As the economy’s strain on the climate has become unsustainable, new business models emerge with the emphasis of producing less and reusing more. Take Vinted, for example, a second-hand clothing marketplace with over 45 million users. Ovoko brings this revolution to the most unexpected of places – the used car parts industry. In Lithuania, they operate under the brand RRR.LT.
The industry has its unique challenges but is no less significant in bringing the world closer to the circular economy and ultimately tackling the pressing environmental challenges. To move their industry forward, RRR.LT develops advanced technology, including AI, computer vision, automation and analytics solutions. Their Head of R&D, Mindaugas Šipelis, talked to us about how they streamline the scrapyards’ workflow and raise the customer experience with tech.
The Interview
🔵 To start at the very beginning, what business problem does RRR.LT solve? Who are your customers?
RRR.LT is an online marketplace for used car parts, serving customers all across the European Union and beyond.
We innovate on a historically very undigitized industry, illustrated by the fact that 90% of used parts for sale are not available online. Our goal is to provide a frictionless purchasing experience for customers and open new market opportunities for our suppliers.
As a marketplace, we connect both sides: sellers and buyers. We currently have 780 suppliers on our platform – scrapyards, dismantlers, recyclers. They are crucial to us because the marketplace wouldn’t have any inventory without them. And the buyers are both regular car owners and car repair workshops.
🔵 So basically, RRR.LT takes a very undigitized industry into the 21st century?
Buying a used spare part before RRR.LT was complicated, even if the part itself was as good as the new one. The customer had to find the correct scrapyard, book an appointment via phone, physically drive there and figure out if the spare part fits your car. It might take multiple days to get the component you need.
On RRR.LT, you simply have to insert your car model, and we show you the complete list of available spare parts from our sellers with photos, descriptions, quality information, etc. You can order the spare part in a matter of clicks, and if it doesn’t fit, return it free of charge. So yes, in short, we are trying to offer this service as it should be in the 21st century.
🔵 You talked about how you make buyers’ life easier, but how does RRR.LT benefit the suppliers?
Historically, selling used car parts has been a very local business, but our marketplace allows suppliers to reach more potential customers and create more value. Sellers benefit from better service and customer support that boosts buyers’ confidence. Finally, they also gain access to market data to make informed business decisions about what cars to buy and dismantle.
🔵 Great, let’s move on to the innovative technical solutions. How have you applied computer vision in your business?
We have over 3 million different spare parts on the platform, each uploaded individually by our sellers. To list a new item, they have to pick it up on the scrapyard, recognize what part it is, inspect it, write a description and take photos. This process is quite time-consuming and requires a fair amount of detailed knowledge of the spare parts.
Our computer vision product is one way to speed up the process. When you open the camera inside our app, our technology recognises the spare part and prefills as much information about it as possible.
🔵 How did you design the product from a technical standpoint?
Reading the spare part number could seem like an easy task, but there are some tricky elements. At first, we had to build an enormous database of images of spare parts and information about them. Our advanced artificial intelligence model trains the computer vision product based on this dataset.
Also, we had to solve some challenges that are more specific to car parts. Sometimes, there might be a lot of technical information on the label of a spare part, so you have to find the correct number. Or in the case of plastic spare parts, the part number is usually imprinted, which makes identifying it with computer vision very difficult, as there could be lighting issues or unexpected shadows. Our model is already advanced enough to recognize these cases.
🔵 What’s the accuracy rate for recognizing the part number?
Accuracy is quite satisfying already – about 80 to 90%. We count the accuracy with all the edge cases where even humans can’t read the part number because it’s worn, dusty or rusty. Our model is sometimes still able to detect the number, so I think we have great accuracy.
Of course, the more popular cars have higher accuracy rates than the less popular ones. Also, it depends on the quality of the image and the specific spare part.
🔵 How far have you developed the computer vision product? When can customers start using it?
We currently run applications in the testing environment, but we’ll make them publicly available soon. Initially, it will enable our suppliers to list the spare parts on the marketplace faster, as discussed earlier. In the future, we would like to make it available to the buyers as well. If some part of your car breaks, identifying the correct component without technical knowledge is strenuous. Our computer vision product can help with that.
🔵 I understand that in addition to computer vision, you have engineered a semi-automated photo booth. How does it work exactly?
Computer vision product solves one step of the identification and listing process, but we are thinking about our suppliers’ entire workflow. There might be hundreds of thousands of spare parts in the warehouse, but as long as they sit on the shelf, the chances of someone buying them is very low.
Our conveyor project helps to speed up another step of the listing process. We custom built a machine that, in addition to identifying the spare part with computer vision, also automatically weights it and takes photos and measurements. We developed this product to be very user friendly, even if you are not very familiar with spare parts. As a supplier, you merely have to print out the label for the product and wait until someone purchases it online.
🔵 And this also aims to solve the supply problem, to get more inventory to the platform?
We currently see that uploading new products is a bottleneck for us: Although you can already sell spare parts internationally with our platform, the uploading process still relies heavily on manual labour. We try to provide our suppliers with the technology to make it faster. Another benefit for us is that we get way higher quality images and more structured data.
🔵 How much quicker can a person upload the products to RRR.LT with the conveyor?
If we compare it to the suppliers with the quickest upload speed today, this solution will be at least 3-5 times faster. But for those who are uploading the products at a slower pace, this might be up to 10 times faster.
But actually, we are doing two things at once. As mentioned previously, considerably higher data quality is an additional benefit on top of the faster process. The photos will have a higher resolution and better lighting, and the item’s description is always correct and structured.
🔵 And finally, analytics! I assume that before RRR.LT, the data about the used car parts market was challenging to find or maybe didn’t exist at all?
Yes, data and analytics are essential for us. Without the data, we couldn’t be so confident in what we are building. Currently, the scrapyards are working from their experience, but we have the data to detect trends in the car market. For example, if some car brand’s sales rise, then we know that some years down the line, the demand for spare parts will also grow.
Another example would be unmet demand in the market for some car brands. We can find those opportunities from the data and point them out to our suppliers, who can fill the gap and make good margins. The opposite also might be true – if there is already a lot of inventory for some models, then we can tell the suppliers that it’s not realistic to sell that many items.
🔵 What’s the long-term vision that you are building towards with all these new products?
Excellent question! We could change the customer experience for both sides in only a few years by bringing advanced digital technology to an industry that has never had it before.
The big vision for the future is that we will conserve resources by producing fewer new car parts. Instead, we can increasingly learn to reuse things. It will be even more crucial in the future, considering the environmental challenges we face. I think that reusing car parts will become the new normal and the customer’s first choice. By extending the usable life of the vehicle parts, we step closer to the circular economy.
🔵 Let’s jump to another subject for the very last question. You simultaneously develop both physical and digital products, including AI solutions. Does it get difficult to cover all these technical areas at once?
Yes, it’s actually very challenging to combine the different technologies. We need very high competencies on all fronts. For instance, developing hardware could look straightforward at a surface level. But when you start analyzing your specific use cases, drawbacks in the existing solutions become apparent, and you have to rethink and redesign everything.
So it is very challenging at times, but then again, we are the R&D team, so complex technical challenges are our bread and butter. Of course, it’s only possible if you have a highly competent team.
Check out Ovoko’s open positions on their career page.
