Size recommendation for e-commerce powered by AI using neural networks


Artificial Intelligence (AI) has become a buzzword. Many e-commerce, trades and businesses claim to improve their services, or introduce new products that are made better thanks to newly developed technologies using intelligent algorithms. At the heart of AI are simple, but yet very powerful systems called Neural Networks.

What are neural networks and what do they do ?
Artificial neural networks, also called connectionist systems, are computing algorithms that are inspired by the biological brain of animals, including humans. At a cellular level, the brain is formed of neurons that manage the flow of information, converted in small electrical currents, from sensory inputs to decisional output. Artificial neural networks consist of the juxtaposition of layers of neurons. The input layer receives the input signal, or data. The hidden layers perform mathematical operations based on the signals received, which will determine if they fire more signals down the chain. Ultimately, a resulting signal will reach the output layer with the outcome or result. The tricky part is to determine the exact mathematical operations to apply at each neuron in order to reach the correct outcome. Teaching using known input-output pairs is used to calibrate the neural network setup and get it ready to process unknown input to predict the most likely outcome.

neural network schematics

Progress in computer technology but also access to vast amount of data has allowed to unlock the potential of “deep learning”, which is the process of teaching neural networks to identify, recognise and take decisions based on pre-existing, and/or newly created, data.

Prime AI bespoke implementation
Google, Amazon and universities around the world have been developing neural networks platforms for a varied number of applications. Most companies would use these pre-existing platforms to create and train their models. However, these tend to be heavy, slow and overly complex due to the fact that they have to address a wide spectrum of potential applications.

Prime AI uses no third party platform and has developed its very own proprietary neural networks technology. The result is a fast, agile and accurate implementation that can run on a basic personal computer and be interrogated in micro-seconds via a web-browser.

Application to fashion retail
When Prime AI size recommendation widget is first rolled out to a retailer’s website, it will straight away be as accurate as the old fashion size charts, but will offer a much more intuitive, reliable and easier user interface for customers to get their size. Resulting in an immediate increase in customer confidence to complete a purchase, and/or buy more items. The neural network training and evolution does not stop here ! Prime AI gathers successful sales transactions and all returns, which are fed back to the neural networks to fine tune and evolve their understanding of how customers of a specific retailer relate to garment sizes down to SKU level.

body shape to size

Continuous training using sales and returns
Prime AI uses a retailer’s standard size charts (or similar data) to perform the initial training of its neural networks, by informing them on how a customer’s biometric data relate to its theoretical garment size for each garment (at SKU or category level as required). Also, the technology is capable to integrate with retailers CRM database and loyalty program to learn from customer’s past transactions in order to deliver quicker reduction in returns.

body shape to size

Prime AI innovative and exclusive neural network implementation for the purpose of matching customer body shape and garment size enables very fast re-training. Experience has shown that meaningful shift in the correlation can be obtained with as little as a handful of new data points. Leading to a very agile, adaptive and responsive technology capable to cope with the fast pace of fashion retail where new SKUs are introduced, sold and replaced within short space of time.

The accuracy of the tool has been validated by a measured reduction in returns but also an increased number of customers following the recommended size when shopping with the help of the intelligent size recommendation tool.

Add the most advanced size recommendation technology to your website
Adding Prime AI size recommendation widget is as easy as adding 2 lines of code to a retailer’s online store. We only need information to train the initial neural networks, most of the time in the form of already existing standard size charts. Prime AI does all the rest. Our server handles all the data, all the codes and all calculations, adding no weight on retailers’ e-commerce platform.

Contact us now to take advantage of the most advanced AI powered size recommendation tool on the market.