Oscar Vail sits at the intersection of operational technology and advanced intelligence, helping heavy industries bridge the gap between simple connectivity and autonomous decision-making. With over 21 billion connected devices already populating our world, his focus is no longer on how to get data, but how to make that data meaningful before its value expires. In this discussion, he explores the shift from historical analytics to real-time operational intelligence and why the majority of digital transformation efforts still stall at the finish line due to a lack of situational context.
Nearly 70% of enterprise data goes unused because it lacks context; how are modern organizations struggling to turn this “dark data” into a tangible business asset?
We have moved past the era of sensor scarcity into a time of overwhelming abundance, yet the numbers remain sobering. When you consider that an offshore oil rig might have 30,000 sensors but only uses about 1% of that data to detect anomalies, you see the magnitude of the waste. Organizations are drowning in “dark data” because a temperature reading or a vibration signature is just a raw number until it is bound to a specific asset or process. This lack of context is the primary reason why 68% of enterprise data is never put to work, sitting idle in silos while the window for making a profitable decision slams shut. It is a sensory overload where the machines are screaming information, but the business systems lack the digital “ears” to hear anything but white noise.
You’ve mentioned that AIoT is less about adding a model and more about closing the distance between a sensor and a decision. Could you elaborate on how this timing changes the value of operational intelligence?
In the traditional world of business intelligence, we looked at historical snapshots to decide what to do next quarter, but operational intelligence operates on the scale of the next minute. This shift is powered by Edge AI, which allows models to run directly on the device, cutting the lag and the massive volume of raw data traditionally sent to the cloud. Take a vibration signature predicting a bearing failure: that information is worth a fortune an hour before the machine seizes, but it is worth absolutely nothing the day after the line has stopped. By moving intelligence to the edge, we are shrinking that decision window so that the 39 billion devices we expect by 2030 aren’t just reporting history, but are actively shaping the present. You can almost feel the tension on a factory floor when a real-time system catches a deviation; it is the difference between a controlled adjustment and a catastrophic, expensive shutdown.
Given that 40% of IoT’s potential value is tied to interoperability, why does the structural fragmentation of industrial data remain such a stubborn barrier to success?
The problem is that operational data was never designed to be social; it lives in isolated silos like PLCs, SCADA, and ERP systems that literally speak different languages. When these systems cannot share information, we forfeit nearly half of the total value that IoT could actually deliver, as the connections between systems are where the real optimization happens. To fix this, you have to do the unglamorous work of creating a unified namespace or an asset model that translates those disparate naming formats into a common tongue. Without that translation layer, adding more sensors just creates more noise for your IT team to store and manage at a high cost. It is like trying to run a global corporation where every department uses a different currency and alphabet—the friction eventually brings the entire operation to a grinding halt regardless of how many sensors you install.
When organizations finally get the data foundation right, what are the most striking performance gains you’ve seen across sectors like manufacturing or energy?
The results aren’t just incremental; they are often transformative, as seen in the World Economic Forum’s Lighthouse sites where productivity gains frequently exceed 50%. In the realm of energy, Google DeepMind famously used thousands of sensor readings to cut data center cooling energy by 40% initially, settling into a 30% average saving once the system became autonomous. Predictive maintenance is another heavy hitter, consistently delivering 30–50% reductions in machine downtime and slashing maintenance costs by up to 40% in some research. These aren’t just theoretical projections; they are hard numbers reflecting a reality where integrated data allows a plant to operate at its absolute peak efficiency. Seeing a facility move from reactive “firefighting” to a state of calm, optimized flow is perhaps the most rewarding part of this digital evolution.
With 84% of IoT initiatives stuck in “pilot mode” and 95% of generative-AI pilots failing to impact profit, what is the fundamental mistake these companies are making?
The most common mistake is a total inversion of priorities, where companies fund a flashy dashboard or a complex model before fixing their underlying data quality. Cisco found that only 26% of decision-makers considered their IoT projects a complete success, often because they ignored the foundational integration needed across teams and systems. We are seeing a pattern where 80% of AI projects fail—roughly twice the rate of standard IT projects—simply because the technology is ready but the data is not. Gartner predicts that 60% of AI projects will be abandoned by 2026 if they lack AI-ready data, which proves that you cannot “math” your way out of a bad foundation. It is a heartbreaking sight to see millions of dollars poured into an algorithm that eventually chokes on un-contextualized, siloed information that it cannot even read.
What is the specific sequence of investment that separates the few successful companies from the many that fail to scale their AIoT projects?
Success follows a very strict hierarchy: context comes first, analytics second, and AI third. The 26% of companies that actually succeed spend their “durable budget” on the invisible work of mapping protocols and naming assets so the data is trustworthy from the start. If you buy the model before you fix the data, you end up with a bespoke solution that has to be rebuilt every time you add a new machine, which is why 60% to 84% of projects never leave the proof-of-concept phase. A proper AIoT platform should act as a unifying layer for connectivity and modeling rather than just being another silo to integrate later. By investing in the layer that compounds—the contextualized data—you ensure that every future use case can plug into the same reliable foundation without starting from scratch.
As we move toward “agentic AI” where autonomous systems make decisions at machine speed, how does the role of data readiness shift from a business value concern to a safety concern?
This is where the stakes become incredibly high, especially since 23% of organizations are already starting to scale agentic AI. If an autonomous agent makes a decision based on un-contextualized data, the error doesn’t just sit on a dashboard waiting for a human to notice it on Monday morning; it executes at machine speed. Gartner expects about 40% of these agentic projects to be cancelled by 2027, partly because the risk of a “wrong” autonomous action is too great without a perfect data foundation. When a model acts on bad data in a physical environment, the loss is immediate and potentially dangerous, moving the conversation from ROI to basic operational safety and control. We are reaching a point where having “good enough” data is no longer an option if you want to let the machines drive the factory.
What is your forecast for the future of industrial AIoT?
My forecast for the future of industrial AIoT is a massive consolidation where the “connected thing” is replaced by the “connected operation.” By 2030, with 39 billion devices in the field, the competitive advantage will belong solely to those who can bridge the gap between OT and IT, turning raw telemetry into autonomous actions in real time. We will see a sharp divide between companies that treat data as a byproduct and those that treat it as a refined fuel for their autonomous agents. Ultimately, the industry will move away from the hype of the “model” and back to the rigor of the “foundation,” where the real winners are the ones who mastered their data context years before they turned on their first AI agent.
