[{"data":1,"prerenderedAt":277},["ShallowReactive",2],{"insight-/insights/ai-investment-casino-gambling":3},{"id":4,"title":5,"author":6,"body":7,"category":259,"ctaTarget":260,"date":261,"description":262,"extension":263,"image":264,"imageAlt":265,"keyphrases":266,"meta":267,"metaDescription":268,"navigation":269,"path":270,"readingTime":271,"seo":272,"seoTitle":273,"stem":274,"tags":275,"__hash__":276},"insights/insights/ai-investment-casino-gambling.md","Are you a casino gambler with AI investments in your own business?","Francois van der Merwe",{"type":8,"value":9,"toc":244},"minimark",[10,14,25,28,31,36,39,42,46,49,58,62,71,86,90,99,102,106,115,124,128,137,140,143,147,150,165,169,172,181,185,200,203,212,216,225,234,238,241],[11,12,13],"p",{},"I think a lot of executives are making the same mistake with AI right now, the kind of mistake that turns up whenever an industry gets over-excited: the spending chases whatever looks impressive rather than whatever works.",[11,15,16,17,24],{},"The casino picture comes from Hock Tan, the Broadcom CEO. At a conference in Singapore on 12 October he said that \"a lot of investors have no clue what they are investing in\", as ",[18,19,23],"a",{"href":20,"rel":21},"https://wolfstreet.com/2026/10/12/ceos-of-broadcom-intel-say-the-quiet-part-out-loud-about-ai-and-roi/",[22],"nofollow","Wolf Street reported",", and described investors setting \"very extreme expectations\" for AI only to find \"it's almost like they are in a casino\". He was talking about investors, not the budgets inside your own company, so I am borrowing his picture and not his argument. He is not calling a bubble either: he also said \"there is more demand for compute infrastructure than we can supply\", and expects that to last until 2029, probably into 2030.",[11,26,27],{},"He had a different picture for enterprises, and it is the one that applies to your budget. \"When you are a kid you go to a video arcade, you play. Your parents give you $10 of tokens, you finish it in 10 minutes\", and he went on: \"you haven't learned anything. That's what's happening in enterprises today.\"",[11,29,30],{},"As I see it, the tokens go on a demo, a pilot or an announcement, because each one looks like progress at the time, and the question that matters, what the organisation learned and what a customer got, comes later. The rational and the reasonable in the room need to prevail. That means steering away from the wow factor, the impressive slideware, the latest ChatGPT release, and the robots racing, boxing and playing football at the so-called robot Olympics, and back toward business fundamentals.",[32,33,35],"h2",{"id":34},"business-fundamentals","Business fundamentals",[11,37,38],{},"We know what drives a successful business. It creates significant value for its customers, it does that consistently, it scales that value, and it does so at a price point that lets it make a profit, both gross and net. Nothing about that changes because the technology underneath is new. An AI initiative that cannot say which customer gets the value, how reliably it arrives, and what it costs to run at scale has no business calling itself an investment. Value, consistency, scale and margin are four tests, and I would put every AI proposal through all four.",[11,40,41],{},"I am not anti-AI. I think it is powerful, and the places it earns its keep are usually boring: removing cost from work that already exists, or improving something a customer already pays for. That is where the fundamentals show up, in the numbers rather than in the pitch.",[32,43,45],{"id":44},"value","Value",[11,47,48],{},"Significant value means a customer somewhere is measurably better off, not merely impressed. An internal assistant that everyone enjoys using is not yet value. It becomes value when you can point to the customer who is served faster or better because of it, and measure the difference.",[11,50,51,52,57],{},"The ",[18,53,56],{"href":54,"rel":55},"https://newsroom.ibm.com/2025-05-06-ibm-study-ceos-double-down-on-ai-while-navigating-enterprise-hurdles",[22],"IBM Institute for Business Value"," surveyed 2,000 CEOs across 33 countries in early 2025, and those CEOs reported that only 25 percent of AI initiatives had delivered the expected return on investment over the last few years. IBM sells AI itself, and this is CEOs grading their own programmes rather than an independent audit, so I take it as a signal rather than a verdict. Even so, by their own account, three in four initiatives did not deliver the return expected of them. A business case that cannot name the customer and say what that customer gets has not cleared the first test, and nothing that follows will rescue it.",[32,59,61],{"id":60},"consistency","Consistency",[11,63,64,65,70],{},"Consistently means the value has to turn up every time, not only on the good days. In June 2024 ",[18,66,69],{"href":67,"rel":68},"https://www.nrn.com/quick-service/mcdonald-s-is-ending-its-ai-drive-thru-test-with-ibm",[22],"McDonald's told franchisees"," it would end its automated order-taking partnership with IBM, which had been tested in about 100 restaurants since the two companies first partnered in 2021. Its own explanation was measured: \"While there have been successes to date, we feel there is an opportunity to explore voice ordering solutions more broadly.