← All food signals
Infrastructure

The Model Kitchen

At packaged-goods makers, a new product now meets a model before it meets a person. Mondelez reports fewer samples; Colgate’s concepts face digital twins before real consumers.

What is it?

Packaged-goods makers are putting a model early in product development. Nestlé’s Food Genie predicts how recipes might perform. Mondelez’s tool creates recipes that a human expert then assesses. Colgate tests concepts on digital consumer twins, and General Mills takes feedback on ideas from digital personas.

Why now?

Nestlé’s chief technology officer described Food Genie to Reuters in August 2026. Mondelez told the same agency in July that its AI tool reduces the samples typically generated. Constellation Research published General Mills’ account of its digital personas that month. MIT Sloan Management Review described Colgate’s twins in January 2025.

Who’s building it?

Nestlé calls its Food Genie platform internal; Colgate-Palmolive created its own digital consumer twins to test concepts.

Food GenieMondelez AI tooldigital consumer twinsdigital personasconsumer simulations
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — Nestlé’s internal platform team and Colgate’s consumer-twin builders.
Buyers — scientists and product developers, who Palzer says use Food Genie on recipes and reformulation.
Gatekeepers — Mondelez’s human experts and the real consumers Colgate treats as gold standard.
What’s at stake
Whether a model’s early read of a recipe or concept decides which ones a person goes on to test.
Whether real consumers remain the gold standard for concepts.
Whether a small recipe archive can be modelled at all.
What’s unsettled
Can a digital consumer twin stand in for a real consumer?
How many samples does a recipe model remove at Mondelez?
What happens when a twin’s read and a real consumer’s opinion diverge?
02 · How the demand matureda progression, not a timeline
Novelty
digital twins, unlike focus groups, read scores of concepts at a time.
Retention
Nestlé’s tools run across a database of roughly 120,000 recipes.
Willingness to pay
Colgate tests scores of concepts at a time against its digital consumer twins.
Category arc
Mondelez says its tool reduces the samples typically generated through innovation.
03 · Where the market goesthe forks, not a prediction
→The model’s read comes first. More recipes and concepts are predicted or simulated before anyone makes a sample or asks a consumer.
→The real consumer stays the standard. Twins widen the field, and people still validate concepts.
→Twin and consumer diverge. A concept the twins favoured fails with real consumers, and concordance is retested.
→Data decides who can model. Palzer says a thousand-recipe archive is very difficult to model.
The forks turn on how closely a twin’s opinion matches a real consumer’s, and on how much recipe data a model needs.
04 · The Pre-Synthesis Screen desksame mechanism, a different industry
The Screened Molecule
Fragrance ingredients get screened before synthesis.
Osmo auctioned ten AI-developed fragrance ingredients at a 2026 congress.
BeautyMatter
The Screened Molecule
Real venture capital now backs an AI-screening claim.
Osmo raised $130 million total, including a $70 million round.
Premium Beauty News
The Perpetual Launch
Beauty makers report AI shortening product creation.
L’Oréal says it can now create products four times faster.
Reuters (via MarketScreener)
05 · The Wire
01
Food Genie predicts how Nestlé recipes might perform.
Reuters (via RTÉ) ↗
02
60% of Mondelez’s AI biscuit recipes performed better.
Reuters (via The Star) ↗
03
Colgate’s digital twins test scores of concepts at a time.
MIT Sloan Management Review ↗
04
General Mills uses digital personas for feedback on ideas.
Constellation Research ↗
05
Palzer: a thousand recipes is ‘very difficult’ to model.
Reuters (via RTÉ) ↗
The panel on this signal
They agreed on
–Three packaged-goods companies run a model early in product development, and General Mills describes digital personas.
–A human check follows at Mondelez and at Colgate.
–The stage differs by company: recipes at Nestlé and Mondelez, concepts and ideas elsewhere.
The sensory-evaluation scientist disagreed on
–A predicted recipe score cannot register an off-note that a trained sensory panel would catch on first tasting.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.