AI Is Quietly Cutting Trial-and-Error in Bakery and Snack R&D
There's a point in every developer's week when a project stalls so hard you start questioning physics. You swap syrups, tweak water activity, adjust fibers, and nothing moves. That stall is expensive. AI is finally helping teams hit it less often.
- 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 set to help R&D teams work faster, smarter and with fewer dead ends.
The stall no one talks about
From the outside, launches look smooth. Inside an R&D lab, you see the hidden grind: doughs that tighten without warning, fats that melt too early, bars that crumble after conditioning. Most failures are predictable in hindsight. AI's value is seeing that pattern before you burn weeks of bench work.
Why the Barry Callebaut-NotCo partnership matters
On paper, it's a surprising match. Barry Callebaut brings deep process context-scale-up realities, equipment quirks, and what actually runs on lines. NotCo brings an AI engine trained on thousands of ingredient behaviors: how proteins buckle with heat, how fibers destabilize systems, where moisture migrates, which blends collapse on contact.
Together, they're aiming at the question product developers ask every day: can we avoid the blind alleys that eat half the timeline?
A new development pattern
This isn't about machines inventing snacks. AI can't tell you whether a biscuit's snap feels premium in your market. But it can flag weak candidates early. Instead of running 20 benchtop trials to confirm a hunch, teams can filter out the formulas most likely to fail structurally-before batching, baking, or bar forming.
The result: fewer loops, tighter learning cycles, and more time on promising paths.
The pressure cooker R&D is working inside
Timelines are squeezed. Cocoa prices spike. Starch supplies wobble. Retailers push for faster turns while marketing wants cleaner labels and stronger claims. Regulations shift and the definition of "healthy" keeps moving.
AI won't calm the market, but it can save weeks where delays hurt most-during reformulation and scale-up. For many teams, that's the difference between hitting a window and missing a season.
You can already see the fingerprints
Look at recent high-fiber cookies, protein-enriched bakes, and plant-based bars. You'll notice better cohesion, fewer gums on labels, and stability that holds through distribution. That's not luck.
It's smarter screening, better modeling, and supplier data flowing earlier into the process. The Barry Callebaut-NotCo move formalizes a shift already underway: intuition supported by data, not replaced by it.
What AI isn't
- A replacement for product developers or sensory panels.
- A guarantee against flawed prototypes or tricky scale-ups.
- A shortcut to creativity.
What it is: a way to stop spending time on formulas that never had a chance.
Inside the Barry Callebaut-NotCo deal
Barry Callebaut is integrating NotCo's AI into its chocolate R&D to speed development, sharpen hit rates, and help customers handle cocoa cost and supply pressure. Peter Feld, CEO of Barry Callebaut, says the partnership "reflects our commitment to creating the best customer experience by boosting innovation and speed to market. By combining our deep chocolate expertise and global reach with NotCo's advanced AI capabilities, we're aiming to unlock speed for breakthrough recipe solutions - from health-forward formulations to functional ingredients and Nutri-Score-friendly options."
Matias Muchnick, CEO of NotCo, says combining the two data sets creates "a new standard for the industry." While the focus is chocolate, the implications extend to baked goods and snacks-where time, sustainability, and cost pressures already define the brief. NotCo's wider network includes Grupo Bimbo, Mondelez, Ferrero, and Kraft Heinz.
What to do next: practical steps for product developers
- Codify constraints first. Translate label goals, allergens, cost ceilings, and equipment limits into rules your tools can use. Garbage in, garbage out.
- Digitize your formulations. Centralize specs, process notes, and sensory outcomes. Even simple spreadsheets improve screening quality.
- Build a "no-go" library. Capture past failures with context (hydration, shear, bake profile, bar matrix). AI learns fastest from clean negatives.
- Start with screening, not generation. Use AI to eliminate weak candidates before briefed DoE trials. Keep human judgment on concept and sensory.
- Pull suppliers into the loop early. Request machine-readable data: particle size, thermal curves, water activity impact, solubility, fat profiles.
- Protect scale-up. Add line-specific quirks (mixing energy, oven zones, cooling curves) to your model inputs. Theory dies on real equipment.
- Track reformulation velocity. Measure time saved from first pass to stable pilot. Report that internally-budget follows proof.
The bottom line
AI isn't the headline. Your hit rate is. With the right data and guardrails, teams can cut dead ends, keep labels clean, and move faster without burning out. That's the win.
If your team is upskilling on practical AI for product development workflows, explore curated options here: AI courses by job.
Your membership also unlocks:
