alcemy, the Berlin-based AI company focused on the cement and concrete industry, recently announced a major update to its offering for customers and unveiled plans to further support producers on the way to autonomous cement and concrete production.
The announcement covered three major initiatives:
- alcemy Foundation Partnership - a new enterprise AI transformation offering.
- alcemy for Cement - the roadmap toward autonomous mill operations.
- alcemy for Concrete - the roadmap toward autonomous recipe control.
In order to find out more about these exciting developments, World Cement sat down with alcemy’s CEO, Leopold Spenner, and General Manager, Oliver Kanders.
The alcemy Foundation Partnership, your new Enterprise AI transformation program, is a logical, but quite a significant, expansion of your offering. Why did you feel that now was the right time to take that next step?
Leopold Spenner: There are both external and internal factors. The external factor is the emergence of powerful models from Anthropic, Google, and Open AI and the entirely new magnitude of possibilities they bring. These large language models are trained specifically on writing software code and interacting with data, which makes them 10x or 100x more useful in practice than earlier iterations.
The internal factor is that our customers have been pushing and asking for this. They see some AI successes here and there in individual plants, but they want to take it to scale. They know that without the underlying databases, every new plant and every new AI use case will turn into a new six-month implementation project. The effort-benefit equation simply will not be there.
In your announcement, you mention that many cement producers struggle to make the jump from successful pilot projects to company-wide adoption because they lack the organisational and technical foundations to scale AI across multiple plants. What exactly are those missing “foundations”, and how can alcemy help provide them?
Leopold Spenner: It is about establishing consistent data across different IT systems, plants, countries, and continents.
To take a simple example, the VDZ has given the industry a powerful data standard for assessing kiln and mill efficiency globally. Yet, at many plants, the underlying data is fragmented and poorly contextualised. That’s why audits can still require dozens of back-and-forth interactions just to clarify basic details. The missing foundation is harmonising those nitty-gritty data points and their context across all plants and IT systems so standards like the VDZ’s can be used to their full potential, which is currently not a reality.
The tagline for your new Enterprise AI offering is “Your data. Your people. Your platform.” Focusing on the “people” section, tell us how you’re working to ensure that plant personnel are brought along as willing and enthusiastic partners in the scaling up of AI.
Leopold Spenner: In this first phase we will only work with a maximum of one or two partners that are all in. That commitment requires the producer to put people in place who are fully responsible for the AI transformation and adoption - roles many producers do not even have yet.
We’re primarily seeking quality experts and process engineers who are well-respected today to take on this task. We envision them becoming the agent of change responsible for transforming the whole organisation.
The main tool for doing this is the agentic AI setup, where anyone at a plant can use natural language to quickly build things. That building process becomes extremely powerful and effective if you have the entire data backbone to build upon - the whole ERPsystem, all process data, and all lab data across the cement and corresponding concrete plants.
There are enough tinkerers and curious people in our industry who, once they have access to all this data in a homogenised way, can understand how powerful it is for solving their daily problems. This is exactly what we are trying to achieve with this programme.
One of the other parts of your announcement was that you’ll be taking further steps towards fully autonomous mill operations. One of the ways you plan to achieve this is via integrating additional process data and optimisation variables. What kinds of additional things are we talking about? What’s getting added to the picture?
Oliver Kanders: We decided to look more holistically at our full capabilities, but I can provide two specific examples:
Mill feeder data: Right now, we are relying on the producer to manually report the actual impact of our optimisation on exact throughput. We want to automatically and more precisely calculate that data.
Grinding aids and strength enhancers: These are becoming much more important. In quite a few regions, you might not typically be allowed to decrease clinker content further, but you can do so if you manage to apply powerful activators.
These do not currently appear in standard laboratory analysis simply because of the very small amounts used. Getting that data documented and into our models makes total sense because we want to include it as a steering variable.
The announcement also touched on your plans to deepen relationships with leading automation providers and mentioned the example of your strategic partnership with ABB. How important are these kinds of partnerships to achieving autonomous milling operations?
Oliver Kanders: They are super important. Philosophically speaking, we are doubling down on splitting responsibilities.
