AWS for Industries
Tapestry advances its analytics and transforms its Data Exchange on AWS
Global fashion house Tapestry Inc. (Tapestry) has dismantled its data silos and modernized its Tapestry Data Exchange system using cloud and fully managed solutions on Amazon Web Services (AWS). Before undergoing a full digital transformation of its Data Exchange and infrastructure, the company used an on-premises infrastructure, which was limited in its ability to scale across its global operations. By securely migrating to the cloud, Tapestry was looking to expand its data science efforts by building out its machine learning (ML) capabilities and using APIs to deliver data across the supply chain.
“We offer tens of thousands of products across our brands,” says Josh Ainsley, senior director and head of data science at Tapestry. “Being able to forecast demand is one of the big areas that we focus on.” For instance, Tapestry strives to accurately predict how many products it needs to ship to each of its stores so that it can optimize shipping costs and avoid surpluses or shortages in inventory. It also uses ML to segment customers, detect similar products, and identify bottlenecks in its supply chain.
Investing in its data science by migrating to the cloud
Tapestry consists of three fashion brands: Coach, kate spade new york, and Stuart Weitzman, which have 1,500 retail locations globally. Originally, Tapestry was known as Coach Inc. However, it changed its corporate name in 2017 and added the Stuart Weitzman and kate spade new york brands to its portfolio of brands through acquisitions in 2015 and 2017, respectively. Because of these acquisitions, Tapestry’s data systems and infrastructure were fragmented and disparate, making it difficult to use the full breadth of its data to unlock key insights and analytics that would give the company a 360-degree view of its customers. Tapestry takes customer data privacy seriously. Customer data it collects is managed in accordance with regional, state, or local privacy regulations. Tapestry Data Exchange is also used to process General Data Protection Regulation and California Consumer Privacy Act requests it receives from its customers and help it to quickly respond to new privacy regulations globally and deliver a privacy-first-oriented data solution.
As the basis for its forecasting, ML, and data storage, the company created the Tapestry Data Exchange, which processes and stores data across 100 systems and acts as a single source of truth. This Data Exchange also serves as the foundation for Tapestry’s ML capabilities. In 2018, the company decided to modernize its Data Exchange system by adopting a set of scalable cloud services using AWS. “We wanted to build strong foundations for our data science,” says Fabio Luzzi, vice president of data science and engineering at Tapestry. “So, as we move forward, we can scale, develop new data products, and deploy them at scale.”
Using APIs to transfer data across its Data Exchange system
The Tapestry Data Exchange system consists of three primary pillars: a data lake, a massively parallel processing system, and an operational data store (ODS). To transfer data across these three environments, Tapestry adopted Amazon API Gateway, a fully managed service that makes it easy for developers to create, publish, maintain, monitor, and secure APIs at virtually any scale. Using Amazon API Gateway, the company can deliver data to its applications, such as its point-of-sale systems, almost instantaneously. “We get half a million API calls per day,” says Muhammad Chaudhry, head of data engineering at Tapestry. “Our average response time is 200 ms.” During busy shopping times, the company has scaled to over 1.5 million API calls daily, peaking at 10,000 API calls per minute.
For its data lake, Tapestry uses Amazon Simple Storage Service (Amazon S3), object storage built to retrieve any amount of data from anywhere. Its data lake stores 200 TB of both structured and unstructured data from all three of its brands. To prepare its data for ML models, the company runs its data through multiple data processing services. During the first stage, Tapestry cleans its data using AWS Glue, a simple, scalable, and serverless data integration service. Then the company pulls its data into its massively parallel processing system using Snowflake Data Cloud, an Independent Software Vendor solution offered in AWS Marketplace. Using Snowflake Data Cloud as a core data processing engine, the company can model and process customer product and other enterprise datasets with elasticity. In this stage, Tapestry also uses Amazon Redshift—which helps companies accelerate their time to insights with fast, easy, and secure cloud data warehousing at scale—to model and further process its data.
For the final leg of its data processing, Tapestry pushes its data to its ODS using APIs. On any given day, Tapestry processes about 15 TB of data through its ODS. To power its ODS, the company uses Amazon EMR, which gives companies the ability to easily run and scale Apache Spark, Hive, Presto, and other big data workloads. “Working in the cloud helped us achieve two things,” says Justin Bussen, vice president of data engineering and data solutions at Tapestry. “First, we were able to get rid of data silos that existed because we previously hosted our infrastructure in different places. Second, we can scale our resources, whether it’s database power or compute power, to accelerate the completion of tasks.”
Advancing its analytics using ML on AWS
When its data has been fully prepared, Tapestry uses Amazon SageMaker to build, train, and deploy ML models for virtually any use case with fully managed infrastructure, tools, and workflows. Tapestry has built several ML models to perform near-real-time customer segmentation, product ideation, and supply chain optimization. “Using AWS, it’s simple to deploy our ML models at scale and infuse our data across different touch points in the customer experience,” says Luzzi.
For instance, the company can predict how well a product will sell by comparing it to other similar products that it has offered in the past. To accomplish this task, Tapestry uses embeddings for each of its products, which are sets of numerals that represent the physical characteristics of a product. Using these embeddings, Tapestry can accurately predict how well a new product will perform, along with how many units it needs to stock at each of its store locations. Tapestry also uses ML to identify VIP customers and bolster its ecommerce.
By running its ML and analytics on AWS, Tapestry has deepened its data science capabilities and gained a better understanding of customer behavior. “We can automatically evaluate several transactions and give our employees insights into our customers on a daily basis,” says Ainsley. “That isn’t something we could do without using AWS as our cloud provider.”