Businesses that rushed to invest millions in AI tools are finding that the profits they expected are late, small, or not there at all. This is happening in many industries and on many continents. The technology is moving quickly, but the financial case is not as clear-cut as many flashy presentations made it seem.
AI’s shine meets the spreadsheet
For a long time, people have been saying that AI is the next big thing in productivity, able to lower costs, speed up work, and create new ways to make money. Boards were afraid they would be left behind if they didn’t act quickly. Consultants, tech companies, and investors all pushed the idea that businesses should automate now and make money later.
New information tells a different story. PwC did a survey of 4,454 business leaders in 95 countries. It found that more than half of the companies that put a lot of money into AI have not seen the financial gains they were promised.
PwC says that 56% of executives say that AI has not helped them make more money or cut costs in the most recent fiscal year.
That number goes against what most people think. Company reports and keynote speeches at conferences still talk about “transformational potential” and “huge efficiencies.” The mood in the finance department is not as happy.
There are stories of success, but they don’t seem as big as the headlines make them out to be. About 30% of the leaders who were asked said that AI projects have helped them make more money. Only 12% say they have reached the holy grail: AI has helped them make more money and cut costs.
The illusion of quick AI profits
The difference between what people thought would happen and what actually happened is partly because of how AI investments were presented. Many boards thought AI was a sure way to save money quickly, especially by cutting staff. Some companies openly bragged about getting rid of a lot of workers and replacing them with AI systems.
Those tests often didn’t work. Companies learned that chatbots and automation tools had trouble with nuance, judgment, and complicated real-world situations. Customer service got worse, internal processes came to a halt, and hidden costs rose in the form of manual fixes and damage to the company’s reputation.
At least with the tools and data we have today, AI is better at helping people do their jobs than replacing them completely.
The cost of widespread AI adoption is much higher than just paying for a software subscription. Companies have to pay for cloud space, specialized hardware, integration work, new data pipelines, cyber security, and ongoing model tuning. Those costs show up on the profit and loss statement long before real savings do.
Why AI isn’t “plug and play”
A lot of executives now see that they treated AI like a simple extra. You get a license, connect it to your systems, and your productivity goes up. Reality is more like a long, messy renovation than a new gadget plugged into a wall socket.
In many companies, AI is still stuck in a bunch of small pilot projects: a customer service prototype here, a forecasting test there, and an internal assistant for emails. These projects often look great in demos, but they never really get fully deployed and start making money.
Innovation labs, not core business units, are where projects happen.
It’s not clear who owns what because IT, data teams, and business lines all want different things.
AI is added on top of processes that aren’t working well, so they don’t work better.
Staff aren’t trained well, which makes people not trust them and not use them enough.
Last year, a report from MIT that was widely cited pointed in the same direction for generative AI in particular. It said that 95% of the time people have tried to use generative AI in business settings, they have not yet seen a quick increase in revenue. That doesn’t mean failure forever, but it does show how new and experimental most deployments still are.
The technical problems that executives didn’t plan for
AI technologies have their own problems that make them less useful for businesses, in addition to the problems they cause in organizations. Modern generative systems are known for “hallucinations,” which means they give outputs that are sure but wrong. That can cause confusion and put you at risk of legal or compliance issues when dealing with customers.
Models can miss context or mess up edge cases, even for simple tasks like summarizing documents or suggesting answers. After that, human staff has to double-check the results, which slows down the promised efficiency gains.
Instead of fully automating their processes, many companies end up with a “human plus AI” workflow that costs money and takes time to get stable.
Another problem is data security. AI tools need a lot of business information, like private documents and personal data, to work well. Leaders are still worried about where that information goes, who can see it, and whether a model might make sensitive content available to someone else.
Even when vendors offer private instances and strict controls, legal and compliance teams don’t move quickly. AI projects are stuck while teams work out how to share risks, do audits, and make sure contracts are followed, things that weren’t possible a few years ago.
Even though there are doubts, investment is going to grow.
