Madrid bakery uses AI forecasting

A Madrid bakery is a good place to see what AI is actually useful for right now. Not robot bakers. Not a vending-machine future. Just better guesses about how many loaves, pastries, and roscones people will want on a given day — which turns out to be a much harder problem than it sounds. Pan Delirio, the Madrid artisan bakery run by Javier Cocheteux, said it started working with AI forecasting in 2019 to predict sales and avoid both waste and empty shelves. ### What problem was the bakery trying to solve? Fresh bread is brutal inventory. If you bake too much, you throw money away by the end of the day. If you bake too little, customers walk in, see the shelf empty, and maybe do not come back. Cocheteux put the daily waste range at 10% to 14%, which is a huge drag for a business selling perishable products with tight margins. ### Why is forecasting bread so hard? Because demand is not just “more on weekends.” The bakery said sales move with weather, the day of the week, holidays, and city events. A marathon can change foot traffic. A demonstration can change how people move around Madrid. That means the bakery is dealing with a messy local-demand puzzle, not a simple trend line. ### What did the AI actually do? Basically, it was a demand-prediction tool. In 2018, before the rollout, Cocheteux asked a specialist company to build a system that could estimate how much the bakery would sell of demand by product over different time horizons so the bakery can plan production and raw-material purchases more accurately. ### Did it actually work? At least once, very visibly, yes. Cocheteux said the program “clavó” — basically nailed — sales for December 19, 2019, during a period when demand was especially difficult because of the run-up to Roscón de Reyes season. That matters because holiday bakery demand is exactly where bad forecasts hurt most: you either miss sales on your busiest days or drown in unsold product. ### So why didn’t this become the permanent system? The pandemic broke the historical data. When COVID hit in March 2020, lockdowns and changing consumer habits made the old patterns much less useful. Cocheteux said the bakery effectively lost the historical baseline the model had learned from. That is the catch with forecasting systems — they work best when tomorrow still looks enough like yesterday. ### Is the bakery still using AI now? Yes, but in a lighter way. Pan Delirio said it no longer uses that original AI setup for market forecasting, yet it still uses tools like ChatGPT for job postings, client presentations, and internal training materials. So the bakery did not become an “AI bakery.” It became a normal business that uses AI where it helps. ### Why does this story matter beyond one bakery? Because this is the unglamorous version of AI that may stick. Small and midsize food businesses do not need a humanoid chef. They need fewer bad guesses. If a model can cut waste, reduce stockouts, and help a 100-plus-person bakery plan production without touching the artisanal part of the work, that is a real business use case — boring on the surface, but valuable where it counts. ### Bottom line? The Madrid bakery story is not about AI taking over the oven. It is about AI doing back-room math well enough to keep the bread basket full and the trash bin lighter. For a craft business, that is probably the sweet spot.