In a recent report, Gartner estimated that only 15 % of AI projects are successful, a strikingly low figure. In contrast, Citrine Informatics has found that 72 % of its customers generate value within the first year.
This paper draws on the experience of two experts who have participated in hundreds of AI projects within chemical companies, extracting lessons on what works and what does not.
The Experts
Johannes Benkhoff is an Advisory Partner at SPROUT, a Frankfurt-based innovation and sustainability consulting company. He has spent over 25 years in the technology and innovation domain within the chemical industry, serving in various roles, including scientist, innovation manager, and executive leader.
His career includes work at Ciba Specialty Chemicals, BASF, and a ten-year tenure at Clariant that ended in 2021.
At SPROUT Consulting, he now applies this experience by advising chemical companies undergoing digital transformations, providing external support for their change management efforts.
Kyle Killebrew is the Chief Operating Officer at Citrine Informatics, an AI software company that enables faster development and deployment of next-generation chemicals and materials.
During his time at Citrine, Kyle has worked with some of the world’s largest chemical producers, placing advanced AI tools directly in the hands of product scientists and helping transform product portfolios.
For this work, Citrine has been repeatedly recognized as a global leader in innovation, AI, and climate technology by organizations including the World Materials Forum, Fast Company, Time Magazine, and others.
Educate Yourself
Imagine you have been tasked with identifying the benefits of AI and determining what requirements your company might have for this new technology. What would be your first step, and why?
Johannes: It is essential to understand what AI, or material informatics in product development, is, what it is not, and how it differs from tools such as ChatGPT. This understanding must then be communicated throughout the organization. There is little to no “magic” involved, and it is important to bring expectations back down to reality.
My first step would be to educate myself, not to gain deep technical expertise or present myself as an authority, but to build a clear understanding and personal conviction regarding what AI could do for the organization. Only by being convinced myself can I effectively convince others.
By identifying and examining industry case studies from reliable internal and external sources, I can help inspire and persuade others. By clearly demonstrating business impact and thinking on a larger scale, leaders and experts become enthusiastic and are more likely to engage in the AI journey.
What common applications and notable outcomes have you observed in the chemicals and CPG space?
Kyle: Our customers cover the entire value chain, from producers of intermediates and commodity resins such as LYB and Braskem, to companies producing specialties and formulated products like Eastman, Syensqo, Synthomer, and Lanxess, and finally to end-market manufacturers such as B Braun, Panasonic, and Rolls Royce.
Many of these companies use AI to rapidly reformulate products. Faster reformulation allows them to win customer RFPs more quickly, respond more effectively to market demands, anticipate regulatory changes, or address gaps in raw material supply chains.
One current example is the industry-wide effort to remove PFAS, often referred to as forever chemicals, from formulations. One Citrine customer achieved this for an in-market adhesive in just 40 % of the time typically required to reformulate a product.
Customers also continue to discover new applications for AI that unlock significant value.
Some use AI to optimize processing, reducing energy consumption, minimizing waste, or increasing yields. One specialty chemicals customer achieved the highest yield they had seen in more than two decades, reaching that result in 75 % of the time compared to their internal benchmark. They estimate approximately $1 million savings from that single project.
Several multinational customers have used Citrine to adapt formulations to local ingredients, reducing supply chain costs and Scope 3 emissions, or to rationalize ingredients and lower inventory expenses.
A customer supplying paints and coatings is using the Citrine Platform in technical sales to generate multiple formulations that meet customer specifications during meetings. In a recent Handelsblatt article, the company stated that this approach helped increase revenue by 30 %.
Some customers are also developing entirely new products and expanding what is considered possible in significantly less time. However, because this receives less emphasis across the industry, it is typically not the primary use case.
Are there common misconceptions about AI that you encounter during initial discussions with potential customers?
Kyle: AI in chemicals and CPG often falls into what I refer to as “The Valley of Misperception.” Some people equate AI-driven chemical innovation with technologies such as ChatGPT, IBM Watson, or Google’s DeepMind. These systems rely on massive datasets, which generally do not exist in industrial chemicals and product development.
