Applied AI & AI in business /

How Bower is scaling up with AI to recycle in new markets

May 18 2021/4 min read

We built an AI solution that helps us review crowdsourced images of recycling stations, which cuts manual review time by 75%. Let us explain more how we did it and how this led us to win an AI competition by Peltarion.

This is a guest post by Simon Asp, Design Technologist at Bower. Check out how their app works at

02/ The business case

We have a digital recycling app that lets you get rewarded each time you recycle everyday packaging. Our idea is that you should be able to recycle and get rewarded anywhere, any time.

We use a location-based patented technology to ensure that the user actually recycles their packages at a recycling station. Each month, the users of our app recycle around 1.5 million packages, and all of them are recycled at any of our 30k+ recycling stations that you can find in the app.

How the Bower app works

Some of these stations we have added ourselves, but we can’t add private recycling stations since the data is not available — so we use a crowdsourcing model where the user can upload the station from their home.

— But how do we make sure that the information is correct?

We have been handling user-uploaded stations manually, which has been great for a small-scale solution. But as we’re entering new markets, and rapidly growing with more customers — the manual labour is getting harder, to say the least, especially since most of our stations are crowdsourced.

03/ Solution: Machine learning

We’re now proud to say that we built a solution that cuts down the time we spend on reviewing these stations by 75% and also won a competition from the leading AI company Peltarion!

We built an algorithm that can predict if an image is a recycling station or not. It then takes a decision based upon the prediction — and based on that, we can semi-automate our review process, meaning that we’re still in control in certain cases. And the good part is that as we continue to review images, we also train the algorithm to become better.

The ML solution can predict if an image is a recycling station or not - Original image from

04/ Scaling

The solution for our app is great in many ways. Most of which, being able to scale to new countries faster. Our vision is simple:

We want to create the first generation free from trash, and anyone should be able to recycle anywhere in the world with the Bower app.

Our company is based in Sweden, and recycling infrastructure and habits in our neighboring countries are similar, but it always differs to some extent, and even more so by going beyond the Nordics. By training a machine learning model to understand how recycling stations look in each country, we can open up our digital solution much faster, with fewer human resources.

05/ For the user

One aspect of the ML model is that we get a much better user experience in the app. As of now, a user has to wait to get their recycling station registered since we have to look at it. If the image is not accepted, you would have to go back and take a new photo.

Imagine going out in the rain to take a photo of your waste bin for the second time. Not so fun! With the AI solution, you can get instant feedback if the image did not meet our requirements.

Bad image? — Image from Aaron Burden.

06/ For us

We shouldn’t forget that we spend much time going through these images, which a quite boring task if you ask me. It just has to be done. I wouldn’t say that you learn anything after looking at thousands of images of stations. Although, sometimes there can be images that you wouldn’t expect (you never know what users send in).

07/ The AI for Startups-program

The whole thing started when Peltarion, a leading no-code AI platform hosted a program and competition called “AI for Startups”, and adding to that, we just got contacted by a brilliant student, David — Interaction design student at Umeå University, who wanted to write his master thesis at Bower.

Peltarion, a no code operational AI platform.

The competition lasted for six months, where we could utilize Peltarion’s platform free of charge, plus an additional six hours of expert consultation.

I can personally say that the platform is really great — it lets you whip up machine learning models in minutes, from training, evaluating and deploying them. For us, this meant being able to create multiple models quickly, which we could try out easily with the one-click deployment. We could then run measurements on the model with real-life data, and tweak the solution even more.

Thanks, Peltarion for a great platform and great help, and David for being an awesome master thesis student.

We can’t wait to develop this; to grow faster, give a better experience for our users, and to relieve our support team!

Bower is based at the Norrsken Impact Hub in Stockholm.

  • Simon Asp

    Design Technologist at Bower

    Simon is the Design Technologist at Bower – a digital recycling app that lets you get rewarded each time you recycle everyday packaging. The winner of Peltarion's Startup Program, their machine learning solution built on the Peltarion platform cuts the time spent reviewing crowdsourced images of recycling stations by 75%.

02/ More on Business & Applied AI