As traditional high street shopping takes a back seat in retail and e-commerce continues to lead the way, online marketplaces such as eBay and Lalafo continue to grow and further establish their place in the industry as consumers become more and more comfortable buying products from third party suppliers. Whilst eBay is the world's ninth-largest internet company by revenue, the market size is increasing with consumer demand and competitors are thriving with similar platforms.

Everyone buys new products, but only a fraction of consumers go on to sell them afterwards. Think of those trainers you bought for the marathon than you never ran - they’ve not even left the box. It might be a small inconvenience to list, package and ship the items, but the potential return is huge. Lalafo are currently researching user behaviour in peer to peer trade and have identified several barriers that stop people from selling items on a regular basis and are working to see if they can remove these with the help of AI. Methods such as automatically classifying products by image and text, predicting the prices, providing content related recommendations have all been done, but how does this transfer into a valuable practice?

Lalafo focus on specific challenges in user-generated data cognition and the application of DL models in a C2C marketplace with millions of users. At the Deep Learning in Retail Summit in London next March 15 & 16 we will hear from Yuriy Mukhin, Co-Founder at Lalafo. In advance of his presentation, we asked Yuriy and Denis Troyanov, Lead Data Scientist, about some of their current work on the platform.

They explained that Lalafo have been working in C2C classifieds space for 8 years in very successful classifieds with tens of millions of users, however they’re seeing that most people still don’t sell anything. People keep thousands of unused items in boxes making their lives cluttered and stressful. ‘We believe this happens because selling experience is full of friction.’ The creation of the AI powered marketplace that knows what you have, how you use it, how much it costs and who would like to buy it is overcoming this problem. ‘We want to help people own only things they really enjoy and get rid of all other stuff instantly in one tap.’

We analyzed points of friction and applied AI to remove them. First off, we decided to teach Lalafo to see. When our user posts an image, we instantly know what kind of product it contains, its category, brand, year of manufacturing, colour, shape and many other characteristics of the item. We call this technology, built in-house, - Product DNA. Based on Product DNA other Lalafo products suggest prices and match items to people who would like to buy them.

Of course there are challenges in building deep classification systems and the ecosystem around it. Denis explaind that when you have tens of thousands of classes it’s all about smart architecture, computing power and lots of data. There are so many online retailers that aren’t structured enough and Lalafo are working to overcome this.

The implementation of AI completely transforms the user experience, making it infinitely easier to sell and buy online. ‘Imagine an assistant that knows what you have, how much it can be sold for, who would like to buy it and helps you make effortless deals.’ However, the lack of data, unbalanced datasets, a latent hierarchy in data-manifolds, very noisy data, limited generalization ability of wide-known classification models, and a limited time for re-training the models under the conditions of constantly evolving product catalog all contribute to the challenges in perfecting the model.

If you’re keen to hear more about how Lalafo are working to overcome these challenges and implementing AI to transform the ecommerce marketplace, join us at the Deep Learning in Retail Summit, March 15 & 16. Other confirmed speakers include Alessandro Magani, Data Scientist at Walmart Labs, Kostas Perifanos, Lead ML Engineer at Argos and many more who you can view here.