BlogGuides
How much to bake tomorrow: what a study says about AI in bakeries

One click, and Google will show my articles more often in your results.
A 2024 life cycle assessment published in the Journal of Industrial Ecology helps answer how much to bake tomorrow with AI: on 2022 sales data, it measures a 30% average drop in bread and roll returns. It's the first academic study to confirm a figure close to what vendors themselves claim, with one limit worth stating upfront: the data analyzed comes from the vendor's own customers.
An academic study lands on the same order of magnitude as vendors
In 2024 the Journal of Industrial Ecology published a study by Nicolas Hübner, Justus Caspers, Vlad Constantin Coroamă and Matthias Finkbeiner, researchers at TU Berlin, focused on Foodforecast, an AI-based demand forecasting service that several German bakeries use to decide how much to produce or order each day. The study is a life cycle assessment: it measures both the software's direct environmental impact, servers included, and the indirect benefit of returns avoided compared with manual ordering. On 2022 sales data, the authors find an average 30% reduction in bakery returns, mainly bread and rolls.
It's the first peer-reviewed study to confirm a figure close to what vendors claim on their own in promotional material, the 30-34% Foodforecast itself publishes in customer case studies on its site: we wrote about that in what the data actually shows about waste in bakeries, where the only independently reported case was a Swiss one with no percentage attached.
Why it isn't a fully independent check
The study's authors are university researchers at TU Berlin, not Foodforecast employees, and the work went through peer review at an indexed scientific journal: that's a guarantee a vendor-published customer story doesn't carry. But the assessment analyzes sales and returns data supplied by the vendor's own customers, not a sample the researchers collected independently: the authors themselves write that the results remain constrained by methodological challenges and data uncertainty (source: the study's abstract, Journal of Industrial Ecology, volume 28, 2024, accessed 22 November 2026). The full text sits behind a paywall: this article relies on the public abstract and the bibliographic records that quote it, not the complete paper.
These public sources don't let us establish how many bakeries were in the sample, or how many tonnes of waste were avoided in total: some figures on this circulate online, but we couldn't confirm them in any source we could reach, so we're not reporting them.
A second project, in Germany, still without numbers
Since 3 May 2024, Germany's federal environment ministry (BMUV) has funded a pilot project, under the Green-AI Hub Mittelstand program, involving Berlin pastry maker Brammibal's Donuts and Münster start-up FoodTracks: the goal is to use demand-forecasting algorithms to cut food waste in retail transport, according to the BMUV press release of 3 May 2024. The release announcing the project contains no measured figures, only goals: more than a year and a half later, the research behind this article found no public result specific to this project, even though the wider program has published results elsewhere tied to other participating companies, not this one.
What it means for a bakery or pastry shop owner
A 30% average reduction, measured across one vendor's customers, isn't a promise for every single business: it depends on how much organized sales data the business already has, because these systems learn from history, they don't guess. It's the same principle behind the opposite case, that of a Stockholm café run by AI with no human check, which bought more than four times the pastry the café was able to sell.
Before looking at which system to choose, it's worth looking at your own till data: if it's been organized for months, a study like this gives a realistic idea of what to expect; if it hasn't, no software fixes that on its own. If you want to know whether your business already has the groundwork for a tool like this, get in touch: I'll tell you straight, without selling you something you don't need.