The Plastic Whale Experiment
Recently, ING WBAA completed its second yearly experimentation week — a week full of energy, curiosity, anticipation for outcomes and celebrations. It’s such a pleasure working in a team that provides the freedom to explore interesting possibilities by dedicating valuable time to experimentation.
This time, I was one of the twelve people who chose to work for the Plastic Whale experiment. The “captain” — as we call the initiator of each idea — was Marc Rooding — Chapter Lead Software Engineering at WBAA. I was very enthusiastic from the first moment that Marc expressed his desire to run an experiment for such a valuable social good cause, and this enthusiasm was even greater for all of us once we realised we would work with Plastic Whale’s representatives, throughout the whole process. That made each step of the experiment even more valuable, realistic and tangible.
Plastic Whale (PW) is a social enterprise and their mission is focused on plastic-free waters worldwide. They achieve this by showing others that economic value can be created from plastic waste, involving as many people and businesses as possible. Their motto, “Stop Talking. Let’s start doing”, was apparent to PW people’s mindset. They are energetic, curious to learn, and ready to take action with positivity and trust — although experimentation cannot guarantee a deliverable. They want to make progress via a greater goal, and that is by changing people’s mindset and by enabling circularity. Here’s what Sarah de Beurs — Head of PW —has to say on her experience working in the experiment with her team:
“Knowing about ING WB Advanced Analytics’ capabilities and after discussing with Marc, I realised Plastic Whale had to leverage this opportunity and actively participate in this experiment. We were pleasantly surprised when the team delivered a holistic, savvy and innovative solution to our business problem in such a short timespan. We are very happy with the prototype but also with our effective collaboration with the experiment group in every step of the process. It was a pleasure exchanging insights and working together with people who share the same energy, culture and values, towards a greater goal, that is Plastic Whale’s mission for a circular economy and a plastic-free world. We wish to further continue our collaboration with WBAA in the future”.
PW aims to raise awareness in circularity and the ways we can get value out of waste. One way this can happen is by making people aware of the types of plastic materials — what can be recycled and whatnot. This is where Marc and WBAA could offer a helping hand:
“While talking to Plastic Whale about their challenges in scaling and the opportunity that technology and I could bring them, I immediately saw an opportunity in which we could utilise the strengths of our tribe. Not only would we be using our skills for doing something good, but it was also the perfect use case combining user research, data science and engineering. From the get-go, we knew that we had so many paths to explore, but so little time to do so. We were with a group, which allowed us to focus on 3 different goals” — Marc.
Plastic Whale experiment’s goals were the following:
- To develop an MVP plastic scanner capability which would be able to recognise a few different types of plastic.
- Based on the assumption that such a scanner capability is accurate and quick enough, we wanted to show PW what a possible use of that scanner could look like. We decided to focus on an Android application that would allow the user to scan a plastic item, it would try to recognise the object on the photo and based on the predictive algorithm, offer educational and informative information about that specific type of plastic.
- To do proper user research with PW. This is where we believe you should always start with. Due to the nature of the experiment week, we had limited time to conduct user research and deliver a complex working prototype as well, so we decided to work in parallel. We loosely followed a Google Design Sprint way-of-working to discover potential challenges for PW and their customer base and ideate on solutions.
From day one Marc had designed a roadmap for our way of working so that we are all assigned to tasks that were both interesting and relevant to each of us. Therefore, during our first meeting, we chose on which part of the experiment we would like to work. We also agreed on daily sync moments between the 3 teams, and between us and PW.
On day one, we had a kick-off in which we went over the 3 goals one more time, a lead was assigned per team, and then we were off. Using the daily sync moments we were always informed about every team’s progress, and we could quickly adjust or offer help to each other if that was required.
“It was really impressive to see how much work our group was able to achieve within a mere 4 days. Through the design sprint we were able to offer PW new insights and we also managed to deliver an end-to-end working mobile application which was utilising the plastic scanner that was running in Google Cloud” — Marc.
User Research
ING WBAA is a design-led tribe and therefore, we believe proper user research is where you should always start from when developing a project. We wanted to help PW create a desirable, feasible, and viable solution and worked together in this process to leverage each others’ insights.
