6:34
AI was supposed to be the biggest economic story in history. A new revolution, trillions in spending, "Worldwide AI spending will total $2.59 trillion in 2026," So, naturally, someone should be able to prove it made them money. Right? Well… not really. 6,000 senior executives were surveyed across the US, UK, Germany, and Australia. 69% said their firms are actively using AI. But 89% reported no effect on labour productivity. Another study found, despite $30-40 billion in spending, 95% of AI pilots produced no measurable impact on profit. We've been using AI for a while now. So, has anyone actually made money from AI? In some areas, it sure seems to be doing the opposite.
Why is it so difficult to show that AI made any impact? Are the end businesses using AI, actually seeing financial returns? To be clear: AI has made some companies a lot of money. Nvidia. Cloud providers. Model vendors like Anthropic, consultants, AI API providers, etc. What I'm talking about is the end customer. Those paying for ChatGPT and AI tools, and integrating them into the end business. So, what are the results? Well… A report by MIT found that, despite $30-40 billion in enterprise spending, 95% of Generative AI pilots in its sample produced no measurable
P&L impact, aka profit and loss impact. 95% of deployed AI pilots produced zero return. Something very interesting is happening. An enormous amount of AI activity is failing to survive the journey from impressive demo to an actual financial result… Another 2026 study by the National Bureau of Economic Research, which surveyed about 6,000 senior executives across the UK, US, Germany, and Australia, found that 69% of their companies were actively using AI… But, nine out of 10 executives reported no impact on employment or productivity at their own firm over the prior three years. These executives also predict, over the next three years, AI will boost productivity by an average of 1.4%, raise output by just 0.8%,
and cut employment by just 0.7%. So, not much change. In McKinsey's 2026 global AI survey, 37% of respondents said AI contributed positively to enterprise-wide EBIT. Sounds promising, but this is unchanged from the survey from the prior year, despite deployment increasing. And the proportion that were AI high performers, meaning a company who attributes at least 5% of EBIT to their use of AI… was just 6% of all respondents. So, why is AI, this technology that is seemingly changing the world, so at odds with "increasing" revenue. AI has clearly made some companies money, but doesn't it seem there's a big missing piece, that justifies all this crazy spend?
The companies buying AI, the ones meant to become more productive, more profitable, and more innovative, are mostly struggling to show it in the numbers. Why doesn't AI show up in revenue?… And if genuinely useful AI does not reliably translate into profit… where does the value go? AI is making a change in some places. Customer support, for example. AI help resolved 14% more issues per hour, on average, and the least experienced saw gains of up to 35%. Of course, a "resolved" ticket,
doesn't always mean a satisfied customer. But, this isn't an actual economic result. A support agent handling 14% more issues per hour does not mean the company made 14% more money. In other words, there are only very specific cases where AI sees tangible, economic benefits. Klarna claimed its AI assistant handled 80% of customer-service chats in 2025, and said consumer satisfaction did not decline, which resulted in an estimated $58 million in annualised cost savings. So repeatable, digital, high-volume tasks and a clear metric, like "resolved", seem to work well with AI. This sounds like the answer,
but there are a few caveats. Most tasks in work, don't fall into this specific case: repeatable, high volume, clear metric, that perfectly correlates to revenue. And secondly, and this is the big one, cost avoidance, is not an increase in revenue. Klarna may be making more profit from AI, but it is not evidence that AI is generating more revenue. Most of the AI revolution has been sold as growth: new products, new markets, and radically higher output-not merely as more efficient customer-service. And, that's mostly what we're seeing. Let's have a closer look at those surveys again. From the MIT study, where 95% of organisations have no return, they found: "These tools, [ChatGPT and CoPilot], primarily enhance individual productivity,
not P&L performance. Most fail due to brittle workflows, lack of contextual learning, and misalignment with day-to-day operations." And I think that last part, is very interesting. We see this in the McKinsey 2026 survey: "individual employees are reporting real impact: Eight in ten respondents say AI has improved their own productivity. However, enterprise-level financial impact hasn't followed the same trajectory." So we see that AI is great at improving "efficiency", for an individual… but this is the problem. That doesn't directly translate to more money.
Think about AI just being plugged into an organisation, as is. AI writes sales emails faster, but does that improve how many people will buy your product? AI writes code faster, and you might ship more products quicker, but there's still the bottleneck of QA, architecture, security, product decisions, and the most important one: customer adoption and demand. AI summarises meetings… but, those meetings still happen. It helped some employees do tasks faster, but the company doesn't actually generate more revenue.