\" I do not know the internal numbers and I will not guess at them. The general point stands without them: a system that works often enough to impress a visiting executive is not the same as one that works every time a customer is hungry and in a hurry.",[11,72,73,74,79,80,85],{},"Air Canada shows a harder side of consistency. In February 2024 British Columbia's Civil Resolution Tribunal held the airline responsible for inaccurate bereavement-fare information that the chatbot on its website gave a customer, and ordered it to pay a total of C$812, including damages, interest and fees, as ",[18,75,78],{"href":76,"rel":77},"https://www.foxbusiness.com/lifestyle/air-canada-chatbot-gave-customer-wrong-info-airline-pay",[22],"Fox Business reported",". It is a small-claims tribunal decision, not a court precedent, but ",[18,81,84],{"href":82,"rel":83},"https://www.mccarthy.ca/en/insights/blogs/techlex/moffatt-v-air-canada-misrepresentation-ai-chatbot",[22],"the tribunal's line"," is worth keeping: \"It should be obvious to Air Canada that it is responsible for all the information on its website.\" The sum was small. The lesson I take from it is bigger: a business can be held to what its AI told a customer, including the time it was wrong.",[32,87,89],{"id":88},"scale","Scale",[11,91,92,93,98],{},"Scaling that value means the one success has to become the normal case across the business, not a pilot that stays a pilot forever. A ",[18,94,97],{"href":95,"rel":96},"https://www.ciodive.com/news/AI-project-fail-data-SPGlobal/742590/",[22],"survey by S&P Global Market Intelligence"," of more than 1,000 respondents in North America and Europe found that 42 percent of businesses said they were scrapping most of their AI initiatives, up from 17 percent the year before, and that organisations on average reported scrapping 46 percent of their AI proofs of concept before they reached production.",[11,100,101],{},"Producing value once is hard enough. Multiplying it across a few hundred branches, stores or teams is a different and harder problem, with its own cost and its own ways of going wrong, and a demo never tests for it. I suspect this is where a lot of AI budgets stall: not in the pilot, which usually looks good to the people running it, but in the long, unglamorous work of making the pilot ordinary everywhere, which is the part nobody puts in the board pack. Scale is a unit-cost question before it is a rollout question, which brings me to the last test.",[32,103,105],{"id":104},"margin","Margin",[11,107,108,109,114],{},"The last test is the one executives skip fastest: a price point that leaves room for gross and net profit, not just revenue. AI has a running cost that grows with use, because every question, every document and every automated step has to be computed, and the people who build these products have been caught out by that themselves. In early January 2025 Sam Altman posted that OpenAI was losing money on its $200-a-month ChatGPT Pro plan, writing, as ",[18,110,113],{"href":111,"rel":112},"https://fortune.com/2025/01/07/sam-altman-openai-chatgpt-pro-subscription-losing-money-tech",[22],"Fortune reported",": \"insane thing: we are currently losing money on openai pro subscriptions! people use it much more than we expected.\" That is one plan at one moment, not a verdict on OpenAI's finances. But if the people who build the models can misjudge what heavy use costs, a business case built on the cost of a small pilot can misjudge it too.",[11,116,117,118,123],{},"Gartner expects inference, the cost of actually running a model, to account for at least 70 percent of a model's lifetime costs, and predicts that through 2028 at least half of generative AI projects will overrun their budgeted costs, ",[18,119,122],{"href":120,"rel":121},"https://thejournal.com/articles/2026/06/22/report-half-of-gen-ai-projects-could-exceed-budget-by-2028.aspx",[22],"as The Journal reported",". Those are forecasts, not results, but they point the same way. I think building the thing is often the cheaper part, and running it everywhere, every day, is where the margin can go. The margin that matters is yours, after the AI bill has been paid.",[32,125,127],{"id":126},"the-veil-of-techno-excitement","The veil of techno excitement",[11,129,130,131,136],{},"One of the most important things here, I think, is the veil of techno excitement that gets drawn over executives' eyes. It arrives as a pitch: look, this is how we will meet the AI mandate the board expects you to deliver on. The mandate is real, and so is the pressure behind it. A ",[18,132,135],{"href":133,"rel":134},"https://www.dataiku.com/stories/blog/confessions-from-500-global-ceos",[22],"Harris Poll survey commissioned by Dataiku",", an AI