Parisa Khanipour Roshan — Senior User Researcher and our lead, Veenam Jain — Product Manager — and myself, comprised this team and worked together on the user research part. We started by conducting stakeholder interviews with the PW team to better understand their goals and challenges. Based on our conversations, we created a research plan for user research and conducted in-depth interviews with people who had gone on a plastic fishing trip with PW at least once, to better understand their experience with PW, but also their environmental concerns and recycling habits in the broader context.
In this phase, we spoke to 12 individuals in total, analysed the findings, and identified emerging themes. We then presented all the insights back to the teams (including our partners in PW) and together brainstormed on the identified problem areas, and ideated and prototyped different possible solutions.
“It was amazing to receive so much participation from our partners at PW. They were very engaged throughout the process, a great help in recruiting participants, and actively involved in the ideation sessions. It was also a joy to work with my colleagues from different squads who I don’t normally get to work with so closely. Everyone was so enthusiastic about research and I’m very impressed by the wealth of insights we were able to uncover and the diversity of the ideas generated in such a short time” — Parisa.
Technical Background
Our data science team, led by Fariborz Ghavaniam — Data Scientist — tried to solve the identified problem by creating an app that takes a picture of a piece of plastic and identifies its plastic-type. Based on research, the plastic-types of interest — meaning those types that can be recyclable are HDPE, PET, PPE.
Identifying the type of plastic by only looking at its image is a daunting task. One needs to, at least, know the product type and its brand to correctly map the object to its material. Due to the four-days time constraint to develop and deploy the algorithm, we needed to dramatically simplify the problem.
Through discussions with PW representatives, we realised that some plastic products are of specific plastic-types, regardless of which company produced them. For instance, bottle caps, bottles, and straws are most of the time of plastic-types HDPE, PET, and PPE. We decided to go on with this simplification. We might not be able to identify all, for instance, HDPE plastics but we identify those that are in the shape of a plastic bottle cap.
“We developed a machine learning model that takes an image and classifies it into the plastic-type categories of interest. The model development procedure is described on TensorFlow’s website. To train such a model, one requires training data. We made use of a dataset of trash objects called TACO. This dataset contains images of litter in the wilderness and are categorised into several categories including plastic bottle, plastic bottle cap, and plastic straws. The trained model is then deployed to Google cloud’s platform, where the mobile application could call its API and identify the plastic-type in the snapped picture” — Fariborz.
The Plastic Whale Scanner app
While Fari's team was focussing on building the actual scanner capability, the second technical team started on ideating what a potential use-case for such a capability could be. We decided to focus on an Android application that would allow the user to scan a plastic item, try to recognise the object on the photo and based on the predictive algorithm, offer factual and educational information about that specific plastic-type.
Within the limited time of the experiment week, the team was able to create a proof-of-concept application that would take a picture, upload it to the model running in Google Cloud, process the result, and present some basic information about the plastic-type.
The application itself was written with Kotlin, using AndroidX APIs for camera usage, navigation and state management among others. Last but not least, the integration with the plastic scanner model was done using Firebase Storage for uploading photos and retrofit for querying the predictive model.
Our Vision
Being provided with the freedom to experiment can be key to success for analytics teams and organisations. Being part of the Plastic Whale experiment turned into an amazing experience with great learnings for all of us. On top of that, experimenting and working on a project that could potentially contribute to a social good cause feels like a stepping stone to a better world.
“Looking back at the experiment week, we barely scratched the surface of what’s possible. The potential in which we at WBAA can contribute to a social good initiative like PW is enormous. I, for one, would be very interested to set up a more lasting collaboration with Plastic Whale to see how AI and technology can assist them in their ambition” — Marc.
ING WBAA’s experiment week is a week of creativity, collaboration, hard work, and joyful celebrations. We all anticipate the moment when we will present our prototype, and will also check our colleagues’ solutions to their respective experiments — ending up to the awards ceremony (we get to vote and have as many awards categories as the number of experiments since we do not see this as a competition).
If you are an individual, team or organisation with a smart idea that matches our advanced analytics capabilities, and if you wish to see your idea turn into a smart prototype via experimentation, feel free to contact us here.