On the opposite side of this valley is the belief that small data environments, such as laboratories, cannot deliver value quickly. In reality, the lab is a uniquely well-suited environment for AI. Data in this context is expensive and therefore sparse when compared to the datasets used to train models like GPT.
There are long-established models and techniques that Citrine uses to allow customers to start training models with as few as 20 to 30, or perhaps up to 100, data points rather than hundreds of thousands.
The data is often messy, handwritten, unstructured, or physically stored in cabinets, and frequently scattered across multiple silos. Citrine combines several established approaches to rapidly make use of this information.
Although structured data may be limited, chemical developers possess a deep reservoir of knowledge in the minds of experienced product experts. Citrine AI can learn from this expertise in much the same way a new graduate hire would.
The AI does not need to rediscover the laws of physics, as in some cases it is already pre-trained to understand them. In more specialized situations, users can explicitly provide those principles by typing or verbalizing them to Citrine.
This point is worth emphasizing: Citrine AI is designed to work alongside product experts, just like any other member of the team.
This approach is taken for three reasons:
- It enables rapid iterations when product experts have new ideas, without the need to wait for intermediaries. Citrine AI typically learns a product’s chemistry in two or fewer experimental iterations.
- It allows AI usage to scale easily, since data science resources are no longer a limiting factor.
- It ensures that expert knowledge is captured during routine use of the software, codified within the platform so that less experienced employees can use it and learn from it over time.
This combination of diverse AI techniques and an intuitive user experience is Citrine’s secret sauce, enabling product developers at any experience level to quickly train and learn from the Citrine Platform using minimal data and with a minimal learning curve.
Create a Common Vision
So, we have educated ourselves. What comes next?
Johannes: In practice, many organizations never move beyond this initial phase, which represents a critical barrier we call the “getting started dilemma.”
AI is widely discussed, experts focus on science and technology, and there is excitement across the organization, yet no concrete steps are taken to move forward.
What is often lacking is a shared organizational vision.
It is essential to translate what AI can do in general terms into what it can specifically do for your organization, and to clearly articulate a compelling future state. Ultimately, the purpose of adopting AI is to strengthen competitive advantage and create sustainable, long-term business value.
Overcoming this hurdle requires bringing together business decision-makers and product experts to identify opportunity areas, map potential applications, and evaluate how AI can deliver value in both the short and long term.
From this process, an inspiring yet realistic vision should be developed. This vision generates enthusiasm among teams and individuals and helps secure resources, budgets, and organizational commitment. People will be motivated to participate.
Pay Attention to People and Culture
How do you translate an inspiring vision into the necessary internal momentum? What steps follow?
Johannes: It is critical to understand the organization’s current state, including not only processes and data, but also, and especially, the people involved.
At an early stage, it is important to identify individuals who are enthusiastic about adopting new technologies and are likely to lead the change as early adopters. It is equally important to recognize those who may resist or slow the initiative.
Understanding their concerns, potential threats, and fears, both within and outside the team, is essential. These issues must be addressed directly through active change management.
Peter Drucker’s statement that “culture eats strategy for breakfast” is particularly relevant here. A culture that encourages openness, listening, and addressing concerns can mobilize people, while the absence of such a culture can lead to disengagement.
I have learned this through experience.
When I first began working in this area, I mistakenly assumed that scientists would naturally be enthusiastic about new technologies and eager to learn. However, researchers are human, just like everyone else.
Let me share a “funny” story.
We launched an initiative to build new capabilities among scientists by introducing more advanced experimental planning tools.
These included tools such as DOE, or design of experiments, intended to replace what I call “Excel sheet planning.” The goal was to achieve a faster, less biased, and more scientific approach to experimentation, thereby accelerating lead generation and time to market through focused development cycles.
Our assumption was simple: “train them, and let them run”, give them freedom to experiment, and success would follow.
That assumption proved wrong.
Despite initial training, capability building, and ongoing expert coaching, within a few months, only a few scientists were consistently using the new tools and integrating them into their daily workflows. Most had reverted to the original, more familiar methods.
