Energy costs are rising and for industrial manufacturers, that pressure is only going one direction. This case study explains how we helped one of the Europe’s largest brick producers cut gas consumption by more than 10% by combining AI with hands-on engineering knowledge.
We’ve built a scalable, data-driven AI solution that engineers fully trusted and adopted so production KPIs started improving fast.
Key results at a glance:
- +10% in energy cost savings thanks to a significant reduction in gas consumption
- New insights into production from 700+ process variables
Context and Challenges
Our client is one of the Europe’s largest brick manufacturers. Firing bricks require a substantial amount of energy, and gas is one of their biggest cost lines. With energy prices going up and sustainability targets getting harder to ignore, reducing consumption was not just a nice-to-have, it was crucial for the business.
The company’s goal was clear: minimize energy use and adopt a more sustainable, data-driven approach to production management.
The engineers had their suspicions. Certain parameters, and certain moments in the process, felt like potential sources of inefficiency. However, suspicion alone is not enough to change how a plant runs. Without data to back it up, every adjustment was essentially a guess, and the results were hard to measure, difficult to repeat, and easy to argue against.
Achieving this goal came with three technical hurdles:
- Complex machine data: More than 700 interconnected process variables made it difficult to identify what was actually driving energy consumption.
- High volume of streaming data: Real-time data from plant operations needed to be captured, processed, and analyzed at scale.
- Identification of actionable insights: Beyond data collection, the team needed clear, understandable outputs that engineers could act on directly.
Our Approach
We designed and implemented a comprehensive analytics solution tailored to the client’s operational environment. Our approach combined infrastructure, collaboration, and usability to deliver results that were both technically robust and practically adopted by the engineering teams.
Starting with workshops on the plant floor
Before building anything, we held in-depth workshops with the plant floor engineers, to fully understand the processes and operations from the people who actually operate the machines.
Turning conversations into testable hypotheses
From those conversations and workshops, we jointly defined a clear set of hypotheses about what was driving energy consumption. Before writing a single line of code, we tested whether those hypotheses were realistic, both in terms of potential impact and how difficult they would be to act on. We focused on delivering business value and on whether the solution could be scaled to other plants.
Building a scalable data and AI platform
The data & AI platform ran on Azure and was built to handle the full data of a large manufacturing operation: billions of records, real-time streams, live production conditions. We built it to scale from the start, so adding more plants later would not require starting over.
Keeping models simple enough to trust
We kept the models simple by design. Engineers needed to recognize the output as something that made sense, not simply trust a number. We validated every result against basic process physics, and where the model and reality disagreed, we adjusted the model.
Deploying directly onto the machines
The models were not just sitting in a dashboard somewhere. Once we gained trust into the models, they were deployed directly into the process itself.
Validating the results
Finally, we designed a structured process to track outcomes and check them against the original hypotheses together with the process engineers.
Scaling up
With the results proven and the engineers on board, the focus has shifted to what comes next. The solution is being rolled out to additional plants, and the models themselves are evolving. Now that there is a solid baseline and a team that understands and trusts the outputs, it makes sense to go further: more complex models, more variables, more opportunity to reduce consumption even further.
Key Benefits
Our collaboration delivered measurable impact across sustainability, operational efficiency, and long-term scalability.
Reduced Gas Consumption
The most significant outcome of the project was a reduction of more than 10% in natural gas consumption. By identifying the main drivers of energy use across hundreds of process variables, the team was able to take targeted action to lower consumption—reducing both environmental impact and energy costs.
New insights into production
The client now has direct access to production KPIs they previously lacked visibility into, including quality, throughput, and moisture levels. These insights support more informed, data-driven decisions at every stage of the production process.
Full Integration into Azure Infrastructure
The solution integrates fully into the client’s existing and now extended Azure infrastructure, ensuring operational continuity and minimal disruption. This also means the platform benefits from the security, compliance, and scalability features already in place within their environment.
A Scalable Foundation for Future Growth
The architecture was designed to scale. The same approach—combining low-code data pipelines, streaming analytics, and user-centered dashboards—can be deployed across additional plants without starting from scratch. This gives the client a reusable blueprint for driving energy efficiency across their entire operations.
Team Involved in This Project
Six months and a small multidisciplinary team: data scientists with an engineering background, data engineers, a cloud architect, and a project manager. The focus was on delivering real value, gaining the trust of the plant engineers, and building something scalable enough to grow with the client.
Technologies Used
The platform was built on Microsoft Azure, providing the scalability needed to handle billions of data records across live plant operations. On top of that, we combined open source tools with SAS Viya.
The tech stack was deliberately chosen to be modern but approachable. Process engineers needed to be able to work with the data and the models day to day.
