Last month, I had the privilege of speaking at the Keynotion Smart Manufacturing World Summit in Stuttgart on a topic that’s close to my heart:
In previous blog posts, I’ve covered the challenges and opportunities in the manufacturing industry and why the process industry should care about AI in general. A common thread in many sessions at the Smart Manufacturing Summit was the growing interest in closed-loop AI. Closed-loop AI refers to AI systems that not only analyse data and make predictions but also automatically feed those insights back into industrial control systems to optimise operations in real time. The concept was popularised during the rise of Industry 4.0, emphasising autonomous decision-making, continuous learning, and self-optimisation as key drivers of smarter, more efficient manufacturing.
But while the potential was widely acknowledged at the event, the implementation gap became crystal clear.
During my session, I asked the audience:
“How many of you have already implemented closed-loop AI - or have started planning to?”
To my surprise, not a single hand went up. This prompted a discussion around the barriers to adoption. I ran a poll to get a better understanding of these challenges. The top hurdle by far?
Access to and trust in industrial data.
This finding is echoed by several industry research firms.
These reports underscore the critical importance of addressing data quality, integration, and governance to implement AI successfully in industrial environments.
So, if a lack of trust in data is the bottleneck, how can we fix that?
If you’re running a legacy plant without a complete sensor setup, the key is to start with valuable use cases - don’t install sensors just because “AI needs data.”
For instance, if your goal is to optimise product quality, begin by identifying equipment in the production line that your quality engineers believe may influence the outcome. Then, focus on installing sensors that measure both critical process parameters and the quality of the raw materials and final product. This ensures you’re collecting the right data to drive meaningful improvements.
Even if your machines are equipped with sensors, it’s common to see data being generated but not stored. That’s a missed opportunity.
For AI models to be trained effectively, you typically need at least 6 months of historical data - ideally 9 to 12 months, depending on your use case. That said, don’t wait! You can still start analysing what data you do have to understand what happened and why.
Many plants have years of historical data sitting in their historians - but often at low resolution. That’s because storage was expensive back then. Not anymore.
First, increase your data resolution going forward. More granular data means better insight, and better models.
Second, to move toward real-time optimisation and closed-loop AI, you need streaming data. Most modern DCS and SCADA systems support OPC-UA, a protocol that makes data sharing and streaming efficient and secure.
If your system is too old, consider installing an OPC-UA bridge to enable real-time streaming without needing to overhaul everything.
A common reaction we hear is: “AI will never control our plant - it’s too risky.”
And that’s a fair concern. But the key is to build trust in AI first, before moving to full automation.
One powerful way to do this is to stream AI model predictions back to the control room, where operators can view and use the insights - but still remain in control. Instead of relying on gut feeling (which is still very common), operators can make decisions based on data-driven recommendations.
This approach helps create trust in the AI models, and it's a major first step toward closed-loop AI - a goal that no longer feels like a distant dream.
The true power of closed-loop AI lies in its ability to drive continuous process optimisation - unlocking improvements that directly impact the bottom line.
Across industries, we’re seeing manufacturers use AI to:
🌎 Real-world Use Case - Optimising product quality and energy consumption
One of our Food & Beverage customers in the dairy sector saw a need to optimise protein powder yield without compromising on quality - a key requirement for selling to premium buyers in the sports nutrition market. They knew what great looked like, however, there are many process steps and settings that will affect the end product. With sensor inputs from across the production line, their quality engineers created an AI model to forecast how protein levels would evolve up to one hour into the future. Leveraging the Intelecy no-code AI platform, the engineers were able to do this without writing a single line of code.
The predictions from the AI models were streamed in real-time directly to the control room, allowing operators to make proactive adjustments and stabilise the process. This resulted in:
After building internal trust in the system and proving that the AI consistently delivers accurate predictions, the manufacturer is now preparing to fully close the loop—automating setpoint adjustments directly in the control system to optimise production in real-time.
🌍 Real-world Use Case - reducing chemical use and increasing sustainability
Another Intelecy customer, **VEAS - Norway’s largest wastewater treatment plant**, processing 3,000 litres of sewage per second—faced a common challenge: the impact of chemical dosing could only be measured 45 minutes after application. While experienced operators had long managed the process manually, VEAS saw an opportunity to enhance decision-making with AI.
Using Intelecy’s no-code platform, their engineers built a forecasting model that combined key variables like chemical dosing, water quality, flow, and temperature to predict future turbidity - enabling more precise and proactive chemical optimisation.
This enabled the process engineers to:
With an annual chemical budget of NOK 60 million, even small improvements in efficiency can translate to significant cost savings - all while maintaining regulatory compliance and water quality standards.
Next step? Fully implementing closed-loop AI, allowing the system to automatically adjust chemical dosing based on real-time forecasts - maximising efficiency with minimal manual intervention.
These examples prove that closed-loop AI isn’t just a futuristic idea - it’s a practical, high-impact tool that’s already delivering measurable results in the real world.
At Intelecy, we’re proud to support manufacturers at every stage of their digital transformation journey. Here’s how we can help:
If you're curious to explore how to take the first step, or the next one - toward smarter, greener, and more profitable manufacturing, we’d love to talk. Let’s talk!