Pre-develop recipe variants with AI

Food manufacturers develop new recipes and flavour variants

Future prospectAI Act: no extra obligationsDesigning & simulatingFood & consumer goods

Fits: Research & development

Aim: instead of cooking one variant after another in the pilot plant, the AI uses your own recipe and test data to suggest the variants most likely to succeed. To be honest: this depends entirely on your data. Without many of your own recipes with properly recorded sensory and analytical values, it remains a matter of collecting ideas, and taste predictions have so far only been demonstrated in individual research studies.

Is this for you?

  • You regularly develop new variants of a product group, e.g. different sweetening, less salt, plant-based substitutes.
  • From years of development, you have recipes with test, analytical and sensory values, including variants that never went to market.
  • You can run a limited project on one product group without a product launch depending on it.
First step today

First check whether your data is sufficient: use the template below to check a made-up or anonymised overview of your development data for one product group to see what is enough for predictions and what is missing. The result is a list of requirements, not yet a recipe.

How to go about it

  1. Define the development goal and hard limits
  2. Provide your own recipe, analytical and sensory data
  3. Have a limited number of variants suggested
  4. Check recipe, legal and technical aspects and select
  5. Produce in the pilot plant and test for sensory quality
  6. Feed panel and analytical values back

Templates to copy

Try it now
in the training
Analyse
in the training
Plan the rollout
in the training
Included in the training

The complete guide

  • 6 steps explaining exactly what to do
  • 3 proven templates to copy into your own AI
  • Safe use: data, checking, deciding, informing
  • Pitfalls, effort and how to recognise success

With the training you get all 430 guides for as long as your access runs.

See the trainingOne-off per person, no subscription

Same approach, other use cases

15 more use cases follow the pattern “Designing & simulating”.

Develop image ideas and a mood board for a consistent feedCreate furniture designs and cutting lists fasterDesign lighter components with generative designEstimate crash and flow behaviour of early variantsModel price increases and supply failures in advancePre-plan photovoltaics and heat pumps from aerial images and consumptionRun stress scenarios for the portfolioWork through design variants and floor plans fasterHave AI pre-select drug candidates computationallyHave AI suggest hypotheses and experimental designsHave synthesis routes suggestedPilot AI-suggested structural options
‹ PreviousSee all use casesNext ›

LK-01 · Legal status 08.10.2026