Article Icon

Translating Flavor Across the World

Jason Cohen
Published on
June 21, 2022
Populations and Sub-Populations

In 1801, Guinness decided it was time to take their beer international. The master brewers came up with a new version of the classic recipe, made using more hops so the beer would be able to survive long voyages in warm climates. Though their goal was practical — make a beer that would still taste good after weeks at sea — luckily, the new beer was a hit. West Indies Porter, with its delicious caramel and chocolate taste, along with the distinct bitterness of classic Guinness, remains a popular beer today, and many people prefer it to normal draft Guinness.

Preservatives, packaging, and travel times have all improved since 1801, meaning that CPG companies don’t have the same constraints Guinness did. Now companies can make product decisions that prioritize the flavors their target demographics will enjoy.

But how can you know what a new market wants?

Historically, to get an idea of consumer preference, you’d have to conduct tests in all the different locations and sub-locations you wanted to launch in. Consumer tastes vary so much across demographics that, of course, you couldn’t apply data from one group to draw conclusions about another. Until now.

Because the Gastrograph AI platform uses models built on primary tasting data from consumers all over the world, we’re able to translate flavor perception from one culture to another. We can reliably predict how different consumers will respond to a product — without even getting them to taste it.

Launching a product in a new location is risky and expensive. Gastrograph AI removes that risk because we help companies find which flavor combinations will be successful with different demographics. All without anybody having to get on a plane.

International Product Development Pre-AI

Without AI systems, launching a product in a new location is difficult. Either you launch without testing your product (super risky), or you conduct a lot of consumer tests (expensive, slow, and still pretty risky).

As Central Location Tests are typically a reserved resource, you’ve got to make some tricky decisions. Let’s say you have a brand that’s successful in the US. Now, imagine you want to launch a product in Europe. Before spending money on the launch, you want some evidence that this product might actually be successful in Europe. So, where should you do your tests?

“Europe's a big place. Do you do your testing in Germany? Do you do your testing in Holland? Do you do your testing in Italy? Do you do your testing in France? Or do you do your testing in all of them? That's very expensive.” — Jason Cohen, CSO of Gastrograph AI

Let’s say you do the research and find it performs well in Germany, poorly in Italy, and medium in France. “Now you're in a bind,” says Jason. “Do you launch a product that's gonna do good in one country? Do you launch a product that's gonna have a range of outcomes across the EU block?”

If you want to tweak the product or do any kind of iterative development, you’d have to do a whole new round of tests in all the different locations. Let’s say you want to make your product saltier because you think that the cultural preference in Italy is for salty food. You’ll have to do another round of testing in Germany to check if the salty version is still popular. Sounds costly and exhausting, right?

How Gastrograph AI Translates Flavor

The Gastrograph AI system allows us to predict how a new product will perform without getting anyone from the target demographic to taste it. Here’s how our system works.

Understand How Demographics Perceive Flavor

The first step is to understand how the different demographics perceive flavor differently. To do this, we’ve developed AI models that map perception for different demographics.

Not everyone from the same demographic has the same perception. You might be more sensitive to sour flavors than your siblings, for example. But together, each demographic has a unique distribution in their perception of flavor. Our AI system can learn that distribution.

We break down demographics into subpopulations (a segment of the population) and cohorts (clusters of consumers who share similar preferences).


Population, subpopulations, cohorts

Perception (which flavors you notice) varies across demographics and cohorts. Preference (which flavors you like) varies in response to differences in perception. First we model what consumers perceive, then we model their preferences.

Collect Primary Data

To train the AI system, we collect a lot of primary data. We’ve collected sensory data over the last 10 years and have the largest and most diverse sensory dataset of on-market food and beverage products.

To bring a new region onto the system, our Forward Deployed Tactical Global Panel Team recruits local consumers to taste 80–100 products over five days. We regularly collect data to grow our sensory database and update the system so we can track changes in perception and preference.

We’ve used all this data to train the AI platform, so it can understand how various demographics perceive flavor. From there, the AI system creates a baseline, so it can make predictions about how each demographic will perceive new flavors.

Translate Perception and Predict Preference

To understand how we’re able to translate perception, think about Google Translate. Type a sentence into Google Translate in French, and it maps each phrase to the equivalent in English.

It knows that:

Je → I

suis → am

grand/grande → tall

So it can send you back the fully translated phrase: “I am tall.”

Gastrograph AI works in the same way:

  • We have models for consumers all over the world (that’s the knowledge of each language that Google Translate has access to)
  • One group brings a new product onto the system (the equivalent of typing “je suis grand/grande” into Google Translate)
  • Gastrograph AI maps how they perceive the flavors onto another demographic (like translating a written sentence into a different language)

Look at how these different demographics perceive the flavor of celery soda:


The perceptions follow similar patterns with some differences — just like language

Let’s say we want to know how consumers in London will respond to celery soda:

  1. We already have perception and preference models for consumers in London and New York — that means we know how both groups respond to different flavors
  2. We bring the new soda onto the system by conducting a tasting with consumers in New York
  3. Based on what the New York consumers perceive, the system predicts how the Londoners will perceive the same product
  4. Once we have a prediction of how the Londoners will perceive the soda, the system can predict how much they’ll like it

Launch Successful Products

Companies can use data from the AI system to make smart product decisions. Gastrograph AI can predict perception and preference across six demographic factors: age, sex, race, socio-economic status, past tasting experience, and smoking habits. We can combine any of the demographic factors to define a target demographic and then predict how they’ll perceive a new product.

Gastrograph AI predictions mean, without conducting tests in multiple locations, companies can:

  • Develop new products to target a specific demographic
  • Optimize existing products to make them more likable for a specific demographic
  • Find out which demographic will like their product the most

With this information, they can launch products they know will have a high chance of succeeding.

The Corn Chips That Traveled

We worked with a company that has a successful product: a spicy corn chip that's popular with Hispanic Americans in Mexico and the USA. They wanted to launch a version of their product in Europe.

The process of traveling all the way to Europe from the US to host multiple in-person CLTs would have been immensely time consuming and expensive. Instead, the corn chip company hired Gastrograph. We were able to run predictions and get them the information they needed to optimize their product for European audiences. They didn’t even need to leave their office.

After bringing the corn chip onto the system, we researched how it would be perceived in different countries in Europe and predicted what the preference would be. The results were mixed. Some countries liked it (for example, Germany), and others didn’t (for example, the UK).

Then, we tried out some country-level optimizations. We tested changes to the flavors that would make the product more likable in each country whilst remaining within the brand of the original corn chip.

The goal was to launch a single product for the whole of Europe. We did optimization at a regional level to try to make the product likable for all countries in a region. “The country-level optimizations got better results because, if you're targeting a narrow consumer group, you can make it better for that narrow group,” explains Jason.


Are you hungry yet?

In the end, the company went for a compromise. They decided to ignore the countries that had a really low preference for the product, and they picked one of the country-level optimizations that did best across several countries.

Time for a Taste Adventure

With the Gastrograph AI system, it’s quick and easy to predict how a demographic will respond to a new flavor combination. That means new product development doesn’t have to be expensive, stressful, or risky anymore. Let your product development be a creative adventure. Let’s experiment, play, and explore. Request a demo.

Read our Whitepaper

Our double-blind validation study shows how Gastrograph AI outperforms traditional product testing.

download here