At a glance
The impact of artificial intelligence on fashion styles and branding has been immense in recent years. Apart from large eCommerce retailers or brands like Amazon, many online fashion startups are now using machine learning algorithms to understand fast-changing customer needs and expectations.
That’s precisely the case of this innovative LATAM-based online fashion startup that uses a recommendation engine to personalize the customer experience and send curated boxes of clothes directly to women’s houses.
The Startup’s business model combines customer-provided data with human fashion stylists to curate boxes of clothes based on customer’s style preferences. Internally the startup uses information that customers provide about their style, fit, and budget in a Web form. However, to scale the business model and generate better recommendations, they needed a faster way to collect personal style preferences.
The startup partnered with Zigatta’s Data Scientists to create an AI-powered game to communicate style trends and gather customer preferences. Then, the team used that data to train a Machine Learning recommendation engine that would assist fashion stylists in curating boxes of clothes they proactively send to customers.
Zigatta’s Data Science Team created a game that shows clothing items to know what customers like (and dislike) based on a recommender system.
- For the recommendations module, the data scientists developed an end-to-end ML workflow combining AWS technologies such as Step Functions, AWS Glue preprocessing Jobs, Batch Transform Jobs and SageMaker Factorization Machines.
- After that, the Data Engineers created an AWS Glue integration job for the daily generation of raw training data to feed the algorithm and architected a Serverless GraphQL API using Node.js, Apollo Graph and AWS Web Services.
- Finally, the UX and Front-end Development Team implemented a React client interface for playing the game.
With the AI-game solution, the startup was able to 5x speed-up over the initial data gathering process for style personal preferences. In addition, item returns decreased by 11%, boosting customer satisfaction, after improving the deliveries of professionally curated clothing with the machine learning recommendations module.