Why AI might fix the slowest part of R&D

Key takeaways:

• AI is emerging as a practical tool to reduce trial-and-error in bakery and snack development.
• Barry Callebaut’s partnership with NotCo signals a broader push toward data-supported formulation across the industry.
• As ingredient volatility and cleaner label demands intensify, AI is poised to help R&D teams work faster, smarter and with fewer dead ends.

There’s a moment in every bakery or snack developer’s career when a project stalls so completely that the team starts questioning everything: the brief, the ingredients, the oven, the laws of physics. You try a different syrup. You adjust water activity. You tweak the fiber blend because someone thinks it might help. And then nothing moves. You’re stuck in reformulation limbo again.

It’s one of the industry’s open secrets: for all the excitement around new launches, the process of getting a product to behave the way it should is often long, unpredictable and exhausting. So when Barry Callebaut quietly announced a partnership with NotCo last year, it barely registered outside the chocolate world. But inside the broader R&D community – particularly among the people who actually spend their days wrestling with doughs and bar matrices – the move sparked genuine interest.

On paper, it’s an odd partnership.

Over the past decade, NotCo has built an AI engine trained on thousands of ingredient behaviors – not just flavor notes but structural mechanics, too: how proteins buckle under heat; how fibers destabilize a system; where moisture migrates; which combinations collapse before they ever make sense.

Barry Callebaut, for its part, brings something the AI world usually lacks: context. Real-world constraints. Practical understanding of scale-up; of how equipment quirks in one factory can ruin a theoretically perfect recipe.

Together, the two companies are trying to answer a question most R&D teams ask quietly but constantly – is there a smarter way to avoid the trial-and-error swamp that eats half our timelines?

A new development pattern

It’s worth saying that this isn’t about machines creating snacks. AI still can’t tell you whether a biscuit’s snap feels premium or cheap. It doesn’t know what ‘indulgent’ means in a culture-specific context. But it can tell you, with surprising accuracy, which ideas aren’t worth pursuing.

That alone changes the opening act of product development. Instead of developers running 20 benchtop trials just to confirm a suspicion they had on day one, AI can filter out the formulations that will fail structurally. This is the part nobody outside R&D sees: the hours lost to what-ifs and maybes. A dough that tightens unexpectedly. A plant-based fat that melts too fast. A protein bar that refuses to hold together no matter how gently you coax it. AI’s role won’t solve all these puzzles, but it will reduce how often you walk into them blind.

The pressure cooker

Part of why this moment matters is the environment bakery and snack brands now operate in. Everything moves faster than it did five years ago. Cocoa prices spike. Starch availability drops. Retailers expect quicker turnarounds. Marketing teams push for bolder claims and cleaner labels. Regulatory frameworks shift. And consumers seem to redefine ‘healthy’ every six months.

In that chaos, R&D timelines haven’t just been stressed, they’ve buckled. Most teams are already working at the limit of what’s feasible. When an ingredient shortage forces a reformulation, it doesn’t just delay a product; it sometimes derails an entire pipeline.

AI won’t fix market volatility, but it can shave weeks from the parts of the process where delays hurt most. And that’s enough to matter.

You can already see the early fingerprints

Look closely at some of the recent launches in high-fiber cookies, protein-enriched baked goods and plant-based bars. You’ll find products that hold together better than versions released a few years ago. Cleaner labels with fewer gums, but no drop in stability. R&D teams aren’t suddenly superhuman. They’re getting help: from modeling tools, from ingredient suppliers bringing new tech to the table, from systems that reduce the ‘let’s try it and see’ phase.

The Barry Callebaut-NotCo partnership simply formalizes what’s been building quietly: an industry shift away from intuition-backed iteration toward intuition-supported-by-data.

Historically, only large manufacturers could afford the kind of exploratory modeling that AI now enables. They had the budgets, the analysts and the time. Smaller and mid-sized snack dollars often didn’t. With AI capability embedded into the system, a brand with five technologists suddenly has access to the same early filtration power as a global player.

That could reshape the category in ways far more significant than any single product launch.

When a supplier as established as Barry Callebaut integrates AI into its long-term innovation strategy, it sends a message to the whole industry that the old R&D timeline doesn’t match the new commercial realities.

A quieter, more incremental shift is happening – one where product development slowly becomes faster, more informed, less punishing.

AI isn’t the story. But it might be the tool that lets the real story – the work of the people who build the next era of snacks – move at the speed it deserves.