News Production & Lab
Low-VOC and AI: “Experimental capacity is the rate limiting factor”
The push for VOC reduction continues to challenge the coatings industry. In this interview, Ece Karaarslan. Materials Informatics Engineer at Citrine Informatics, explains how AI is reshaping formulation processes, allowing for faster raw material screening and innovation. She also shares insights into data quality requirements. Interview by Vanessa Bauersachs
VOC reduction has been on the industry’s agenda for years. How much has actually changed in the way formulators approach this challenge, and where are the remaining bottlenecks?
Ece Karaarslan: There are two approaches that are being used by our customers. The first one is switching to water-based formulations, and the second one is reducing the use of high-VOC additives. I don’t think this approach has changed. But what has changed is the ability to explore a wider search area with many more raw material options in a shorter space of time using AI. In terms of bottlenecks, now that the data analysis can be done in less than an hour, experimental capacity is the rate limiting factor.
AI is increasingly discussed as a formulation tool. What can AI contribute to low-VOC formulation?
Karaarslan: There are two opportunities here. The first is a simple chemical substitution. AI can be used to understand the aspects of a molecule that are positively affecting the final properties of a formulation, creating a fingerprint as it were, and then identifying alternative raw materials that perform the same function. It can predict final properties, even when a new raw material has not yet been tested, and therefore screen options and reduce the number of experiments needed to remove a high-VOC additive.
The second is switching to water-based formulations, which requires a match of surfactants and binders. Our platform has been used by Perstorp, a specialty chemical manufacturer, to determine the appropriate surfactant blends for alkyd emulsification. They won the American Coatings Award for their work on that this year. They solved 17 alkyds in the space of eight months, something that would have historically taken them five years. It was great to see their hard work translate to higher lab efficiency, faster customer response, and an accelerated commercial pipeline.
What data quality and quantity are needed before AI tools can deliver meaningful results?
Karaarslan: We advise our customers to start with about 25-30 data points. Each data point is an experiment where you have recorded the exact formulation, processing conditions, and final measured properties. The data points should be accurately measured and recorded, and there should be diversity of formulations and processing. Thirty experiments all using the same surfactant will not enable AI to learn about the effects of different surfactants. However, AI can cope with sparse data, so if you have experiments where a property was only measured in half the batch, you can still use that data.
AI for materials and chemicals is different from LLMs. Materials and chemicals data sets tend to be small. Eighty-five percent of our customers come to us with fewer than 100 data points. The aim of the game is not to use hundreds of thousands of data points to create a perfectly accurate model, but rather to have enough data to create a model that can improve your next decision on what to take to the lab for testing. Quality of the data matters more than volume.
You will be presenting at the EC Conference Sustainable Coatings in November. What can attendees expect from your presentation?
Karaarslan: I’ll be talking about a couple of different case studies from our customers’ work. I’ll focus not just on low VOCs but also on the efficient removal of PFAS from formulations. I’ll also discuss what we have learned as a company over the last decade and provide hints and tips for getting started with AI. As 70,000 experiments are now predicted each month on our platform, we are privileged to see different approaches to introducing and scaling AI across businesses and have a view of best practices which we’d like to share.
Event tip:
The Sustainable Coatings Conference, which takes place on 3 – 4 November 2026 in Amsterdam, Netherlands, will provide practical insights into low‑carbon technologies, circular economy approaches, bio‑based and water‑based systems, and robust assessment methods such as LCA and mass balance. Learn how the industry is responding to regulatory pressure, customer expectations, and material constraints – and how sustainability can become a measurable business advantage rather than a compliance burden.
