Businesses in the gambling industry tend to die due to over-regulation, litigation or exhaustion – almost never by choice. Yet on 31 July Nigel Eccles, the co-founder of FanDuel, will voluntarily shut BetHog, the crypto casino he launched with several FanDuel alumni, while it is still functioning, still growing and still, by his own account, a very good product.
The reason is not failure but refocus. Everything BetHog built – its engineers, its technology, its lessons – is being folded into Sentient Studios, a spin-off whose sole product is a photorealistic AI dealer: a conversational avatar that deals blackjack in real time, with no studio, no shift patterns and no human behind the cards. Eccles believes it is the biggest opportunity of a 25-year career in gambling products.
“We ran the risk that if we didn’t put 100% behind it, we’d fail at AI live dealer – the one thing we were really excited about,” Eccles tells iGB.
The story of how he and his company arrived at that conclusion says a good deal about where online casino is heading.
Is AI casino dealer a niche too small?
Let’s start with the disappointment that made the decision easier. BetHog’s founding thesis, hatched with Eccles’ long-time collaborator Rob Jones, was that a superior product would win in crypto gambling. It did not.
“I think we built a really great product,” Eccles says. “What we found was that crypto casino doesn’t really compete on product as much as marketing.” In a market ruled by streamer deals and bonuses, craftsmanship counted for little. And even victory would have been small. “We can definitely carve a niche in crypto casino, but crypto casino is probably less than 10% of online casino. And we have a product here that is really unique,” he notes.
That product began as a side project last summer, when the team started deploying an AI dealer on BetHog itself. The data defied expectations. Live dealer is conventionally a high-roller’s game – “A very small number of players, but they play high stakes, high volume,” as Eccles puts it.
The AI tables behaved differently. “The game plays more like an original,” he says, drawing in lower-staking players who dip in and out. In industry terms, it was mass market – the same quality that made slots-style “originals” the engine of modern casino economics.
Ask the dealer
Player interviews explained why. Many people find human dealers intimidating. “New players find it a little intimidating, because they think, ‘I don’t really know what I’m doing,'” says Eccles, who himself has played blackjack for 30 years.
Sitting on a 15 against a dealer’s 10, novices did something they would never dare at a human table: they asked the dealer what to do. Sentient turned the behaviour into a feature – a button on the interface.
The turning point came in April, when the company began talking to operators: a mix of crypto casinos and, more tellingly, “large global regulated operators”. The regulated giants harbour deep frustrations with live dealer. Some jurisdictions ban it outright; others allow it only under certain conditions, such as requiring streams from a physical casino.
The inflexibility of traditional live casino development
However the main problem was inflexibility. An operator wanting to rebrand a live studio around a flagship sports partnership is typically quoted three to six months, at great cost, with long commitments attached. “We could probably do that in one to two weeks max,” says Eccles. “We can change the uniform of the dealers. The dealers themselves [can even change] – how they talk, their personalities, what they look like.” Capacity becomes flexible too: “There’s no ‘you have 20 tables’ – you can have one table or 10,000 tables. If you need several hundred of them on a Friday night and hardly any on a Saturday morning, that’s not a problem.”
Players loved it; operators wanted it; several tier-one licensees are expected to go live later this year. At that point, running two companies in parallel stopped making sense, Eccles explains. The 15-strong engineering team, all based in Britain, had in any case been migrating naturally. “Gradually everybody moved over to AI live dealer, because that was the exciting thing to be working on,” Eccles says. “It was one of those decisions where, as soon as I thought about it, I was like: this is obvious.”
The making of a virtual croupier
None of this means the product was easy to build. Generating photorealistic video has become almost trivial; while generating responsive photorealistic video has not, Eccles explains. The slick clips flooding social media take minutes to render. A dealer has seconds.
When a player is dealt two aces, the avatar must register the hand, compose a remark, choose an emotion, sync her lips and coordinate hands and posture – all before the moment passes. “When we started, we didn’t appreciate all of that,” Eccles concedes.
Early models overacted: “If you want her to be shocked, she looks terrified – and you’re like, no, just slightly frightened.”
The first version, launched in October, was deliberately cartoon-like; photorealism arrived in April, and fortnightly releases have gradually made the characters feel more natural since. Underneath it all, the base models keep improving. “Trillions of dollars have been invested in these base models,” Eccles notes.
