Clean Up Australia: Empowering a million volunteers to create a national litter dataset

CLIENT

Clean Up Australia

Clean Up Australia runs over 16,000 litter-collection events a year with approximately one million volunteers pitching in. The litter data those events generate feeds Clean Up Australia’s national waste analysis and provides evidence to hold packaging industry and government to account. Getting that data back reliably had never been solved.

BRIEF

Build a tool that captures structured litter data from volunteers in the field, without adding to their workload.

Clean Up Australia engaged Conduct to design and build a data capture solution that could work at scale, with volunteers of any technical ability. The tool needed to feed Clean Up Australia’s 80-plus litter categories accurately and consistently, where previous methods had failed.

The problem with the existing process

When an event concluded, volunteers were asked to empty a bag of collected rubbish and manually tally the contents against a category list. After hours of picking up other people’s waste in often unpleasant conditions, few people did it. A digital version of the same form improved response rates marginally.

The barrier was not the format. No amount of digitisation would change the fact that asking exhausted volunteers to re-engage with collected rubbish at the end of a worthy Clean Up day was not going to work.

Clean Up CAM app onboarding and litter photo guide screens designed for first time Clean Up Australia volunteers

Designing for the field

We designed Clean Up Cam around a different model: volunteers photograph litter as they collect it throughout the event, on their own device. The app runs in a browser with no download or account required. Scanning a QR code on an event flyer registers the volunteer, completes the indemnity process and links their photos to the correct event in a single step.

The design accounts for real outdoor conditions. Photos save locally if connectivity drops and upload automatically when it restores. A supervisor dashboard shows check-ins and submissions in real time. Screen reader support and keyboard navigation are built in throughout.

Volunteer holding phone running Clean Up CAM app to photograph litter in a public park during a clean-up event

Selecting the right AI

Our testing established that photos of collected piles produced poor classification results. Capturing litter at the point of collection, item by item, gave the model a problem it could actually solve.

We assessed twelve models against classification accuracy, cost at scale, output quality and hallucination resistance. Gemini 2.5 Flash was selected because it matched leading models’ accuracy within 5% at roughly 15% of the cost per event. That made it viable across Clean Up Australia’s full calendar. The AI sits in a single decoupled API call, so it can be replaced as better options emerge without touching the interface or data layer.

Clean Up CAM AI classification labels including plastics, paper and cardboard and metal applied to a litter photo

Our impact

Clean Up Cam launched for piloting in early 2026. For the first time, Clean Up Australia can capture structured, geolocated litter data from every participant across their entire event calendar.

Data insights can help drive change. The data feeds into Clean Up Australia’s annual Litter Report, which informs public policy, industry practice and community engagement. The data can map trends by region, site type and season, and drive initiatives to reduce plastic waste and its detrimental impact on the environment. Every participating volunteer receives a personalised event summary the same day, something that had never been possible before.

By designing the AI as a decoupled component rather than building it into the core system, we ensured Clean Up Australia can adopt better models as they emerge without rebuilding the product around them. The growing dataset builds toward a future model trained on Australian litter in Australian conditions: an asset that belongs to Clean Up Australia and improves with use.

This project was completed as part of Infoxchange’s Asia-Pacific Digital Transformation Project, with support from Google.org.