Boardrooms bet big on artificial intelligence, but the dazzling future they were promised is colliding with a much harsher reality.
Executives from New York to New Delhi are discovering that generative AI tools are far from a magic cash machine. Huge projects have gone live, staff have been cut, and yet the numbers on quarterly reports often look stubbornly flat. The question now is not whether companies will quit AI, but whether they can finally make it pay.
The great AI payoff that never came
A global PwC survey of 4,454 business leaders across 95 countries has jolted the AI hype cycle. The majority of bosses who poured money into AI projects with the clear goal of boosting profits say the expected financial lift simply did not arrive.
According to PwC, 56% of executives report that AI has neither raised their revenues nor reduced their costs in the last fiscal year.
That figure clashes sharply with the confident rhetoric of the last two years, when AI was sold as the quickest route to leaner operations and fatter margins. Many boards approved aggressive spending on data infrastructure, software licences and specialist teams on the assumption that payback would be swift.
The reality is more nuanced. Nearly 30% of surveyed leaders do see higher revenue linked to AI initiatives. Yet the real prize – higher sales and lower costs at the same time – remains rare. Only about 12% of companies say they have reached that coveted sweet spot.
This gap between narrative and outcome is forcing some quiet soul‑searching in executive suites, even as public messaging stays upbeat.
From miracle worker to expensive experiment
The faith in AI’s economic power has already led to harsh decisions. Some firms boasted about multi‑million cost savings after replacing whole teams with automation and generative models. Several of those same companies later had to row back, re‑hire, or patch over service failures when the systems did not perform as promised.
Geoffrey Hinton, the Nobel laureate often described as a godfather of AI, has warned that the technology becomes truly lucrative only when it can replace a large share of human workers. That prediction has spooked unions and regulators. Yet the current evidence suggests AI still struggles to fill those roles reliably.
Companies that rushed to swap people for algorithms often faced unhappy customers, operational errors and a quiet increase in hidden costs.
➡️ Pressure mounts on NASA: the space station is nearing its end and the handover is not secured
➡️ For every dessert, its apple! The ultimate guide to finding the right apple to use
➡️ The first cities may not have been in Mesopotamia… but here
Support chatbots that cannot solve basic problems, automated document systems that hallucinate details, or AI‑generated code that introduces subtle security flaws: all of these can erode trust and require expensive human intervention to fix.
The AI mirage in numbers
PwC presented its findings at the Davos summit, framing 2026 as a decisive year for AI in business. By then, many large companies will have spent several years experimenting at scale. Patience, especially on listed markets, is not endless.
At the same time, a study by MIT cited by industry experts paints a similar picture on generative AI specifically. It suggests that roughly 95% of attempts to plug generative AI into enterprise operations have failed to produce rapid revenue acceleration.
- 56% of leaders see no AI effect on revenue or costs
- ~30% report AI‑linked revenue growth
- 12% enjoy both revenue growth and lower costs from AI
- 95% of generative AI pilots have not delivered rapid sales gains
The numbers are not a verdict that AI “doesn’t work”. They show instead that converting pilot success into company‑wide financial impact is far harder than early marketing suggested.
AI is not plug and play
One of the bluntest lessons from the last two years is simple: AI is not a mouse you plug into a laptop. It refuses to behave like a neat, predictable accessory.
Many firms treated AI as a badge of modernity. Having “an AI strategy” helped attract investors, partners and scarce tech talent. In practice, this often meant scattered pilots: an AI assistant here, a prototype customer‑service bot there, some experimental analytics in a single department.
Most corporate AI remains stuck in pilot mode, disconnected from the core processes that actually generate profit.
These pockets of experimentation look good in presentations but rarely touch the systems that move the needle: pricing, supply chains, risk management, product development. Integration is hard. Legacy IT, siloed data, and internal politics get in the way.