Cost savings? Maybe, but an increase in revenue? We're not seeing that. And, this goes even deeper. The personal version of AI is easy, the enterprise version, is very difficult. The MIT study also found: "Generic tools like ChatGPT are widely used, but custom solutions stall due to integration complexity and lack of fit with existing workflows" "The hype on LinkedIn says everything has changed, but in our operations, nothing fundamental has shifted. We're processing some contracts faster, but that's all that has changed." If we look at disruption by industry, note, "disruption" doesn't mean revenue here, we see most industries have very little change.
"Only 5% of custom enterprise AI tools reach production. Chatbots succeed because they're easy to try and flexible, but fail in critical workflows due to lack of memory and customization." So it seems part of the mystery, is beyond "deployment", or "just using more AI", but the entire company and workflow around the AI. "High performers pursue growth and/or innovation alongside efficiency; they fundamentally redesign workflows that are enabled by AI rather than insert AI into existing ones" ChatGPT can draft an answer to a customer question… but, to solve the customer's problem,
it might need to be able to navigate their account, check the order, understand the contract, have authority to issue refunds, interact with the CRM, access data across multiple dashboards. And it's not allowed to do any of those things. And for good reason. With those permissions, mistakes can be disastrous and expensive. But, we're at a weird paradox here. The AI can write the refund email. It just cannot be trusted to refund the right person, for the right amount, under the right policy, without staff writing and building lots of infrastructure and information for it. Enterprise tools stall because of brittle workflows. Maybe not brittle for humans, but brittle for AI models.
Another PwC survey found that most companies were lacking the "AI foundations, such as clearly defined road maps and sufficient levels of investment" to realise a return. To really get the most out of AI, you sort of have to build lots of the business for AI. We've actually seen something very similar to this before. In Stanford economist Paul David's 1990 paper "The Dynamo and the Computer", he found that managers typically overlaid "one technical system upon a preexisting stratum". In the late 19th century, when factories that originally used a waterwheel or steam engine, first introduced electricity, managers simply replaced the old power sources with new electric motors. "They enjoyed modest cost savings on fuel as well
as slightly improved control, but their factories continued to function exactly as before." This is the Dynamo Fallacy, and doesn't this sound familiar? "Electrification changed manufacturing, but only after managers became willing to redesign their entire businesses around the fundamental capabilities of electric machinery." Leaders falsely assume a groundbreaking, new technology, will immediately deliver massive productivity gains. Maybe, AI is struggling, as people are just putting another "technical system upon a preexisting stratum".
Maybe it is because most companies are using AI in the easiest possible way: as a better personal tool, inserted into an old organisation? But, how do you rebuild an organisation around AI? And even if you do, does that actually promise an increase in revenue? If you've watched this far, you probably like our videos! So, consider subscribing! It helps the channel and means you won't miss future videos. It's hard to say what this looks like, or what this looks like for AI. But, there are a few explanations on why this hasn't happened yet.
One is something called the J-curve. It's a theory that explains how productivity can fall or look weak early in rollout, because it takes reorganisation, new processes, and new training before benefits become visible. Like a J, it goes down first, then up. A bunch of people say this is why AI results are disappointing. But, we're not quite seeing signs of the curve turning upward. It just explains delayed results, not how you actually deliver those results. Next, is a more plausible theory I think: David J. Teece's theory of Complementary Assets. Profits from innovation often go not to the inventor or user of the technology, but to
whoever controls the scarce complementary assets. "If the complements are not owned by, or cannot easily be replicated by, the inventor, then rents will likely be drained away toward the complement owner(s), and the innovator will be in no better position than an imitator who controls relevant complements." I think this is fascinating. There are two parts to this, value creation, and value capture. They don't always happen at the same place. Even if AI makes you more efficient, the money goes to whoever owns the "scarce complement". The thing that is needed to
turn a technology into a valuable product or service that is hard to replace or obtain. Distribution, maybe the integrated tool providers, but most likely, data centre owners, and GPU capacity. Companies buying AI may save a little time, while the companies controlling chips, cloud capacity, enterprise distribution, or the customer relationship capture most of the economic upside. The companies convincing everyone that AI will bring about a massive return. So, has AI made anyone money? Obviously. But it's the companies selling chips, cloud capacity, models, and AI services, not the average business deploying AI.
The ones who control the assets, are the ones who seem to capture it. But, AI can be useful without translating to more money. Because, it's one thing to make your whole company more efficient. But, to actually capture that, financially? Or to prove that efficiency made you more money? That's not as clear. Something else very interesting is happening with AI. Namely, the most important part of the industry, is actually moving. Moving away from the models. This is a reason OpenAI is in a lot of trouble. They've promised about a trillion in spend,
but ironically, they're caught in the layer of the stack that might be the least valuable. AI models, might be slowly becoming a commodity. Click here to learn the rest of the story.
