The most expensive item in the fridge may be the food you intended to eat but forgot. A packet gets pushed behind a jar, vegetables lose their freshness and the shopping list keeps growing even though ingredients are already at home. Researchers in Peru tested an AI food waste app that tried to interrupt this familiar sequence before good food reached the bin.
Their research software, ZeroWasteAI, helped participating households track ingredients, notice expiry dates and consider recipes that used what they already owned. In a study published in Sustainability in March 2026, the authors reported a lower food-waste rate by the end of a four-week pilot in Lima. The striking caveat is that only 11 households took part, with no separate control group. It is an interesting early test, not proof that a phone application can reliably cut waste everywhere.
The problem begins long before anyone opens the bin
Food waste is often portrayed as a failure of discipline: people buy too much and throw it away. The reality is usually more ordinary. A working household plans several dinners, an evening changes unexpectedly, someone eats out and the fresh ingredients sit untouched. Another shop takes place before anyone remembers what remains at the back of the fridge.
The problem is also more substantial than a few neglected carrots. The United Nations Environment Programme's Food Waste Index Report 2024 estimated that about 1.05 billion tonnes of food were wasted in 2022 across households, food service and retail globally. That estimate includes households across income groups. The burden is environmental as well as financial because land, water, energy and labour have already been used to produce and move food that never gets eaten.
There is no single explanation, so there is unlikely to be a universal cure. Some waste results from cooking too much; some from confusion over dates; some from changing family plans or food that spoils before it can be used. A useful intervention has to fit the habits of the people in the kitchen rather than assuming that everyone has time to maintain a perfect inventory.
What an AI food waste app adds to the kitchen
ZeroWasteAI was built around a simple idea: if people know what they already have, they may be more likely to use it. The researchers combined a phone-based ingredient record with image recognition, expiry reminders and recipe suggestions. The system was intended to reduce the effort of keeping track of food, a task that often defeats even well-intentioned manual lists.
Participants could capture ingredient information using a smartphone and check the quantities the software estimated. That human check is important. A photo does not necessarily reveal the exact weight of fruit, how much milk remains in a carton or whether an item was already opened. Automatically recording a wrong quantity could make the plan less useful than a scribbled note on the fridge door.
The reminders and recipe ideas were intended to turn information into action. It is one thing to know there is spinach approaching the end of its useful life; it is another to decide what to cook with it when everyone is hungry. Suggesting a meal around ingredients already present could shorten that gap. Even so, the software cannot know whether a household has time, equipment, dietary preferences or enough other ingredients without suitable information.
Privacy and effort matter as well. A system that asks users to photograph every purchase and confirm every item may become a chore. Uploading household inventory data also raises questions about where it is stored and who can access it. The study shows that such a tool can be built and tested; it does not settle the practical trade-offs involved in running a trusted service for a large population.
What the four-week experiment actually measured
The researchers followed the 11 participating households over four weeks and compared an early baseline with later measurements. They reported a food-waste rate falling from 31.3% to 23.0%. That is a difference of 8.3 percentage points, or a relative decrease of about 26.5%. The figures are worth explaining carefully because 'a 26.5% reduction' can sound like a much larger absolute change than the measurements show.
The study also collected feedback about the application's usability and participants' perception of their food-management habits. Users reported benefits such as better awareness and planning. These observations are useful in showing whether the system was practical enough for people to try, but the participants knew they were testing a waste-reduction tool. Merely paying closer attention to the contents of the fridge might change behaviour.
There was no comparison group of households continuing normal routines without the software. With such a small, conveniently recruited group in one city, it is impossible to separate the application's effect confidently from other influences such as seasonal shopping, household enthusiasm or the novelty of being studied. Four weeks also says little about whether people would keep using the service through the following year.
That does not mean the results are meaningless. Pilot studies can establish whether a design is workable and identify questions for larger trials. The point is to recognise the stage of evidence. This is a promising research prototype that merits further testing, rather than a proven consumer solution with a guaranteed saving attached.
Why the clever recipe is not always the best answer
An AI suggestion might use an ingredient technically correctly while producing a meal nobody wants. It could recommend buying extra items for a recipe that consumes one leftover but creates three more. It might treat an ingredient as safe because the printed date looks acceptable, despite poor storage or visible spoilage. These are reasons to keep human judgement central rather than treating a generated recipe as an instruction.
The food-safety distinction is particularly important. 'Use soon' reminders should not turn into advice to ignore signs that food has spoiled. The role of such a system is to support ordinary safe planning, not replace reliable storage guidance or encourage someone to take risks to meet a waste target. LiveAIWire has previously examined the weaknesses in AI food-safety and recipe advice, a related issue when software sounds certain about matters it cannot directly inspect.
There is also a difference between replacing a dinner and preventing unnecessary purchases. A useful app could recognise that a family already has ingredients for a meal, helping it shop more carefully. It should not nudge someone into using food purely because it appears on a list. Households make choices according to taste, budgets, work schedules and care needs. Technology succeeds when it adapts to these circumstances instead of issuing instructions from a perfect imaginary kitchen.
The supermarket is only one part of the journey
Waste also occurs before food reaches home. Storage, transport, packaging and forecasting can all affect how much is lost. Our coverage of AI in food supply chains considers the upstream problems businesses try to solve. The Lima experiment is narrower and arguably more relatable: it asks what technology might do at the kitchen table once the goods have been purchased.
These two levels connect. A retailer that predicts demand more accurately may prevent some surplus, while a household that knows what is in the fridge may avoid buying a duplicate. The combined impact could matter, but the benefits cannot simply be added together without evidence. Improvements at one stage might change behaviour at another, and tools designed for a warehouse are not necessarily sensible in a family home.
Likewise, digital tools in farming and food production offer a different kind of efficiency. It is tempting to present AI as the answer at every link in the chain. A better question is whether each tool addresses a measurable failure at reasonable cost. In the kitchen, that cost includes the user's time and attention, which may already be stretched.
The humbler fix could be the more useful one
Even without the experimental app, the principle behind it suggests practical habits: check what is available before shopping, keep older items visible, plan a flexible meal around foods that need using and distinguish a product's quality date from clear signs of spoilage. A phone reminder might help with one of those steps, but it need not be the only route.
A next study should compare similar households with and without the tool, follow them for longer and account for differences in family size, food spending and existing habits. Researchers would also need to show that the time spent maintaining the inventory does not outweigh the benefit. The best result is not merely a more accurate dashboard; it is less edible food discarded in ordinary life.
ZeroWasteAI offers a thoughtful answer to a familiar irritation: food disappearing into the fridge until it is too late. The experiment is small, but the question is large. If AI can make the good habit easier than the wasteful one, households may finally have a tool that earns a permanent place next to the shopping list.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.