The conclusion was clear. Even scientists are human, and change can be intimidating. Ultimately, understanding and actively managing the change process is essential for success.
How do you actively manage the change?
Johannes: It is important to concentrate first on early adopters, the people who are highly motivated at the beginning, and then address those who may follow at a later stage.
The first step is to select initial cross-functional teams, diverse groups of employees made up of key roles and skills, to define the first roadmap for early projects within the organization.
These are people who are eager and willing to understand and address the relevant technical and business questions, challenges, and opportunities, and to translate them from an inspiring vision of “what is possible” to what is realistic and feasible within the organization, including cost implications and timelines.
Over time, others will join once initial success has been demonstrated. When success becomes visible, people are more likely to follow, especially when they recognize the personal benefits, such as opportunities to grow, succeed, and stand out.
Choose an AI Partner Well
If you were selecting an AI technology provider to work with, what three things would you look for?
Kyle: First, look for an AI platform that integrates well with your existing team and data and is designed to create value quickly. This means weeks, not months or years.
Second, while standard considerations such as information security and business longevity are assumed, it is critical to ensure that the software can scale quickly. This includes confirming that the provider uses a data model that can grow as AI adoption increases and that it is easy to onboard new business units into the system.
Third, select a provider with an experienced team that understands change management and is willing to partner with you to achieve business outcomes, rather than simply delivering software.
Select Initial Projects Strategically
How do you choose the right initiatives to start with?
Kyle: This is a very important question. Over the past decade, we have learned two key lessons: first, how you start is one of the strongest predictors of long-term success, and second, there is a wrong way to start.
Citrine has built a disproportionately large and experienced team to ensure these points are properly addressed. Even before implementation begins, we work closely with customers to ensure their AI goals are specific, measurable, and achievable.
We then create a plan that clearly defines how value-based outcomes will be achieved and on what timeline. This plan prioritizes AI applications based on two factors: the potential business value and the ease of application. Ease of application depends on the availability of experts, data, and laboratory time.
Making experts, data, and lab time available, as well as aligning on goals and execution plans, all require leadership commitment. Leadership engagement is the strongest indicator of long-term success.
We launch the highest-priority applications first while, in parallel, preparing lower-priority initiatives for later execution.
Start with Minimal Data
Many advisors recommend focusing first on data strategy and preparation before applying AI. Given your experience with sparse data, is that what you would recommend?
Kyle: Our customers who achieve the highest returns, typically three to five times in the first year, work on data and AI in parallel. Prepared data alone delivers limited value. The real impact comes from training AI models and running analytics on that data.
High-return customers begin with whatever minimal data they already have, even if it is sufficient for only a single project. This allows them to generate early value, identify which data is worth digitizing, and begin accumulating returns immediately while continuing to prepare additional data.
Much historical data may be outdated and not worth digitizing.
By selecting an initial application with high potential return and completing it quickly, organizations generate interest and momentum, which in turn supports broader data digitization and additional AI initiatives.
Track Progress and Keep Everyone Aligned
If success in the first projects is critical to the overall initiative, how do you ensure those projects succeed and stay on track?
Johannes: The first one or two projects must succeed. Failure is not an option, as it can significantly slow progress or even jeopardize the entire initiative. What is required from senior and top leadership is a mindset of doing whatever is necessary to ensure success.
The organizational and project framework plays a central role. Three aspects are particularly important:
- Deeply engaged people and teams who begin the journey with a realistic execution roadmap.
- A business leader who is closely involved, actively supporting the team and helping remove obstacles.
- Top executives, ideally including the CEO, who visibly support the initiative. Transformational efforts are most effective when led from the top, with continuous inspiration provided through leadership communication and progress updates, aligning words with actions.
Close attention and coaching of the initiative is needed:
- Capability building to support project teams to learn, practice, and play their new roles.
- Agile working practices should be considered, as classical project management methods tend to fail for such exploratory initiatives.
- Continuously aligning the organization – horizontally across departments, and vertically across hierarchies.
- A cross-functional steering framework to keep the project focused on value and technically realistic.