One interesting thing about the present moment is that disappointment isn’t leading to retreat. At the Davos summit, PwC shared its results and still called 2026 a key year for AI in business. Most of the leaders who took the survey said they would raise, not lower, their AI budgets.
Part of this is the fear of missing out. No business wants to admit that it picked the wrong technology cycle or tell investors why its competitors seem to be ahead of it. AI has also become a way to brand yourself: saying you are a “AI-driven” company can help you get both money and talent.
At the same time, many executives say they have learned from early missteps. The next wave of spending is expected to shift away from isolated experiments and towards deeper integration in a few key processes where value is clearer, such as supply chain optimisation, fraud detection, or targeted marketing.
Rethinking how to measure AI ROI
Traditional return-on-investment calculations struggle with AI. The benefits are often indirect, delayed or hard to measure. Time saved by employees, better decision quality, fewer errors, and increased resilience do not always show up as a neat line on a quarterly report.
Some firms are starting to adopt a more nuanced approach. Instead of asking, “Did AI cut headcount this year?”, they look at a broader set of indicators.
Aspect Example AI impact
Productivity Faster report drafting, reduced time spent on routine emails
Quality Fewer manual errors in data entry, more consistent customer replies
Speed Quicker decisions on pricing, logistics and resource allocation
Innovation New product ideas and prototypes generated at lower cost
Risk Earlier detection of anomalies, fraud or system failures
These metrics still link back to profit, but over a longer horizon and with less linear cause-and-effect. Investors are watching to see which companies manage this transition with discipline and which are simply relabelling old IT projects as “AI transformations”.
What “good” AI integration could look like
Behind the averages, some firms are genuinely extracting value from AI. Their stories share certain traits that other leaders are keen to copy.
- They select a small number of high-impact use cases instead of scattering pilots everywhere.
- They redesign workflows around AI outputs, rather than just bolting tools onto existing steps.
- They invest in data quality, governance and security before scaling models.
- They train staff to work with AI systems and encourage feedback from the front line.
- They accept an initial period of lower productivity during the transition.
Imagine a logistics company using AI to plan delivery routes. If it simply deploys a model and keeps all the old manual checks, drivers, planners and managers may ignore the recommendations. If, instead, it rebuilds scheduling processes, aligns incentives, and integrates AI into vehicle tracking and fuel management, the system can reduce delays and costs in a measurable way.
The same principle applies in banking, healthcare, retail or manufacturing. The technology is only one part of the investment. Organisational change, training and governance often cost just as much, yet they are the pieces that unlock real ROI.
Key terms leaders are wrestling with
Behind closed doors, many executives are still getting up to speed with basic AI vocabulary, which shapes how they think about strategy.
Generative AI refers to systems that create content: text, images, code, audio and more. These tools are powerful for drafting, brainstorming and summarising but remain unpredictable without guardrails.
Hallucination describes the tendency of some AI models to produce information that sounds plausible but is factually wrong or fabricated. For customer-critical or regulated activities, this behaviour demands tight supervision.
Data governance covers the rules, processes and technologies that control how data is collected, stored, shared and used. Weak governance increases the risk of privacy breaches, biased outputs and regulatory penalties.
Scenarios for the next few years
If PwC is right that 2026 could be a turning point, several paths are possible. In one scenario, organisations slowly align their structures, data and incentives with AI tools. Margins improve gradually, not dramatically, but firms that stay the course gain a quiet edge over laggards.
In another scenario, frustration with poor returns and growing regulatory scrutiny pushes some boards to freeze ambitious AI projects and focus only on tightly controlled, low-risk automations. Capital shifts from grand transformation claims towards more concrete, narrow tools.
A more optimistic path combines both caution and ambition. Companies accept that AI is not a magic cost-cutting machine and use it instead as a force multiplier for skilled employees. Under this model, AI handles routine tasks while people spend more time on judgement, creativity and relationship building. Productivity grows, but not at the speed that early hype suggested.
For now, the numbers reveal a simple truth: AI as a business investment remains a high-stakes experiment. The leaders who succeed will be those who treat it less as a shiny gadget and more as a long-term restructuring of how their organisations think, work and measure value.
