits — as a PR response. Whichever moves first sets the template. The deeper structural dynamic (neutral order book, professional capital, retail flow) cannot be solved without breaking what makes a CFTC-registered exchange a CFTC-registered exchange, so the underlying concentration is unlikely to materially compress without participation segmentation that would itself trigger a different regulatory question.

---

Adkuu's intelligence layer ingests live prediction-market feeds and scores defection-risk and ex-fish return cohorts for regulated iGaming operators — letting you surface fair-pricing comparisons, segment the highest-LTV recreational sports bettor by behavioural profile, and route them into RG-credible re-acquisition flows. Learn more about our prediction market infrastructure.

Last updated: May 2026