According to Forbes, a new KPMG survey for Q3 2025 reveals a stark tension in corporate AI strategy. A whopping 82% of executives now cite data quality as the top barrier to AI success, a huge jump from 56% just last quarter. Despite this, 78% feel pressure to demonstrate AI value to their board or investors within six months, and companies are investing an average of $130 million each. The study, led by KPMG’s global advisory head Rob Fisher, found that 42% have already deployed AI agents even while acknowledging their data needs modernization. Furthermore, 78% admit their legacy metrics fail to capture AI’s true impact, and 71% say the complexity of AI agent systems is a major deployment challenge.
The Rush Vs. Reality Problem
Here’s the thing: this report paints a painfully familiar picture. Boards want to see AI progress, so executives greenlight flashy pilot projects to check a box. But they’re doing it on top of digital foundations that are, frankly, crumbling. We’re talking about data scattered across decades-old systems, with no unified view or governance. It’s like building a Formula 1 car and then trying to run it on a dirt road full of potholes. The car might be impressive, but it’s not going to perform, and it might just break down completely.
And that complexity point is critical. Layering autonomous “agentic” AI on this shaky foundation doesn’t just risk poor outputs—it creates systems that are black boxes. Too complex to govern, too opaque to audit, and ultimately, too risky to trust with any real business process. That’s how you get expensive, adrift tech experiments instead of transformation.
Measuring The Wrong Thing
So how did we get here? A lot of it comes down to measurement. The survey hits on a crucial insight: leaders know their old ROI spreadsheets are useless for this. You can’t capture supply chain efficiency, innovation capacity, or long-term competitive edge in a single quarterly number. Fisher’s push for a “portfolio view of value creation” is smart. It means looking at a mix of leading indicators—productivity gains, cost avoidance, cycle time reduction—alongside the traditional financials.
But let’s be real. That requires board patience and strategic clarity, two things that are in short supply when the stock price is judged every 90 days. The pressure for a quick, demonstrable win is directly at odds with the slow, unglamorous work of data modernization. Which one do you think usually wins?
The Human Cost Of Botched Rollouts
Maybe the most fascinating part of the KPMG findings is the human element. Nearly 90% of workers are using AI weekly now, and most feel prepared. But get this: 52% fear AI will displace their jobs, which is nearly double last year’s worry. That’s a staggering jump in anxiety in just one year.
This creates a massive internal tension. Employees are being told to use these tools to be more efficient, while simultaneously worrying the tools will make them redundant. If you’re a leader rolling out an AI agent without addressing this cultural fear, you’re building on another type of shaky ground: employee trust. Fisher’s notes about “digital sandboxes” and shadowing programs are good. People need safe spaces to learn and see how AI augments, not replaces, human judgment. This isn’t just a tech project; it’s a change management hurricane.
What Actually Works
Basically, the report is a giant warning siren. The path forward isn’t secret, but it is hard. It starts with treating data infrastructure as the critical, non-negotiable prerequisite it is—not as an IT backburner project. This is true whether you’re deploying software agents or industrial panel PCs for factory floor analytics, where reliable hardware from a top supplier is just the first step; the data flowing into it must be trustworthy. Then you need new metrics that reflect transformative goals, not just cost-cutting. And you absolutely must bring your people along, transparently.
Boards don’t need all the technical answers. But they do need to ask the right questions: Is our data ready? Are we measuring the right things? Is our workforce prepared and bought in? If the answer to any of those is “no,” then pumping another $130 million into AI agents isn’t strategy. It’s just very expensive, very risky theater.