platform vendor, reported in March 2025 that 63 percent of the CEOs surveyed said their boards demand AI results, and 96 percent of those conceded the demands were justified. The IBM study has a related number: 64 percent of CEOs acknowledged that the risk of falling behind drives investment in some technologies before they have a clear understanding of the value those technologies bring, while only 37 percent said it is better to be fast and wrong than right and slow. That figure is about technology in general, not AI alone, but I would be surprised if AI were the exception.",[11,138,139],{},"A proposal built only to answer the mandate is answering a different question from whether customers get more value at a margin you can defend. That, to me, is what gambling with AI in your own business looks like: nobody can say what they expect back, so the excitement does the work the numbers should be doing.",[11,141,142],{},"The reality is that this is how business cases, use cases and AI initiatives that waste money get approved. They look like progress in a board pack because the technology is exciting. Somebody in the room has to ask for the numbers before the mandate gets to spend them.",[32,144,146],{"id":145},"root-causes","Root causes",[11,148,149],{},"When I look at why AI returns keep falling short of what was promised, I think it comes down to two root causes. Either the business made the wrong type of AI investment, chasing the mandate rather than the customer, or it made a reasonable investment but never anchored it to one of the four fundamentals, so nobody can say afterwards whether it worked.",[11,151,152,153,158,159,164],{},"The African numbers make the point sharper. A ",[18,154,157],{"href":155,"rel":156},"https://www.ecofinagency.com/news-finances/0906-56276-big-african-companies-invest-just-2-of-revenue-in-ai-pwc-finds",[22],"PwC study covering 85 large African companies"," found they invest about 2 percent of annual revenue in AI, against about 5 percent among comparison companies in countries such as the United States, China, Germany and France. ",[18,160,163],{"href":161,"rel":162},"https://itweb.co.za/article/ai-roi-remains-elusive-for-african-firms/KzQenqjyj3jMZd2r",[22],"ITWeb reported"," a PwC finding that 82 percent of organisations in Africa have taken part in AI pilots, and ITWeb also reported that among African CEOs who have invested in AI, 23 percent reported revenue increases and 25 percent reported cost reductions over the past year. These are continent-wide and self-reported, not South African figures. Still, I read them as the second root cause at work: plenty of pilots, too few of them anchored to a fundamental. With a smaller budget there is even less room to spend the tokens in ten minutes and learn nothing.",[32,166,168],{"id":167},"the-innovation-lab","The innovation lab",[11,170,171],{},"A good example of money wasted on the wrong type of investment is a physical AI innovation space, or an AI innovation lab at your office, built before any customer value has been proven. That is a waste of money, and I say so as someone whose company runs hackathons and AI programmes. The work is fine. Paying for the showpiece room before the work has produced anything is the problem, because the room exists mainly to be seen.",[11,173,174,175,180],{},"Innovation centres were being built long before AI. In 2017, years before the current wave of generative AI, ",[18,176,179],{"href":177,"rel":178},"https://www.cdotrends.com/story/13587/capgemini-innovation-centers-are-not-delivering",[22],"CDOTrends reported a Capgemini study"," of 1,700 employees at 340 firms: 87 percent had an innovation centre, no firm had reached the top \"optimized\" maturity level, and only 17 percent thought they had a company-wide culture of innovation. Capgemini's Lanny Cohen said organisations need to accept that they cannot just open innovation centres and expect an overnight transformation. Structured, outcome-tested innovation work can pay for itself, and a room on its own does none of that work.",[32,182,184],{"id":183},"the-dropbox-panda","The Dropbox panda",[11,186,187,188,193,194,199],{},"The comparison I keep coming back to is Dropbox's chrome panda. In 2016 Dropbox installed a five-foot chrome statue of its panda mascot in its new San Francisco office, reportedly at a cost of around $100,000, according to ",[18,189,192],{"href":190,"rel":191},"https://www.entrepreneur.com/article/275450",[22],"Entrepreneur"," and ",[18,195,198],{"href":196,"rel":197},"https://betakit.com/hootsuites-ryan-holmes-on-the-wrong-way-to-run-a-billion-dollar-company-in-2016/",[22],"BetaKit",". According to BetaKit, a note posted beside it said, in part, that \"it wasn't the right call\", that \"every dollar counts\", and it asked staff, \"Would I spend my own money this way?