Instead, many are watching costs climb while the promised gains lag behind.
Across global boardrooms, AI is shifting from shiny trophy project to awkward line item on the balance sheet. The hype has not vanished, but optimism is colliding with spreadsheets, and the numbers tell a far messier story than last year’s presentations.
Executives confront an AI hangover
For several years, senior leaders have repeated the same mantra: invest big in AI now, reap huge savings later. That narrative is starting to crack. A global survey by PwC of 4,454 executives across 95 countries shows that a majority are not yet seeing the financial benefits they expected from AI deployments.
According to PwC, 56% of executives say AI has neither increased revenue nor reduced costs over the most recent fiscal year.
This is far from the gold rush many anticipated. Nearly 30% of respondents reported some revenue growth linked to AI projects. Yet only 12% said they had achieved the holy grail: higher revenue and lower costs thanks to AI in the same period.
That 12% figure matters. It shows AI can deliver measurable productivity improvements and margin gains, but that success remains confined to a relatively small group of companies that have managed to integrate the technology properly.
The mirage of easy savings
Some firms moved faster than others, chasing aggressive cost-cutting through automation. A few even boasted about replacing swathes of human staff with AI systems, promising millions in savings. Several of those stories have quietly turned into cautionary tales.
Tasks that looked easy on PowerPoint proved much harder in reality. Automated customer service bots struggled with nuance. Generative models produced confident but wrong information. Operational glitches led to service failures and reputational damage. Several companies that rushed layoffs later had to rehire or rebuild teams, absorbing double costs.
AI still struggles to fully replace human workers in complex, real-world settings. In many cases, the technology adds new costs before it creates dependable savings.
That gap between promise and reality is feeding a new kind of anxiety in the C-suite: not fear of missing out on AI, but fear of never recouping the money already spent.
➡️ Fine hair after 60: these 3 hair colors are the ones that age the face the most, according to a hairdresser
➡️ Details on the mission of 700 French soldiers facing the Middle East’s most volatile border
➡️ Saab launches second Polish SIGINT ship
➡️ For millennia men avoided crossing the Taklamakan Desert; today China raises fish there
➡️ A polar vortex disruption on March 2, 2026 enters official high-impact scenario, “cold Arctic air could spill southward,” explains meteorologist Andrej Flis, mauvaise nouvelle for travel
➡️ What is a baby squirrel called?
➡️ Long-term discomfort for workers who want to work from home: they should come to the office despite financial and environmental reasons – experts warn that remote work does more harm than good
➡️ Why people hang a bay leaf on the door and what it’s for
AI is not plug and play
One of the biggest misconceptions was that AI tools would behave like typical enterprise software. You plug them into existing systems and workflows, train staff, and watch efficiency rise. The experience has been very different.
In many organisations, AI lives inside isolated pilots or small experiments. A team tests a chatbot here, a fraud model there, perhaps a generative assistant for marketing content. These pockets of innovation rarely reach scale or connect to mission-critical processes, where the largest gains would sit.
AI adoption is widespread on paper, but shallow in practice. Many companies are experimenting, far fewer are transforming how they actually work.
Executives interviewed around the Davos summit, where PwC presented its findings, acknowledge the gap. Strategy documents still treat AI as a central pillar of future competitiveness. Budget lines for AI continue to grow. Yet the operational reality on the ground is slower and messier than those ambitions suggest.
Why most AI projects stall before delivering ROI
Several patterns keep emerging when you look at failed or underperforming AI programmes. They tend to cluster around a few recurring problems.
- Fragmented initiatives: Multiple pilots run in parallel, owned by different departments, with no clear overarching roadmap.
- Weak data foundations: Poor data quality, siloed databases and unclear data ownership limit what AI can do.
- Overreliance on vendors: Companies outsource the thinking as well as the building, then struggle to adapt systems internally.
- No process redesign: Old workflows remain intact; AI is simply bolted on top, diluting any potential gains.
- Limited change management: Staff receive new tools, but not training or incentives to actually change behaviour.
On top of that, AI itself is not as reliable as many executives assumed. Generative models still hallucinate facts. Prediction systems fail when the underlying data shifts. And integration into complex legacy IT systems can be slow, expensive and risky.
Generative AI: hype meets hard numbers
A separate report from MIT last year added another layer of realism. It found that around 95% of attempts to integrate generative AI into business operations had not produced rapid revenue acceleration.
Many companies launched generative pilots with great fanfare: automated copywriting, code generation, sales support, internal knowledge bots. While some early productivity boosts surfaced, they often remained at the level of time saved per employee rather than clear bottom-line impact.
