AI is already doing real work in food R&D. It can scan millions of compounds and point researchers toward the bioactives worth a closer look. It is helping cut salt without anyone noticing at the dinner table.
Across the food innovation chain, these tools are already delivering value. The trouble is that the results do not talk to each other.
At the Gottlieb Duttweiler Institute’s International Food Innovation Conference, PIPA founder and USDA/NSF AIFS director Ilias Tagkopoulos put this at the centre of his talk, “Food OS: AI as the Connective Tissue.” His point was straightforward: technology is no longer the biggest constraint. The harder problem is getting outputs/insights to move cleanly from one stage to the next. A connected information system can enhancefood innovation by focusing on prevention, healthspan, and personalised health
New Product Development as a Connected System
An efficient product development process behaves less like a relay race and more like a loop. Take a high-protein dessert aimed at women going through menopause: a mousse that has to carry a meaningful amount of protein and support steadier blood sugar without tasting like a supplement.
A product like this does not move from idea to factory in one clean pass. It develops through a sequence of decisions, with each stage narrowing the product and exposing a different kind of risk.
The work starts with the consumer and the product brief: who is the consumer, what are the consumer struggles and needs, what product format answers these needs, what benefits does it need to deliver.
Only then does the formulation gets more specific: what kind of protein system makes sense, what level is defensible, and how that decision ripples into taste, texture, nutrition, and how it behaves in production. From there, the chef can put together a benchtop version built around that brief. At that scale, it tends to come together well; the texture lands, it reads as a dessert, and there’s something real on the table for the team to taste, react to, and move forward with. The problem is that what behaves smoothly in a small batch doesn’t always survive the jump to equipment.
A protein level that’s perfectly clean in a beaker can turn grainy or chalky once it’s running through a high-shear mixer or a continuous process line. That usually doesn’t surface until the pilot; the first time the recipe actually meets the process at any meaningful scale. And that’s an expensive moment to find out. Pilot and factory slots are scheduled months in advance. Operators are booked, line time is paid for regardless of outcome, and if something needs reworking, you’re doing it in the most costly environment possible, fixing something that would have been a straightforward tweak two stages earlier.
In a connected system, those processing constraints are visible while the recipe is still on the bench. The chalkiness shows up in simulation, not at the pilot plant, so it gets caught and dealt with early. Once scale-up happens, the actual processing steps and line constraints feed back into the system so the next product extension starts from what the equipment can genuinely handle. And when the first testers come back and say, “It’s nearly there, but I’d only have it every day if it came in strawberry”, that goes straight to the people deciding what ingredients to look at next.
In a connected system, those processing limits are visible while the recipe is still on the bench. The chalkiness surfaces in simulation, not at the pilot plant, so the issue is caught and addressed earlier. Scale-up then feeds the real processing steps and constraints back into the system, so the next product extension starts from what the line can actually make.
And when the first testers say, “It’s close, but I’d only eat it daily if it were strawberry flavoured,” that feedback goes straight to the people choosing the next round of ingredients.
How product development as a connected system changes the cost of learning
In our work with Rivalz, a healthy-snack startup, a connected approach cut R&D cycles by more than 95% and brought three products to market in four months instead of years.
The value was not speed alone. It was seeing the real constraints of NPD earlier. A recipe can meet its nutritional ambition and still fail on texture. A benchtop sample can taste right and behave differently on scale-up equipment. A product can look stable early on, while oxidation and off-notes only surface later.
FIOS® reduces those gaps by checking recipe choices against processing constraints before pilot or factory trials. It can show, for example, how adding 4 g of fibre per serving changes viscosity and mouthfeel or how substituting cocoa powder affects bitterness, sweetness perception, and aftertaste.
FIOS® Food Innovation Operating System™ connects the dots across product development through one unified food intelligence model, helping teams streamline and de-risk reformulations and new product launches.
Talk to our team about your food innovation needs.
Event images courtesy of: Gottlieb Duttweiler Institute