“We look at our product from eight months ago and compare it to today and it’s just quite stunning. Every release, it improves.”
Live casino rivals face innovator dilemma
The competitive question writes itself: What stops, for example, Evolution, the Swedish giant whose vast studios and thousands of human dealers dominate live casino, from building the same thing with deeper pockets? Eccles expects imitation – a rival is “very, very close to being live”, he says – and he claims to welcome competition: “I’ve been in markets where I was the only one, and it turned out it wasn’t a very good market.”
But he argues the incumbents face a classic innovator’s dilemma. His analogy is De Beers, which greeted lab-grown diamonds with a marketing war insisting they were not real, before grudgingly selling them as an inferior line.
Evolution has previously been very dismissive of AI live dealers, calling them deepfakes. Embracing them now, Eccles argues, would cannibalise their own product. Eccles’s forecast is immodest: “In five years’ time, 80% of live dealer will be AI-based avatars.”
Trust beyond the table
Sceptics will raise two objections against AI live dealers. The first is authenticity: surely the point of live dealer is that it is live? Eccles counters that the sociability of live tables is largely mythical. Visit a public Evolution blackjack table and “90% of the time you’ll see absolutely zero communication”.
As for trust issues, Eccles recalls: “Live dealer definitely came about in an era when trust was a major factor – people were saying, ‘I don’t know if I trust these sites.’ Physical cards do help drive trust. But the interesting thing is that physical cards aren’t the only way you can drive trust.”
If they were, Eccles says, there’d be no way that slot providers like Pragmatic Play or Hacksaw would work.
“No user understands what their RNG system is and whether it’s fair, but they’re trusted because of the brand. So we think there are other ways you can build trust, not just physical cards. The current industry already uses a lot of virtual elements, and users accept it.”
Trust in Sentient’s dealer has risen as production values have improved, and the games can be made provably fair. Regulators have proved unexpectedly relaxed: strip away the avatar and “at its core, it’s just an RNG product”, Eccles says.
The dealer has no knowledge of upcoming cards, which sit with a separate, certified engine, and responsible-gambling behaviour is trained in: if, for instance, a dealer hears a player admit he is out of money, it will warn him off.
Operating in the cryptocasino grey zone
The second objection is sharper. BetHog operated in the crypto-casino grey zone at a moment when American regulators were hunting unlicensed operators. Was the closure really about opportunity, or about walls closing in? “Almost none of the latter,” Eccles insists: the firm avoided contested markets, used crypto as payment rails rather than as a shield, and was exploring regulated licences before the pivot. Any remaining balances, he promises, will be reimbursed: “If somebody emails me in two months’ time and says, ‘I had $500 in there,’ we would honour it.”
Nor does his marathon lawsuit against FanDuel’s former board – whose fraud and fiduciary-duty claims a New York court recently allowed to proceed – intrude on the new venture. “At this point, it’s not significant. It’s a multi-million-dollar case, so it’s heavily staffed. It is something that rumbles on, but it has no influence on our business otherwise – it’s completely separate.”
Too good for the room
The strangest problem Sentient now faces is excellence, which again ties back to authenticity. Paradoxically, the issue can be that the product simply is too good. Eccles highlights audio quality as an example:
“The audio quality of live dealers is typically very poor – oftentimes you can’t really hear the dealer that well, because the microphones are on the table and they pick up a lot of background noise. When we create our dealers, the audio quality is just way too high. So there’s this weird thing where we have to dirty the sound quality and interweave it with background noise, otherwise it just breaks that theatre.”
As for unease about machines replacing people, Eccles reaches for the ATM, once shunned, and the self-checkout, once cursed. “People say ‘I don’t know, I want it to stay traditional’ – and then it all just kind of moves.”
Punters who have sworn they would never gamble against an avatar may soon f...