When internal resources are limited, independent external support and coaching have proven to be highly effective.
Shout About Success and Avoid “Not Invented Here” Syndrome
Once you achieve success, what comes next, and what obstacles do you typically encounter?
Johannes: Initial success is encouraging, but it does not automatically lead to continued success. Human behavior plays a role here. People often believe their own area is more complex or different, and that what worked elsewhere will not work for them.
This mindset is commonly known as the “not invented here” syndrome.
People tend to prefer trying things themselves rather than adopting proven solutions from others. Unfortunately, human psychology is often more responsive to failure or pain than to positive examples.
I learned this lesson the hard way.
We planned an AI-based initiative to accelerate product development and significantly improve the performance of next-generation materials in an important business segment.
- We identified a highly ranked business leader who was motivated to explore and invest, providing strong sponsorship.
- We selected a top-tier external partner to handle the computational science aspects. By combining computational methods with historical data, we successfully modeled a very complex formulation.
- We developed high-throughput experimental workflows, enabling rapid experimental cycles and the generation of large amounts of new data. Within months, we had built a true in silico screening tool.
- We gained deep and unprecedented insight into material behavior and achieved a significant performance improvement.
It was a clear success, but it did not scale.
The reason was that we focused too heavily on data, technology, and applications where success was highly likely. As a result, other product experts found convincing reasons why the approach would not work in their own environments and chose not to try it.
We failed to create a compelling story that attracted additional leaders and motivated them to support the initiative.
Counterintuitively, it is often more important to win the support of strong business leaders who send clear and inspiring signals than to have the perfect project supported by perfect data.
Scaling should therefore follow the same principles as the initial approach:
- Creating an inspiring and convincing picture of how value and success for the business can be created.
- Ensuring you have engaged leaders supporting and leading teams.
- Continuous communication of progress and success from top leadership.
Strong results spread enthusiasm. Public recognition of success is especially powerful, as it encourages people to volunteer and participate.
Scale Up
How does Citrine help customers scale and move to the next level?
Kyle: We frequently work with customers to broadcast the success of initial projects across the organization, generating interest from additional product experts and organically driving further scaling and value creation.
From a technology perspective, the platform features that enable scaling become increasingly important. Several customers attempted to build their own solutions and struggled when it came time to scale. Their data scientists created systems that were difficult for product experts to use and challenging to deploy and manage at scale.
The Citrine Platform was designed as enterprise software. It uses a graph-based data model to integrate diverse datasets, employs modern usability techniques that require no coding, and incorporates over eleven years of materials informatics experience.
The platform automatically generates a strong initial model, allows product experts to add their knowledge easily, removes dependency on scarce data science resources, and meets enterprise requirements for security, authorization, and regular updates.
As a result, product experts focus on their core expertise rather than software deployment or maintenance. For them, the platform is intuitive and enjoyable to use.
How does Citrine remain involved once AI workflows are embedded into daily operations?
Kyle: The team that supported the initial rollout, including data engineers, data scientists, and change management advisors, remains available to assist with issues or provide advice on new projects.
Customers are generally able to launch and execute new initiatives independently. However, because our team combines scientific expertise with extensive experience in data integration, customers may request deeper involvement again as new teams or products are introduced.
Key Takeaways from Kyle and Johannes
- To initiate progress, educate yourself and develop a compelling vision of business value and the future to engage leaders, experts, and teams.
- Identify and empower early adopters who are ready to lead the way, and support them through training, coaching, and resources.
- Highlight successes and clearly show how individual contributions lead to personal and organizational growth. That will foster imitation.
- AI for materials, chemicals, and CPG innovation works effectively with small datasets to reduce experimental effort, has demonstrated success, and will become widespread across the industry.
- Choose an AI provider that can leverage your existing data, expertise, and capabilities without requiring major data projects or large new teams.
- Ensure the AI solution can scale across the organization, with a flexible data model, reusable assets, and user-friendly software.
This information has been sourced, reviewed, and adapted from materials provided by Citrine Informatics.
For more information on this source, please visit Citrine Informatics.