Sky-high expectations, inconsistent data and siloed pilots can prevent AI pilots from successful real-world deployment. The solution is in industrial intelligence that closes the loop between prediction and plant floor, writes Stephen Reynolds, Industry Principal, Chemicals at AVEVA
Every Formula 1 car generates hundreds of gigabytes of telemetry data during a race: tire degradation, fuel burn, brake temperature, weather changes, even competitor behaviour.
This data is streamed to engineers at the track and at remote labs in real time – so they can analyse performance, optimise race strategy and gain a competitive edge in a sport where every millisecond counts. It’s a reminder that success isn’t about the size of your dataset or how sophisticated your model is, but whether insights are turned into timely, operational decisions.
Subscribe to PII and receive:
Our long-established bi-monthly digital magazine *Process Industry Informer*, featuring exclusive articles, news and updates across the Process Industries, plus commentary from our three industry experts:
Sean Moran
Gavin Smith
Dave Green
PLUS:
- Process Industry Update — our weekly e-newsletter
- PII Podcast
- Special messages from our advertising partners
Chemical plants can outperform the market by channeling this approach as they welcome artificial intelligence (AI) into their hypothetical racecars. Plants already generate billions of data points from sensors, labs and ERP systems. Teams are running AI pilots, testing models, and launching proof-of-concepts, with McKinsey reporting that 78% of companies use AI in at least one business function.
Most AI pilots stall
Yet the vast majority of AI pilots stall before delivering value. A staggering 88% of AI pilots fail to reach production, according to a recent survey. When it comes to generative AI – the hot new kid on the AI block – 95% of initiatives to drive rapid revenue are falling flat, MIT reported in August.
This “AI purgatory” isn’t drag from a lack of imagination. It’s a lack of strategy and action. In other words, the inability to translate insights into real-world action, just as a perfect F1 strategy is useless without tire changes.
AI isn’t plug and play
For starters, teams may be struck by shiny object syndrome, treating AI like a plug-and-play solution. When ambition outpaces infrastructure, models are often applied to tasks they weren’t designed for, and they are fed inconsistent or delayed lab and sensor data. Pilots may also be isolated from operations, so the resulting insights can’t be applied to real-world processes.
“A staggering 88% of AI pilots fail to reach production, according to a recent survey.”
Even when AI predicts fouling in a reactor or inactive catalysts, the value disappears if operators are unable to act because of misaligned workflows. Impatience for instant results and the lack of continuous feedback from new feedstocks can compound the problem. Fragmented technology equals fragmented results.
Getting to pole position, to stay with our F1 metaphor, requires cross-functional collaboration and a connected ecosystem that unifies industrial chemical systems. That means linking together MES, LIMS, ERP, historian and process control systems within a single platform.
This creates a single source of truth that breaks down data silos, so intelligent insights can be fed into existing control loops and delivered where they are needed, from shop floor or the top floor.
“Only with these three elements – curated data, process-aware intelligent models and inspired humans in the loop – does AI move beyond theoretical concepts to real operational and R&D outcomes to deliver lower downtime, higher yields and shorter innovation cycles.”
AI augments, humans elevate
Curating and centralising industrial data is the start. Alongside, models must be designed so they respect upstream and downstream dependencies. Most important, however, is the human element: teams must be able to trust the intelligence they receive, and be empowered to act on them. In this respect, avoiding AI purgatory should be viewed as more of an organisational culture shift than a technological upgrade.
Only with these three elements – curated data, process-aware intelligent models and inspired humans in the loop – does AI move beyond theoretical concepts to real operational and R&D outcomes to deliver lower downtime, higher yields and shorter innovation cycles.
That’s how SCG Chemicals got to 99% plant reliability, with a ninefold return on investment in just six months. Keeping one of Asia’s largest chemical supply chains humming requires dependability at every level. In response, SCG built out a digital reliability platform that embeds AI across its lifecycle.
By integrating predictive analytics, centralised data and digital twin environments in one place, the platform enables teams to make process decisions on the fly, akin to arming F1 racing teams with real-time intelligence.
With a range of dashboards from the business unit level down to individual equipment, SCG’s teams can access actionable information and correlate it with real-time data within 10 seconds. Identifying catches to avoid asset failures has closed the reliability gap, and maintenance costs are down 40%.
Scaling up across the chemicals sector, we see AI use cases extending to improving asset uptime with predictive analytics, hybrid modelling to accelerate product innovation, and even to ingredients discovery, for example to create environmentally sustainable materials.
“When chemicals companies adopt this step-by-step approach, they move beyond reactive troubleshooting.”
Preventing AI pilot purgatory
But success in each area goes beyond treating AI pilots as technology experiments – a guaranteed pathway to join the 95% failure statistic. As digital and analytics tools are being adopted, companies need end-to-end approaches to turn analytics into operational improvements, as Deloitte pointed out in its recent chemical sector outlook.
Overcoming AI purgatory requires a shift in perspective, not least by way of cultural change. The first step is defining the KPI you want to change and quantifying its workflow impact. Pilots that try to do everything achieve nothing.
Then, build out the data-first connected ecosystem, integrating historians, MES, LIMS and vendor programs. Success rests on quality data, as Arthur D Little notes, so schemas must be standardised, metadata annotated and lab protocols determined.
The next step is selecting the right AI, and making it observable. For example, pattern recognition to forecast equipment failures, LLMs for compliance document search, or hybrid modeling for innovative applications, such as formulation.
Then, productize and scale, one use case at a time. The final step is cross-functional evaluation and governance – McKinsey recommends tasking senior leaders with oversight – to reduce model drift and adoption risk.
When chemicals companies adopt this step-by-step approach, they move beyond reactive troubleshooting. Operators can anticipate fouling, adjust reaction conditions and prevent downtime. R&D teams can accelerate formulations while ensuring consistent scale-up. The industry can finally extract real value from its AI investments, just as F1 teams convert telemetry into split-second, race-winning decisions.