博彩行业的企业往往因过度监管、诉讼或精力耗尽而消亡——几乎从未主动选择退出。然而，7月31日，FanDuel联合创始人Nigel Eccles将主动关闭BetHog——他与几位FanDuel旧部共同推出的加密赌场——而该平台仍在运营、仍在增长，且据他本人所言，仍是一款非常出色的产品。
原因并非失败，而是重新聚焦。BetHog所构建的一切——其工程师、技术、经验教训——正被整合进Sentient Studios，一家分拆公司，其唯一产品是逼真的AI荷官：一个能实时发二十一点的对话式虚拟形象，无需演播室、无需轮班、牌桌后也无需真人。Eccles认为，这是他25年博彩产品生涯中最大的机遇。
“我们承担的风险是，如果不100%投入，我们就会在AI真人荷官上失败——而那恰恰是我们真正兴奋的事情，”Eccles对iGB表示。
他和公司如何得出这一结论的故事，在很大程度上揭示了在线赌场的发展方向。
AI赌场荷官是否是一个太小的细分市场？
让我们从那个让决策变得更容易的失望说起。BetHog的创立论点是，与Eccles长期合作伙伴Rob Jones共同构思的——在加密博彩中，更优质的产品将胜出。但事实并非如此。
“我认为我们打造了一款非常出色的产品，”Eccles说。“我们发现，加密赌场的竞争与其说靠产品，不如说靠营销。”在一个由主播合作和奖金主导的市场中，精工细作无足轻重。即便胜出，规模也很小。“我们肯定能在加密赌场中开辟一片天地，但加密赌场可能还不到在线赌场的10%。而我们这里有一款真正独一无二的产品，”他指出。
这款产品始于去年夏天的一个副项目，当时团队开始在BetHog自身上部署AI荷官。数据出乎意料。真人荷官通常是豪客的游戏——用Eccles的话说，“玩家数量非常少，但他们下注高、玩得多。”
AI牌桌的表现则不同。“游戏玩起来更像原创游戏，”他说，吸引了那些小注玩家进进出出。用行业术语来说，这是大众市场——正是这一特质让老虎机式的“原创游戏”成为现代赌场经济的引擎。
问问荷官
玩家访谈解释了原因。许多人觉得真人荷官令人望而生畏。“新玩家会觉得有点吓人，因为他们想，‘我其实不太懂自己在做什么，’”Eccles说，他本人玩二十一点已有30年。
当手牌为15点而对庄家10点时，新手会做一件在真人牌桌上绝不敢做的事：他们问荷官该怎么办。Sentient将这一行为转化为一项功能——界面上的一个按钮。
转折点出现在4月，当时公司开始与运营商洽谈：既有加密赌场，更有“大型全球受监管运营商”。这些受监管巨头对真人荷官积怨已久。一些司法管辖区直接禁止； 另一些则仅在特定条件下允许，例如要求从实体赌场进行直播。
传统真人赌场开发的不灵活性
然而主要问题在于不灵活。运营商若想围绕一项旗舰体育合作对真人演播室进行品牌重塑，通常得到的报价是三到六个月、成本高昂，且附带长期承诺。“我们大概最多一两周就能做到，”Eccles说。“我们可以更换荷官的制服。荷官本身[甚至可以改变]——她们的说话方式、个性、长相。”容量也变得灵活：“不存在‘你只有20张桌子’——你可以有一张桌子，也可以有一万张。如果你周五晚上需要几百张，周六早上几乎不需要，那都不是问题。”
玩家喜爱它；运营商想要它；预计今年晚些时候将有数家一级持牌商上线。Eccles解释说，到那时，同时运营两家公司已不再合理。这支15人的工程团队全部驻扎英国，无论如何早已在自然迁移。“渐渐地，所有人都转向了AI真人荷官，因为那才是令人兴奋的工作，”Eccles说。“这是那种一旦我想到，就会觉得：这显而易见。”
虚拟荷官的诞生
这一切并不意味着产品容易打造。生成逼真视频已几乎微不足道；而生成可实时响应的逼真视频则不然，Eccles解释道。充斥社交媒体的精美短片渲染只需几分钟。而荷官只有几秒钟。
当玩家拿到两张A时，虚拟形象必须识别手牌、组织话语、选择情绪、同步唇形并协调手部与姿态——这一切都要在时机过去之前完成。“刚开始时，我们并未充分认识到这一切，”Eccles承认。
