How ASOS Used Azure to Power Its Recommendations Platform

ASOS, the London-based fashion retailer, wanted to grow its business and become the leading online fashion store for twenty-somethings. To achieve its goal, driving its digital offering forward was vital. However, with a customer database of over 15 million customers, it needed a system that could support its innovative plans. The company turned to Microsoft Azure to build a scalable, managed database that would enable it to make real-time product recommendations and instant order updates. The migration to a microservices architecture allowed ASOS a greater ability to scale, innovate and incorporate new customer engagement models. 

What Challenge Was ASOS Facing?

Due to the extent of its operations, ASOS offers a vast selection of fashion items. The online fashion destination provides 85,000 items at any time with 5,000 new ones added each week. The company goal is to help its customers discover new products that they’ll love, but to do this it needed a platform with the power to create recommendations for over 15 million customers. 

The challenge ASOS was facing was to improve its customer recommendations modelling. It needed to provide shoppers with targeted suggestions but also needed to consolidate the three separate solutions it was using for its data models at the time. Ultimately, ASOS needed to overcome the inefficiencies of its data modelling projects so it could focus on fine-tuning its recommendation service. 

Why ASOS Chose Microsoft Azure

The company chose Azure Cosmos DB to manage its vast customer database because of its global distribution. It also valued the solution’s ability to deal with heavy seasonal bursts, scaling easily to meet user-demands. Azure Cosmos DB gave ASOS the ability to provide a more personalised shopping experience, while at the same time speeding up order updates even within the busiest days of the year. 

With Azure Cosmos DB, the online system can analyse a huge volume of product data and produce relevant recommendations instantaneously. What this means for customers is that they have a vastly improved shopping experience. What it means for the company is that its software designers and engineers can focus on building competitive advantage rather than worrying about server infrastructure. 

Azure Cosmos DB is used across various microservices. A low latency data store enables precalculated user profiles to be stored and retrieved, and product machine learning models allow real-time recommendations to be generated. Thanks to Azure’s global distribution, data models can be distributed to be near the relevant microservices, regardless of location. 

The ASOS Recommendations Platform

The ASOS recommendations platform is based on lambda architecture, split into an offline engine and the online service. Within the offline engine, training and storage or a range of machine learning models is supported. The machine learning models calculate in real-time the product relevancy and the chances of a customer viewing, saving and buying a product. The recommendations platform includes several key elements:

• Multiple machine learning models built by batch-processed telemetry.
• Azure Data Lake Store for the long-term storage of user interaction telemetry.
• Apache Spark MLlib machine learning library to generate versions of the user and product vector models.
• Azure Data Factory to bulk-load vector models into Azure Cosmos DB.

The online recommendations service delivers product recommendations in real-time as users browse listings. To deliver this, the service relies upon low end-to-end latency and quick page-load times. The workload was a natural fit for cloud technologies. Azure Service Fabric hosts the computation and personalised decision engine while the precomputed user profile and product vector data is stored in Azure Cosmos DB. The result is a service which delivers millisecond random access reads and is elastically scalable through peak shopping periods. 

Azure Machine Learning Service

As well as needing to manage its database, ASOS wanted to develop a structure for its data science teams. Microsoft Azure Machine Learning service has enabled the company to reduce time-to-market for new recommendation models by delivering some core benefits to its data scientists:

• No longer have to manually configure virtual machines
• Can focus on models and running experiments instead of worrying about infrastructure
• Removal of  barriers for algorithm creation
• Cross-functional teams have become more productive
• Development of best practices for dealing with data science and modelling

By incorporating Azure Machine Learning service, the company has not only solved many technical challenges, but it’s also made huge organisational change. AI transformation has produced a new collaboration structure for its data science teams. By unifying cross-functional teams with the right technology, they have vastly accelerated the modelling process. Development time has been reduced from months to just a few weeks. 

Could Microsoft Azure Consultants Help Your Business?

If like ASOS your business has a vast customer database and needs to be able to process huge volumes of orders on peak days such as Black Friday and Cyber Monday, migrating to Azure could help. Azure Cosmos DB has the ability to accelerate order workflows and event-driven microservices enable customers to easily track their orders. Azure Cosmos DB has allowed ASOS to create a delightful discovery and shopping experience for its customers and has freed up its data scientists and engineers to build further competitive advantage into its offering. One of Azure’s big selling points is its high availability. With Azure your business is able to scale an application architecture over many data centres around the world. 

To summarise, here are the challenges, the solution and the benefits that ASOS experienced:

• Challenges – ASOS, the leading online fashion destination, aimed to help its customers discover, and, ultimately, buy products that they would love. To achieve this, it needed to build a platform that had the power to create recommendations for its database of over 15 million customers. The company needed to overcome the inefficiencies of its data modelling process so it could focus on innovation and building competitive advantage. 
• Solution – ASOS turned to Azure for the solution to its problems. Azure Cosmos DB offered a global distribution that would also be able to scale to meet seasonal changes in demand. Meanwhile, Azure’s Machine Learning service offered the ability to transform its modelling process.
• Benefits – Azure’s database enables ASOS to analyse a huge volume of product data and to deliver relevant recommendations instantly. This, coupled with Azure’s Machine Learning service, has enabled the company to vastly reduce time-to-market for new recommendation models. The result is a scalable, intelligent solution that delivers an incredible user experience.