AI &Circular Economy

AI and the circular economy: what is the connection?

Artificial intelligence can help keep products, components and materials in use for longer. By analysing images, documents and operational data, it can reveal waste and opportunities that an organisation struggles to identify or manage. However, using AI does not automatically make a process circular. What matters is the physical change: preventing waste, enabling repair and reuse, or recovering materials.

The starting question is not “which model should we adopt?”, but “which loss of value do we want to reduce?”. Better forecasting may prevent overproduction; diagnostics may enable a component to be repaired rather than replaced. If the output remains in a report without changing procurement, design or operations, the benefit is still only potential.

This connects digital technologies to the principles of the circular economy. The Ellen MacArthur Foundation’s work on AI and circularity provides a useful starting point.

Applications across the life cycle

Design and materials. AI can support comparisons between alternatives, search technical information and suggest configurations. Qualified people must still verify safety, durability, repairability and material availability.

Production and procurement. Predictive models can help identify anomalies, reduce defects and manage inventories. More accurate forecasts support circularity only when they actually reduce resource consumption, surplus or unsold products, rather than simply increasing output.

Use, maintenance and reuse. Condition and failure data can guide maintenance and repair. Search systems can help match available components with potential users. Compatibility, quality, warranties and logistics remain essential: identifying a buyer is not the same as completing a reuse transaction.

End of use and recovery. Computer vision and sensors can identify materials or guide separation tasks. Output quality and viable industrial markets determine whether recovered materials will actually replace virgin resources. Recycling is part of the picture, not the only possible application.

Technologies and data: AI is not one tool

Computer vision interprets images; predictive models estimate quantities, conditions or risks; generative AI can help consult documents, organise information and develop hypotheses. These tools are not interchangeable. A text assistant cannot replace a sensor, a laboratory test or a quality inspection.

Before selecting a technology, establish which data exists, who has the rights to use it, how representative it is and how it is updated. Inconsistent product codes, missing composition data and incomplete records may be a greater obstacle than model selection. Results should also be traceable to the relevant product, batch and decision.

A digital product passport can structure useful information, but it is not itself an AI application. Likewise, an automated process may work with simple rules. When agents can take actions, permissions, approval thresholds and activity logs are needed. Orders, critical classifications and safety decisions should not depend on unverified outputs.

Three documented cases

Greyparrot describes a camera and AI system that analyses materials on sorting-facility conveyor belts. The purpose is to turn observations of material flows into operational information, not merely to recognise objects.

Refiberd combines hyperspectral imaging and machine learning to estimate textile composition. It illustrates how better material information can support textile sorting.

Hiro Robotics and Iren: an article published by Iren in 2022 describes computer vision and collaborative robotics applied to dismantling flat-screen monitors and televisions.

These cases are documented by the operators involved. They do not prove that the same solution will be worthwhile in every facility. Performance, costs and benefits need to be checked in the intended setting. Our article on these three cases explores the questions to ask before applying similar approaches.

How to design a pilot

1. Define the decision. Identify a specific problem, an operational owner and the expected change. “Reduce waste on one production line” is a more testable objective than “make the company sustainable with AI”.

2. Establish a baseline. Measure the existing process and compare the AI proposal with simpler alternatives: revised procedures, better data collection, automated rules or design changes. Comparisons need equivalent operating conditions.

3. Test representative data and flows. Include difficult cases, exceptions and seasonal variation. Decide in advance which errors are acceptable, when a person must intervene, and which results justify continuing, changing or stopping the project.

4. Connect the test to operations. Estimate integration, training, maintenance and total cost. A successful pilot must be usable by the people who run the process, not just by the team that developed the model.

Measuring results and understanding limits

Model accuracy is only a technical indicator. Circularity assessment needs measures such as materials avoided, products actually repaired, additional useful life, recovery quality and the destination of outputs. These should be considered alongside cost, time, resource consumption and reliability. Measuring circularity helps define boundaries, units and comparisons over time.

The footprint of digital infrastructure and indirect effects also matter. Greater efficiency can encourage more consumption, offsetting part of the gain. The IEA report Energy and AI examines the relationship between AI and energy systems. Benefits should not be assessed separately from the resources required to achieve them.

Other risks include inaccurate information, confidential data exposure, supplier dependence and loss of skills. Technical documents and environmental claims require traceable sources and human review. Always distinguish expected benefits, test results and impacts that have actually been measured.

From opportunities to a concrete initiative

For a business, the first step is to select a small number of opportunities that fit its material flows, products and capabilities. A project brief should identify the problem, available data, alternative solutions, necessary partners, indicators and test conditions. Only then does it make sense to decide whether to build, buy or work with a startup.

Tondo Lab works on strategy, innovation and the development of circular initiatives. AI creates possibilities that need rigorous assessment, connecting technical feasibility, economic viability and measurable impacts. The aim is to turn a promising use case into a decision that can be tested, rather than adopting technology for its own sake. The AI & Circular Economy Lab course also explores the managerial skills needed to develop these initiatives.