In most firms, generative AI has so far behaved more like a smart assistant than a true growth engine.
For finance chiefs, that distinction is crucial. Shaving a few minutes off a task is helpful, but it does not always translate into fewer staff or significantly faster growth. The result is a growing tension between visible experimentation and invisible profit.
Data security and trust weigh on adoption
Even when the business case looks strong, risk officers raise serious concerns. Training or running models often means feeding them with sensitive internal data: customer information, pricing structures, contracts, product roadmaps.
Once that data flows into third-party models, many executives worry about where it might resurface. Could confidential content leak into outputs generated for another client? Will providers use corporate data to train future models? Regulatory guidance is still catching up, leaving firms exposed to legal and reputational risk.
This uncertainty pushes some organisations to keep AI experiments small and low-stakes. That caution protects them from major missteps, but it also delays any large-scale financial payoff.
Will companies slam the brakes on AI?
Despite the disappointing early returns, there is little sign of a full retreat. Boards largely see AI as a long-term strategic necessity, not a passing gadget. The fear of falling behind competitors still outweighs the discomfort of the current ROI figures.
| Executive attitude to AI spending | Typical reaction |
|---|---|
| Early financial disappointment | Adjust expectations, but maintain or slightly increase budget |
| Regulatory and security concerns | Push for more internal control and stricter governance |
| Clear productivity wins in pilots | Scale selectively to a few core processes |
The more likely outcome is a shift in focus. Instead of chasing headline-grabbing automation or staff cuts, companies are beginning to ask where AI can realistically enhance existing teams, reduce specific bottlenecks or open narrow but valuable new revenue streams.
What a realistic AI strategy looks like
Firms that belong to the 12% group with clear financial gains share some common traits. They treat AI as part of broader transformation, not as a magical layer on top of existing practices.
Three moves stand out:
- Start with a process, not a model: They pick concrete workflows where delays or manual errors are costing real money, then design AI around that.
- Invest in data plumbing: They fix data pipelines, governance and access rights before chasing sophisticated algorithms.
- Measure like a CFO: They track not only technical metrics but also revenue, margin, churn and customer satisfaction.
For instance, an insurer might begin with claims triage, using AI to flag likely fraud or fast-track simple cases. The business outcome is clearer: fewer losses, faster payouts, happier customers. A manufacturer might target predictive maintenance to reduce downtime on key machines, directly protecting revenue.
Key terms executives keep debating
Behind closed doors, board members often grapple with jargon that shapes expectations. Two notions come up repeatedly:
Return on investment (ROI): In the AI context, ROI is not just about cost cuts. It can include faster product launches, better risk detection, higher conversion rates or lower employee turnover. Firms that only look for immediate headcount reduction tend to miss more subtle, but durable, gains.
Total cost of ownership (TCO): AI projects carry hidden costs: cloud computing, integration work, security reviews, model retraining, regulatory compliance and new specialist hires. When these are underestimated, projects that look profitable on paper can quickly drift into the red.
Scenarios for 2026: from hype to harder choices
PwC points to 2026 as a decisive year for AI in business. By then, many multi-year programmes will either start paying off or face serious questioning. Several scenarios are plausible.
In one scenario, a critical mass of firms finally crack integration: AI becomes embedded in finance, operations, logistics and customer service. Gains arrive gradually, but they compound. The early disappointment phase then looks more like a learning curve.
In another scenario, only a minority succeed, widening the gap between digital leaders and laggards. Boards at underperforming companies could grow impatient, cutting budgets and retreating to safer, smaller-scale uses of AI, mostly as internal productivity tools rather than growth drivers.
Either way, the next two years will pressure executives to move beyond slogans. The conversation shifts from “How much are we spending on AI?” to “Which specific processes are better, cheaper or faster because of it?”
Practical risks and how firms are responding
As AI spending matures, risk management becomes as central as innovation. Companies are setting up internal AI councils, new approval workflows and stricter principles for what data can feed external models.
Common risk responses include:
- Building private, company-specific models that never leave internal infrastructure.
- Segmenting data so the most sensitive information is never exposed to third-party tools.
- Requiring thorough human review for any AI output that affects legal, financial or safety decisions.
The trade-off is clear: more controls can slow experimentation, yet they also reduce the chances that early AI bets turn into costly scandals or regulatory battles.
For now, CEOs are juggling three pressures at once: investor enthusiasm, operational reality and mounting scrutiny. The return on investment of AI has become less of a marketing slogan and more of a genuine headache, forcing leaders to treat the technology not as a miracle, but as another hard business problem that rewards patience, precision and a clear sense of where value truly lies.