Artificial intelligence is increasingly being touted as a universal solution, capable of replacing roles across manufacturing, from the factory floor to product development. In flavour creation, that idea is gaining traction too. But while AI can support flavour development in meaningful ways, in my view, it cannot and will not replace experienced flavourists.
Flavour isn’t built purely on data or logic, it’s built on human judgement, sensory skill and creative decision-making and that’s where the limits of automation become clear.
Where AI adds value
Let’s start with where AI does make sense. AI and advanced data tools are extremely useful on the analytical and data science side of flavour development. They can help with the analysis of flavour compounds, processing of data, identifying patterns and speeding up certain stages of flavour analysis. But anyone who suggests using AI to make creative formulation decisions, simply doesn’t get it.
Flavour development is creative, sensory and highly contextual. When we create a flavour for a brand, the brief isn’t just about replicating a chemical profile. It’s about emotion, memory, expectation, application, and how that product will be experienced by the end consumer.
Sometimes the most important decisions flavourists make don’t look logical on paper. Experienced flavourists might try adding flavour compounds that, from an analytical perspective, make no obvious sense. They do it because they understand why it could work, based on years of tasting, training, instinct and experience. That ‘why’ is incredibly difficult to define, let alone teach to an algorithm.
AI can follow patterns it has already been taught. It can optimise around existing data. What it can’t do is break rules deliberately to create something surprising, that works.
The biggest limitation
There’s a very obvious issue that often gets overlooked in these conversations: AI can’t taste.
You can analyse compounds down to incredibly low levels, but human taste perception is far more sensitive and far more subjective than any machine could ever be. Taste buds can pick up nuances that analytical equipment simply can’t detect in isolation and taste doesn’t exist in a vacuum.
As we know, the same flavour will perform differently depending on application, process, temperature, texture, environment and context. The only way to know whether or how a flavour really works is for trained experts to taste it, repeatedly, in real products and that’s something no algorithm can ever do.
The risk of over-relying on AI
From a manufacturer’s perspective, there are real risks in over-relying on AI-led flavour development. If companies allowed AI to drive formulation decisions, they would end up with flavours that closely followed analytical averages. They might be technically correct, but they’d also be predictable, flat and interchangeable.
If multiple flavour suppliers used similar AI-driven approaches, the market would quickly fill with products that tasted broadly the same. There’d be no differentiation, no character and no reason for consumers to choose one brand over another.
Ironically, in this market, the very tools being sold as a route to innovation could lead to a world of bland uniformity.
Digital twins and virtual consumers
There’s also growing interest in concepts like digital human twins – using virtual models to predict how products might be received by consumers. It’s an interesting idea, and as an early-stage screening or risk-reduction tool, it may have some value. But we should be careful not to get caught up in the hype.
Taste is subjective. If you ask 20 people to describe the same flavour, you’ll get 20 different answers. Even trained flavourists won’t always agree. Digitising flavour profiles doesn’t remove that subjectivity, it just disguises it.
At best, these tools can support early decision-making. They cannot replace real-world tasting, real consumers, or expert sensory judgement.
The future of flavour development
Over the next three to five years, I expect AI to become more embedded in data handling, analytics and workflow efficiency across flavour development. That’s a good thing. What I don’t believe is that the creative heart of flavour development will, or should, be automated.
The most successful flavour houses will be those that use AI to enhance human expertise, not side-line it. They’ll invest in technology that gives flavourists more time to think, taste, experiment and collaborate, and they will be transparent about how they use AI.
Unfortunately, not every business will take that approach. In parts of the industry, flavour houses are run by finance-led teams who see AI primarily as a cost-saving opportunity. They may reduce headcount and increase automation without fully understanding the creative risks involved. I think many of them will regret that decision.
What to ask flavour suppliers
If you’re a food or beverage manufacturer talking to flavour suppliers about AI, I’d suggest asking a few direct questions:
- Where is AI being used – and where is it not?
- Who makes the final formulation decisions?
- How is flavour performance tested in real applications?
- What role do experienced flavourists still play in development?
For me, a major red flag is any claim that AI is driving formulation decisions. Using technology to support physical manufacturing or data analysis makes sense. Using it to replace sensory judgement does not.
A line that shouldn’t be crossed
AI will continue to reshape manufacturing, and flavour developers shouldn’t be afraid of that. But there’s a clear line between using technology to support expertise and using it to replace it. When that line is crossed, flavour stops being creative and starts becoming generic. We believe that food manufacturers – and consumers – deserve more.
Flavour isn’t built on data, it’s built on human judgements that are based on tasting, adjusting, and making decisions that don’t always look logical on paper. That’s the part AI can’t ever replicate. Manufacturers who understand that distinction will keep creating products that stand out. Those who don’t risk a future where everything tastes technically correct and exactly the same, mediocre at best.
