The Tell
Poker players have a word for the involuntary habit that gives away a weak hand: a “tell.” Boards have developed one too. Whenever a public company needs to explain a bad quarter without using the word “bad”, it now reaches for “AI-first restructuring”. Similarly, when trying to justify overpaying for a business, Boards use “synergies”. I have yet to see a deal that achieved the level of synergies promoted during the acquisition or merger process.
In May, Gambling.com Group cut about a quarter of its staff, told investors the move was building a flatter organisation around “AI-first ways of working”, and in the same results release cut its full-year guidance and reported a 43% fall in adjusted EBITDA. The market knocked a quarter off the share price on the day. Nobody in that boardroom thinks AI caused the miss. Somebody clearly thought AI would sound better in the release than “costs are too high and search traffic is down.”
I have sat in a few boardrooms over the years and recognise the tell from the inside. A Board asks management what the company’s “AI strategy” is. After a bit of head scratching, somebody produces a very good deck. Nobody asks the harder questions: What has actually been built, by whom, on top of which data, and measured how?
Julie Averill, formerly Lululemon’s chief information officer, wrote about exactly this pattern in the New York Times this month and coined two names worth borrowing: “AI wishing”, the belief that a hard operational problem can be waved away with the right vendor demo, and “AI washing”, think “green washing” but with AI.
Claiming you are using AI when you are not has legal consequences. The SEC fined the investment adviser Delphia $225,000 in 2024 for telling clients its models used AI and machine learning to analyse their data, when the SEC found those inputs were never actually fed into anything.
What the Numbers Show
None of this is unique to gambling. MIT’s Project NANDA studied more than 300 enterprise AI deployments last year and found 95% produced no measurable financial return, against an estimated $30 billion to $40 billion in enterprise spending behind them. Boston Consulting Group’s most recent AI Radar research comes up with a similar number: 75% of executives call AI a strategic priority, but only 25% say it has created significant value. The quarter of companies that succeed do something specific with their money. They put over 80% of their AI budget into redesigning core processes and building new products, rather than scattering it across small productivity tools, and BCG’s own estimate is that only 10% of the value that follows comes from the algorithm itself. Most of the rest comes from people, process, and organisation, which is consultant-speak for the unglamorous work nobody puts in a press release. Wharton’s most recent enterprise survey found Chief AI Officer roles now sitting at 60% of large companies, though it adds, somewhat apologetically, that most of these are existing executives handed a new badge rather than a new budget. A title, as Averill puts it, costs nothing.
Gambling now has its own scorecard for this, courtesy of UNLV’s International Gaming Institute and KPMG, which surveyed 83 operators and suppliers worldwide for the inaugural ‘State of AI in Gaming’ report. The industry’s composite AI Maturity Index averages 45 out of 100. Strategy is the strongest of its four components at 57. Governance is the weakest at 30. A 27-point gap between how confidently a company can describe its AI ambitions and how well it can actually govern them is Averill’s argument, but with a number attached, and it is not a flattering one.
DraftKings gave the pattern a face this year. Chief Executive Jason Robins announced a hiring freeze, citing tremendous momentum in agentic AI and a shift from managing employees to managing AI agents, with reported potential savings of around $30 million a year. Whatever momentum Robins is seeing internally, the survey suggests he is well ahead of the field: AI agents and reasoning systems are the least adopted category of AI in gambling, used by just 32% of companies, trailing generative AI (81%) and conversational AI (67%) by a wide margin. I am not saying Robins is wrong about where this eventually goes. I am saying a headcount decision justified by a capability the rest of the industry has barely started using is precisely the kind of claim that is worthy of follow-up questions — and results, not confidence, are the only acceptable answer.
The Governance Gap
The survey’s more uncomfortable numbers sit behind the governance score. Less than 20% of gambling companies have a dedicated AI governance or ethics position and only 8.4% plan to hire one, despite data privacy and governance ranking among the industry’s own top-cited risks. The regulators surveyed for the same report say, in effect, that they do not trust the industry to close that gap itself: limited visibility into what licensees are actually doing with AI, low confidence in their own oversight capacity, and no confidence the industry can self-regulate.
That scepticism looks earned. In March, an investigation by the Guardian and Investigate Europe prompted five mainstream AI chatbots — Microsoft Copilot, Grok, Meta AI, ChatGPT, and Gemini — for advice on gambling and every one of them produced lists of unlicensed offshore casinos sitting outside GAMSTOP, the UK’s self-exclusion scheme. Copilot called several of the sites “reputable”. Grok suggested paying in crypto to avoid the identity checks that might have stopped a self-excluded player at the door. The findings reached Parliament’s House of Commons within a fortnight. None of these chatbots belong to a licensed operator, which is rather the point. Governance built only around your own AI stack does nothing about the AI systems the most vulnerable gamblers are actually talking to.
Cybersecurity is the other risk gambling companies rank highest in the survey and the industry does not need reminding why. The 2023 breach that took MGM Resorts offline for 10 days, shutting down slot machines, room keys, and booking systems, and reportedly cost Caesars a $15 million ransom the same week, began with nothing more sophisticated than a phone call. The Scattered Spider group found enough of an employee’s details on LinkedIn to talk MGM’s own help desk into resetting the wrong credentials. No AI model failed that day. What failed was a process with no governance built into it, which is exactly the pattern this survey keeps returning to. Give an AI agent real permissions inside a process like that and the bet gets considerably larger.
Land-based operators have their own version of the gap and it is more mundane than governance failure: data. The survey found land-based operators trailing their online counterparts by a statistically significant margin on strategy, infrastructure, and expertise alike and not one of the 20 companies reporting established or advanced AI capability was a land-based operator. This will not surprise anyone who has sat, as I have, in a data room for a casino-resort and asked for a single customer view across casino, hotel, F&B, and retail, only to be handed five systems that each believe they are the master record. You cannot point a language model at a decade of disagreeing databases and expect coherence to emerge from the argument. Somebody still has to do the plumbing and plumbing is not saleable for an earnings call.
Some gambling companies are doing the less quotable version of this properly. Light & Wonder’s leadership describes its AI approach in both offensive and defensive terms: offensively, using AI to speed up platform development and cut technical debt; and defensively, leaning on over $500 million a year of R&D, more than 500 licences, and a library of proven, hard-won math models as a moat that cannot be copied by prompting a chatbot. That is a company treating AI as infrastructure to be built rather than a headline to be issued, backed by a balance sheet that has been quietly paying for the infrastructure for years before anyone thought to ask about it.
“Strategy is the cheap part. Everyone at the table already has one.”
Poker players will tell you the best hands do not need to announce themselves. The gambling companies actually closing the governance gap are, almost by definition, the ones you hear least about, because clean data and a properly resourced compliance function do nothing for a press release. When I look at a company’s AI capability now, I have stopped asking to see the strategy deck. I ask for the org chart, the data dictionary, and the name of whoever owns AI governance — and increasingly, the second and third questions lead to the more difficult ones. The industry’s own numbers make the same point rather more politely: Strategy is the cheap part. Everyone at the table already has one.