We need to look at the “whole” because producers are asking us to resolve a major puzzle piece for them - the autonomous mill - but we are fully betting on doing that as part of a collaboration.
So much has already been built and implemented by companies like ABB and others, and it is obvious that you need these systems to run a mill autonomously. You need both the set points and the automated capability to translate these into the right technical orchestration.
Going back to that point about moving towards “fully autonomous mill operations”, how close are we to achieving that goal?
Leopold Spenner: We are perhaps three to five years away from that. There are probably two parts of the problem to be solved:
Recipe optimisation: Figuring out the optimal grinding fineness, the optimal cement recipe, the optimal chemical admixture, and the larger picture of the recipe.
alcemy is heavily focused on this first problem, where we are already able to do “dual steering” - simultaneously changing either the recipe, such as the clinker versus SCM ratio, or the grinding fineness, depending on what is optimal for the target function.
Equipment execution: Equipment execution: How to run the target recipe as efficiently as possible on the actual plant. This includes finding the optimal filling degree of the mill, the right ratio between separator speed and separator air, and the appropriate amount of material to return to the mill.
The ABB expert system is already able to execute a fineness set point and a recipe set point, and this has only been strengthened through our collaboration. On our side, we want to integrate grinding aids and performance enhancers as new variables. As we integrate more process data, we will see more parts of the recipe - such as sulfate control and different SCMs like slag versus limestone - and new opportunities will pop up.
alcemy has already automated the changing of fineness and recipe set points, but an advanced operator in a plant today plays with even more variables. We are trying to close this gap, and it is probably a matter of three to five years to fully get there.
Turning to the third initiative from your announcement, which covers your plans for autonomous concrete recipe control. What led you to decide that recipe optimisation was the right part of the process to target?
Leopold Spenner: We looked at where we came from: we started with quality transparency all the way to the construction site. This was our beginning, and it is what our clients use and love because they can better defend claims and get transparency on unwanted behaviour on-site, such as water additions or driver errors. It also helps the batchman get on target more easily.
While this was excellent for establishing the reliability of the production process, the next question was how to reach good financial returns. That ultimately comes down to the recipe.
The focus is now shifting to benchmarking all recipes across all plants, using the specific sands, aggregates, and cements available at each site. We want to identify where we are operating close to the limits, where we still have margin, and how we can move closer to the optimum.
We have laid the foundation, and now we want to move the ROI needle.
To achieve the goals of autonomous recipe operation, you’re developing an AI recipe optimiser. Tell us a bit about how that will work in practice.
Leopold Spenner: This will operate in a day-to-day, continuous production setting. With the data we acquire, we can understand and classify the quality and water demand of the sand and aggregates, and evaluate how difficult it is to turn them into good quality concrete.
We can carry out real-time analytics and predictions on workability, viscosity, and compressive strength development because we have a huge number of data sources connected to the system.
For example, in the batching plant, we know all the dosages, temperatures, and moisture levels of the ingredients. We track the power curve, which tells us the actual stiffness of the mixture.
From this, we can infer the water content in the mixture and the expected compressive strength within a real, ongoing production setting rather than a lab environment.
With all this information, we can clearly point producers to recipes or mix designs that are sufficiently robust, yet still offer room to reduce water and cement content based on our current assessment of aggregate water demand. Looking ahead: do you expect to see the full automation of processes across the entire cement and concrete production chain becoming standard practice in the next few years?
Leopold Spenner: I really think so. In five to eight years, we will see many autonomous mills, possibly kilns, production planning, and predictive maintenance systems.
Producers will split into two categories. There will be those who see many plants successfully using these systems, with employees leveraging them to achieve a better overall productivity level, efficiency level, and clinker factor.
Then there will be those who struggle to do so because of their databases and the lack of a proper data backbone. This data foundation, and the people skills built on top of it, will be the differentiating factor. Some will definitely get there, while others will struggle to make this transition or end up in serious dependencies.
We will really see the wheat being separated from the chaff over the coming years.
Click here for free registration to World Cement
Check out the World Cement Podcast