早期模型表演过度：“如果你想让她震惊，她看起来却像吓坏了——你会想，不，只是稍微受惊而已。”
10月发布的第一版刻意采用卡通风格；4月实现了逼真化，此后每两周一次的发布逐渐让角色感觉更自然。 在这一切之下，基础模型不断改进。“数万亿美元被投入到这些基础模型中，”Eccles指出。
“我们看八个月前的产品，再对比今天，简直令人惊叹。每一次发布，它都在进步。”
真人赌场竞争对手面临创新者困境
竞争问题不言自明：是什么阻止了，比如说，Evolution——这家瑞典巨头凭借庞大的演播室和数千名真人荷官主导真人赌场——用更雄厚的资金打造同样的东西？Eccles预计会有模仿——一个竞争对手“非常非常接近上线”，他说——并声称欢迎竞争：“我曾身处只有我一个人的市场，结果发现那并不是一个很好的市场。”
但他认为，现有企业面临经典的创新者困境。他以戴比尔斯作比——该公司对实验室培育钻石发起营销战，坚称它们不是真钻石，后来才不情愿地将其作为低端产品线出售。
Evolution此前对AI真人荷官极为不屑，称其为深度伪造。Eccles认为，如今拥抱它们将蚕食其自身产品。 Eccles的预测颇为大胆：“五年后，80%的真人荷官将是基于AI的虚拟形象。”
牌桌之外的信任
怀疑者会对AI真人荷官提出两点反对意见。第一是真实性：真人荷官的意义难道不就在于它是实时的吗？Eccles反驳说，真人牌桌的社交性在很大程度上是神话。访问一张公开的Evolution二十一点牌桌，“90%的时间里你会看到完全没有交流”。
至于信任问题，Eccles回忆道：“真人荷官确实诞生于信任成为主要因素的时代——人们当时说，‘我不知道我是否信任这些网站。’实体牌确实有助于建立信任。但有趣的是，实体牌并非建立信任的唯一方式。”
Eccles说，如果是的话，那么Pragmatic Play或Hacksaw这样的老虎机供应商就不可能行得通。
“没有用户了解他们的RNG系统是什么、是否公平，但他们因为品牌而信任它。所以我们认为，还有其他方式可以建立信任，而不仅仅是实体牌。当前行业已经使用了大量虚拟元素，用户也接受了。”
随着制作水准的提升，对Sentient荷官的信任也在上升，而且游戏可以被做成可证明公平。监管机构出乎意料地宽松：剥去虚拟形象，“其核心不过是一个RNG产品”，Eccles说。
荷官不知道即将发出的牌，这些牌由一个独立的、经过认证的引擎掌管，并且负责任博彩行为被训练其中：例如，如果荷官听到玩家承认自己没钱了，它会劝他收手。
在加密赌场灰色地带运营
第二点反对意见更为尖锐。BetHog在加密赌场灰色地带运营，而当时美国监管机构正在追查无牌运营商。关闭真的关乎机遇，还是因为四面楚歌？“几乎完全不是后者，”Eccles坚称：公司避开了争议市场，将加密货币用作支付通道而非护盾，并在转型前就在探索受监管牌照。他承诺，任何剩余余额都将退还：“如果有人在两个月后给我发邮件说，‘我在里面有500美元，’我们会兑现。”
他对FanDuel前董事会旷日持久的诉讼——纽约一家法院最近允许其欺诈和信托义务索赔继续推进——也不会干扰新业务。“目前来看，这并不重要。这是一起数百万美元的案件，所以投入了大量人手。它确实在持续发酵，但除此之外对我们的业务没有影响——完全独立。”
好得与场景格格不入
Sentient如今面临的最奇怪问题是过于出色，这又回到了真实性上。矛盾的是，问题可能恰恰在于产品太好了。Eccles以音频质量为例：
“真人荷官的音频质量通常非常差——很多时候你听不太清荷官说话，因为麦克风在桌上，会拾取大量背景噪音。当我们制作荷官时，音频质量就是太高了。所以有个奇怪的现象：我们不得不把音质弄脏，并混入背景噪音，否则就会破坏那种剧场感。”
至于机器取代人带来的不安，Eccles举出ATM和自助结账的例子——前者曾被回避，后者曾被诅咒。“人们说‘我不知道，我希望保持传统’——然后一切就这么变了。”
那些曾发誓绝不与虚拟形象对赌的玩家，可能很快就会……