AI and the circular economy: three cases that show where value is created

A fabric with an unknown composition. A monitor to dismantle that differs from the previous one. Industrial waste that becomes difficult to track once it leaves a company’s premises.

These are different problems, but they share a common feature: recovering value from materials requires knowing what we are dealing with and what we can do with it. When that information is missing, arrives too late or costs too much to collect, even a technically feasible solution can become difficult to implement.

This is where the relationship between artificial intelligence and the circular economy becomes tangible. Not because an algorithm automatically makes a process sustainable, but because it can help identify materials, trace flows and guide operations.

Three examples illustrate different applications and suggest a useful question for businesses: which decision could we improve if we had more reliable information?

A2A Materia: tracing the journey of industrial waste

In May 2026, A2A presented Materia, an industrial waste traceability platform developed with Amazon Web Services and contributions from MIT’s Senseable City Lab. At the time of the announcement, it was operating experimentally at nine A2A Ambiente facilities.

According to A2A, the platform uses machine learning and computer vision to classify incoming waste and trace its treatment and recovery journey. The aim is to provide information on the composition, quantity, quality and destination of materials. Source: A2A’s presentation of Materia.

The interesting prospect is to go beyond recording a waste delivery: connecting what enters a facility with what happens afterwards. For a business, this could help identify opportunities to improve the management of its waste streams.

The quality of the information remains a key question: which data are measured, which are estimated, and how are they checked? The trial documents a concrete application; it does not, on its own, demonstrate an increase in recovery or a financial return.

Refiberd: identifying fibres before choosing how to recycle them

In textiles, knowing a material’s composition is an important step in assessing where it can go next. Refiberd addresses this problem by combining hyperspectral imaging and machine learning.

The system observes how a textile interacts with light. A model processes these data and predicts its composition. The company describes a technology designed to identify fibres and contaminants, including in blended or layered textiles. Source: Refiberd technology.

Here, AI has a specific role: interpreting sensor data. The model does not recycle the textile, but the information it produces can help sort it for a compatible process.

This distinction also matters when evaluating a project. The technical page describes how the solution works, but does not, by itself, provide a verifiable business case for a customer’s facility. Before adopting it, a business would need to test it on its own material streams: which blends can it identify, what errors does it make, and at what cost?

The industrial question is not just “how accurate is the model?” but “does the resulting sorting meet the specifications of the next user of the material?”.

Hiro Robotics and Iren: helping robots handle different products

The third case concerns dismantling flat-screen televisions and monitors. In an article published in 2022, Iren describes its collaboration with Hiro Robotics and a line combining people, robots and artificial intelligence.

The challenge is variability: the products being processed are not identical, and may be worn or damaged. Computer vision and AI help manage these differences, while some operations still require human dexterity and adaptability. Source: Hiro Robotics and Iren.

This example challenges the idea that innovation necessarily means removing people from the process. The outcome depends on integrating recognition, mechanical action and operators’ skills.

Evaluating a similar solution would require looking at the whole process: treatment times, manual interventions, materials actually recovered, safety and maintenance costs. The speed of an individual operation is not enough to establish the economics of the line.

From a technology application to a business project

The three examples differ in maturity and in the results documented publicly. They nevertheless illustrate a useful connection: better identification of a material or product can support better decisions about how to manage it.

At Tondo, we believe this connection should be the starting point for an AI project in the circular economy. Not the choice of software, but a clearly defined problem and an outcome to test.

Before investing, we suggest five questions:

1. Where are we losing value? In sorting, dismantling, missing traceability or difficulty finding an outlet for the material?
2. Which decision will change? Additional information is useful if someone can use it to act on the process.
3. Do we have suitable data and the right to use them? Data must represent real material streams, not just examples that are easy to classify.
4. What happens when the system gets it wrong? Checks, responsibilities and cases requiring human intervention need to be defined.
5. How will we measure the result? The comparison with the starting situation should include the quality and quantity recovered, total costs and the actual destination of materials.

A focused initial trial can help answer these questions before expanding the investment. And if a simpler solution achieves the same objective, AI may not be necessary.

Recovery is only one part of the circular economy

These cases mainly concern the treatment of materials at the end of use. They do not cover the whole circular economy, which also includes design, product longevity, maintenance, repair and reuse.

We propose evaluating AI against these possibilities too: not only how to treat waste more effectively, but how to prevent a product from losing value too soon. These are opportunities to test, not automatic benefits.

The criterion remains the same: a more accurate prediction or a faster process creates circular value when it contributes to a concrete outcome. Environmental benefits also need to be assessed alongside the resources, energy and infrastructure required by the solution.

For a closer look at decisions before recycling, read Repair, reuse or recycle? How to choose and where AI can help.

Learn more and put it into practice