\" Dropbox also said it was keeping the panda as a reminder of \"thoughtful spending\".",[11,201,202],{},"I would not hold this up as a villain. A company that names its own error in front of its own people is doing something many companies do not. My reading, and it is only my reading, is that the panda was frivolous spending to look successful. Looking successful and actually being successful are two different things when it comes to enterprise leadership, and an AI showroom is a very expensive way of confusing the two.",[11,204,205,206,211],{},"The same confusion turns up on the selling side. In January 2025 the SEC found that ",[18,207,210],{"href":208,"rel":209},"https://www.sec.gov/enforcement-litigation/administrative-proceedings/33-11352-s",[22],"Presto Automation"," had made materially false and misleading statements about its AI drive-thru ordering product, Presto Voice. The order says that, in fact, \"the vast majority of drive-thru orders placed through this version of Presto Voice required human intervention.\" Presto consented to the order without admitting or denying the findings. Before you fund the demo, ask what is actually running behind it.",[32,213,215],{"id":214},"earlier-technology-waves","Earlier technology waves",[11,217,218,219,224],{},"In December 2017 Long Island Iced Tea Corp. renamed itself Long Blockchain Corp. In a 2021 ",[18,220,223],{"href":221,"rel":222},"https://www.sec.gov/news/press-release/2021-121",[22],"press release on its complaint",", the SEC said the share price spiked more than 380 percent intraday after the company's press release, and it charged three individuals with insider trading on advance knowledge of that announcement. To me that is an extreme case of a label doing the work a business case should do, and I see the same instinct whenever an AI proposal gets funded on its name rather than its substance.",[11,226,227,228,233],{},"The size of a long technology bet is the other lesson. Meta's Reality Labs, the segment behind its virtual and augmented reality products, recorded an operating loss of $19.2 billion in 2025 on ",[18,229,232],{"href":230,"rel":231},"https://investor.atmeta.com/investor-news/press-release-details/2026/Meta-Reports-Fourth-Quarter-and-Full-Year-2025-Results/default.aspx",[22],"Meta's own figures",", and Meta expects losses of a similar size in 2026. Meta treats it as a long-term investment, and it can fund it from a very profitable core business. I am not calling it a failure. My point is narrower: almost nobody reading this can fund a bet on that basis, so nobody reading this should price their own AI bets as if they could.",[32,235,237],{"id":236},"frugality-and-the-same-quality-gates","Frugality and the same quality gates",[11,239,240],{},"So be more frugal about what you spend on AI, and more thoughtful about why. Insist on the same quality gates you would apply to any other investment, once you strip the AI sexiness out of the proposal. If the case still stands up without the letters AI in the title, fund it, and expect the result to look boring. That, far more than any lab or launch event, is what senior business leaders need to focus on.",[11,242,243],{},"When the next AI proposal reaches your desk, ask who the customer is and what they get. Ask how often the value actually arrives, rather than how well it worked in the demo, and what it costs to run the same thing everywhere rather than in the one pilot everyone has seen. Ask what margin is left once you have paid for all of it, and what the business will have learned when the pilot ends, so the tokens buy a lesson even when they do not buy a result. Then ask the question from the note beside Dropbox's panda: \"Would I spend my own money this way?\" Treat the proposal the way you would treat a new product line or a new market, because that is what it is.",{"title":245,"searchDepth":246,"depth":246,"links":247},"",2,[248,249,250,251,252,253,254,255,256,257,258],{"id":34,"depth":246,"text":35},{"id":44,"depth":246,"text":45},{"id":60,"depth":246,"text":61},{"id":88,"depth":246,"text":89},{"id":104,"depth":246,"text":105},{"id":126,"depth":246,"text":127},{"id":145,"depth":246,"text":146},{"id":167,"depth":246,"text":168},{"id":183,"depth":246,"text":184},{"id":214,"depth":246,"text":215},{"id":236,"depth":246,"text":237},"Opinion","AI Dividend","2026-10-13","AI spending drifts into theatre when executives chase the wow factor and the board's AI mandate. The fix is four ordinary business tests: value, consistency, scale and margin.","md","/images/insights/ai-investment-casino-gambling.webp","A small team standing at a wall of sticky notes in a working strategy session, in a magenta and blue duotone","Primary: AI investment mistakes. Secondary: AI hype; AI business case; AI innovation lab; board AI mandate; AI return on investment; AI running costs; AI pilot to production",{},"Francois van der Merwe on why AI spending drifts into theatre, and the value, consistency, scale and margin tests every AI investment should pass.",true,"/insights/ai-investment-casino-gambling",13,{"title":5,"description":262},"Are You a Casino Gambler With AI Investments?","insights/ai-investment-casino-gambling",null,"rwBCX5rPUbejIJxzAakGf48vG_36OK28KZvgK6ElH08",1791893955238]