^{Tectonic Shifts}
R&D
& AI

Kathleen Timperman and John Walsh discuss
how AI is reshaping beverage R&D

Words: Kathleen Timperman & John Walsh
Images: Ricardo Rey

AI is reshaping beverage R&D — but where will it add most value, what are the risks and could it ever replace human expertise? 1750 Editor Thom Walton sits down with Kathleen Timperman, Chief Science & Innovation Officer and John Walsh, Chief Information Officer to find out…

Where can AI have the biggest impact on beverage R&D?

Kathleen Timperman: Being the geeky R&D person, machine learning — as opposed to generative AI — is where I see big potential. The ability to link our data and use it for questions we weren’t asking the first time we collected it will enable us to make stronger and better decisions. Generative AI still matters too. Tools like Copilot help us get research on ingredient compliance, stability or interactions much faster.

John Walsh: I often describe the time lag from data to insight to action. AI can really help accelerate R&D operations — trend sensing, ingredient screening, formula optimization. It’s not about AI inventing drinks, it’s about helping the R&D team explore thousands of possible combinations and quickly get to what we should be going after.

What are the risks if companies adopt AI too quickly or too casually?

KT: In a science-based function, credible data sources are essential. AI should be a starting point, then you validate. If you rely too heavily on AI, your knowledge doesn’t grow as quickly. You lose the ability to challenge assumptions or identify flawed conclusions. You shift towards surface-level insights rather than true scientific understanding.

JW: There’s a risk of AI becoming a black box. We have to rigorously test outputs. There’s also risk of bias — depending on the datasets, the time periods, the markets. And IP protection is critical.

We need the right security posture around any AI investment. For me, AI is decision support, not decision replacement. You always keep a human in the loop.

KT: Exactly. Erosion of critical thinking, reduced ability to challenge assumptions, and weakened capability to solve novel problems are real risks. We can’t allow that to become how the R&D team operates.

Where will AI add the most value to beverage R&D over the next 5 years?

Accelerating data‑to‑insight workflows
Unlocking value from existing datasets
Enhancing consumer and sensory understanding
Improving supply chain forecasting and operational efficiency
Driving creativity by surfacing new concepts and global trends

Machine learning - as opposed to generative AI - is where I see big potential.”

Where are you already seeing credible, real-world uses of AI in R&D?

KT: We’re starting to see third-party platforms using AI to generate faster insights. We’ve just purchased a subscription to a scientific research platform. This looks at journal articles, patents and news to generate summaries. It takes your library research to the next level. We’re also piloting an AI-backed consumer sensory tool that pairs descriptive trained sensory panel data with consumer data. In theory, this will help identify drivers of preference more quickly and more cost-effectively.

JW: Building on what Kathleen said, consumer trend analytics is another area — keyword analytics on platforms like Instagram that help us to see what ingredients are trending in terms of, say, health benefits. Digital twins are being used to model supply chains and lab equipment. And supply chain forecasting is ripe for AI and machine learning – not necessarily to forecast innovation, but to forecast more predictable products, which give the organization more time to focus on innovation.

Is AI genuinely transformational for R&D, or is it overhyped?

JW: The idea that AI will replace scientists — that’s hype. The reality is that AI will change and improve the R&D operating model. It will reduce iteration cycles, improve how we reuse knowledge, and make experimentation more targeted. But taste, brand fit, manufacturing feasibility and consumer emotion will still require human expertise.

KT: AI is not a replacement for real science. It’s an enhancement; an enabler. There’s a lot of skepticism around AI, but I’m like looking at it as an incredible unlock, helping us be faster and more informed. But you can’t use AI to understand how a product tastes. As a developer, you have to be on the bench and understand how ingredients are contributing to different features in the product you’re ultimately delivering to consumers.

The idea that AI will replace scientists - that’s hype.”

What role should AI play in creative exploration, and where does human instinct still matter most?

KT: Creativity comes from trial and error. We need time on the bench, generating prototypes, doing cocreation sessions with customers. You don’t have the tools to ask the right questions unless you have experience with the products and ingredients.

JW: We should use AI to widen, not narrow, the creative canvas. Tools like Cypris can surface unexpected flavor concepts or global trends. But only humans can validate what makes sense for other humans. AI gives more perspectives — the creativity still comes from the experts.

How do you see AI being used in a decade from now?

KT: AI will be a powerful enabler, helping us move faster, use data more intelligently and make better decisions in real time, especially when it comes to ingredients, quality, cost and functionality. Its value will be in helping us ask better questions, unlock sharper insights and create products that are even more closely tailored to what our customers need.

JW: I can see a future where R&D brings together customer and consumer insight, formulation data, sensory information, cost, nutrition, regulatory and manufacturing data in one high-quality platform. That would allow scientists to ask smarter questions and generate more targeted prototypes faster. But the key is collaboration — using AI alongside our experts, customers and suppliers to create real value.

KT: Ultimately, we can be much faster at helping our customers solve their problems, which is our primary job. We can help our customers create the best experiences for their consumers.

AI is not a replacement for a real science. It’s an enhancement, an enabler”.

The key is collaboration -using AI alongside our experts, customers and suppliers to create real value.”

Any final remarks to leave us with?

JW: All organizations must continue to invest heavily in people because of how rapidly AI is evolving. As Kathleen has already touched on, because of how easy it is to get answers from AI, we need to invest in critical reasoning, making sure that people have the skills they need to critically challenge any output.

KT: I agree, and would add that people will always be better suited to design for people. And our customers and our consumers are people. We need to make sure we give people the skill set they need, and never lose sight of the personal touch.

Finlays Solutions’ AI Pilot
by Aaron Oakes, Sensory Insights & Digital Solutions, Finlays Solutions

In 2025, the Finlays Solutions R&D team launched an AI Pilot to unlock value from existing product data by making it more accessible, structured, and usable for R&D and quality insights. A key challenge was that critical information – such as manufacturing dates, analytical results, and food safety data – was being received from suppliers in inconsistent formats and structures, making analysis time-consuming and limiting its value.

To address this, Douglas Green, Aaron Oakes and a cross-functional team, including Aukse McCoy and Youngmok Kim, piloted an AI-driven approach. The various documents (PDFs, images and Word documents) were ingested into MS Fabric, where a neural AI model was trained to scan and extract key data fields, transforming unstructured content into standardized datasets stored in a Fabric Lakehouse.

On top of this, an AI agent was developed to query the structured data, enabling on-demand insights. This was further extended with an agent integrated into Microsoft Teams, allowing users to ask natural language questions about product and supplier data.

The pilot successfully demonstrated that automating data extraction can significantly improve accessibility and analysis. Just as importantly, it created a valuable foundation for understanding where data structures and consistency can be strengthened. That learning will help us to continually improve, to enable stronger AI use cases, and unlock greater value from our data over time.

Kathleen Timperman
Chief Science & Innovation Officer Finlays Solutions

John Walsh
Chief Information Officer Finlays