Executives commonly underestimate the organisational heavy lifting required. AI systems must be fed clean, consistent data, embedded into workflows, aligned with compliance rules and monitored continually. That means rethinking processes, retraining staff and, in some cases, redesigning products.
Where AI is hitting a wall
Beyond strategy flaws, the tools themselves still pose technical and legal headaches.
- Hallucinations: Generative models can invent facts with great confidence, a serious risk for legal, medical or financial work.
- Basic task failures: Seemingly simple operations, like matching records or extracting figures from messy documents, still confound many systems in real‑world environments.
- Data security: Firms fear that feeding confidential information into cloud‑based models could leak trade secrets or personal data through unintended outputs.
- Regulation: New rules in Europe, and growing scrutiny in the US, raise questions over liability when AI systems misfire.
When these issues surface, companies often respond by adding extra human checks. That protects against damage, but it also eats into the cost savings that AI promised in the first place.
Why CEOs still keep writing the cheques
Despite the disappointments, there is little sign of a pullback. If anything, senior leaders say they plan to spend more on AI in the next three years, not less.
Fear of missing the next productivity wave outweighs frustration with early returns.
No one wants to be the CEO who switched the AI budget off just before a competitor cracked the right formula. There is also pressure from shareholders who see AI as a long‑term structural shift similar to electrification or the internet. Missing that shift could be far more costly than a few failed pilots.
The question is shifting from “Should we invest?” to “How do we stop wasting money?” Many boards are now pushing for clearer metrics: revenue per AI project, time saved per employee, customer retention impact, and compliance risk reduction.
What a realistic AI ROI strategy looks like
Firms that report meaningful returns tend to share a few traits. They are not the ones throwing the most cash at shiny tools, but those that pick their battles carefully.
| Approach | Likely outcome |
|---|---|
| Mass layoffs based on AI promises | Service quality drops, hidden costs rise, brand damage |
| Small pilots with no path to scale | Nice demos, little financial impact |
| Targeted use on repetitive back‑office tasks | Moderate, measurable savings and happier staff |
| Deep integration into core products and pricing | Slower to build but higher potential for lasting ROI |
Practical examples include AI‑driven fraud detection in banking, which can reduce losses while improving speed, or AI scheduling in logistics that trims fuel usage and overtime bills. These are not glamorous chatbots on a homepage; they are tools buried deep in operations.
Key terms CEOs are finally asking about
As the easy stories fade, some technical jargon has moved to the centre of board discussions:
- Hallucination rate: How often a generative model produces incorrect or fabricated content.
- Model governance: The rules, oversight and documentation around how AI systems are trained, tested and updated.
- Data lineage: The ability to track where the data used by an AI system came from and how it has been transformed.
- Total cost of ownership: All the hidden costs beyond licence fees: cloud compute, integration work, staff training, security and compliance.
Getting a grip on these concepts helps leaders move from hand‑waving promises to hard questions about risk, accountability and genuine efficiency gains.
Scenarios: what the next three years could look like
Two broad paths are taking shape. In one, companies keep chasing eye‑catching AI announcements, scattering pilots across departments. ROI stays weak, while fatigue sets in among staff asked to work with half‑baked tools.
In the other, firms narrow their focus. They pick a few critical workflows – for instance, claims processing in insurance or procurement in manufacturing – and invest in careful redesign. They accept a slower rollout but track every pound or dollar saved, every minute shaved off a process, and every legal risk reduced.
The biggest returns may go not to the boldest spenders, but to the most disciplined implementers.
There is also a cultural angle. Companies that treat AI as a partner for staff, rather than just a replacement, often see smoother adoption. Training employees to question AI output, report issues and suggest improvements can turn early mistakes into learning signals rather than PR disasters.
The ROI debate around AI is only just getting started. As the first adrenaline rush fades, the technology is entering a tougher, more accountable phase. For many CEOs, the new anxiety is not that AI will take over their business, but that after all the promises, it still might not pay for itself.