Garment Recommendation System

A Product developed from 0 to 1: An automated data driven approach to determine presentation order of garments on a collection page like Tops, Dresses, Workwear, etc.

Overview

Garment Collection Generation system is an automated data driven approach to determine presentation order of garments on a collection page like Tops, Bottoms, Dress, Workwear, etc. It uses a ML model trained using TensorRec to bubble up the relevant garments to the top based on the user and garment attributes and interactions.

Problem Statement

The Visual Merchandising (VM) team was tasked to manually curate collections in Shopify for every new garment launch cycle. Therefore, the Consumers were largely consuming manually curated collections which emphasize newness. The Covid-19 pandemic caused a decline in the steady stream of new arrivals and slowdown in garment launches. Lacking newness/novelty presented early in the browse experience, Thus, consumers faced the work of manual discovery deeper in the collections. This resulted in reduced satisfaction with the service and increased pressure on retention.

With the increasing number of clients, this manual curation task was anticipated to become excessively burdensome and non-scalable.

Solution

A scalable system to automatically generate an ordered list of styles for collections based on garment type like tops, dresses or garment category like workwear, trendy, etc.
A ML based garment recommender built (using TensorRec) to determine the order of the styles based on user - garment interactions and their attributes, along with the seasonality criterias.
A mechanism to enable the VM Team to switch between manual and automated curation of collections and to set thresholds on attributes like asset counts, demand/supply ratios, retirement rates, etc.

My Role

Define the MVP scope for the 1.0 launch.
Produce technical and functional requirements for the Engineering, Data Science and VM team.
Create a launch plan with minimal risks.
Use a staggered approach using phases and groups of clients and collections, to launch the new system across ~15 clients and more than 50 collections
Planned integration of the consumer services and data systems to successfully deliver the recommendations:
Conduct training and onboarding sessions for the Visual Merchandising, Marketing and Management teams.
SoPs for the VM team to configure and onboard a new collection or modify/delete the collections.
Communicate the Release notes and provide support.
Define Metrics and Dashboards to measure KPIs and track the performance of the new Recommendation System
Identify new features using the collected data and scalability concerns with the rising number of clients and introduction of marketplace concepts..
Define a roadmap for the future releases like global closeting model, segmentation based on near realtime availability of styles, etc.

Key Achievements

Eliminated a week long manual process per client, for ~15 clients
Boosted conversion rate by ~12% across ~15 clients and 50+ collections.
Streamlined and created a scalable and robust recommendation system which currently supports more than 600+ clients

Challenges

During the 1.0 launch, due to lack of infrastructure and minimal scope of the launch, A/B testing was not an option to compare the success metrics. Thus, we had to categorize clients and collections into comparable buckets and measure how metrics performed before and after the release, keeping in mind the statistical significance of the data.
The launch was generating significant apprehension among stakeholders due to its direct impact on user experience. Providing frequent updates on the launch helped alleviate their concerns. Additionally, implementing a phased rollout strategy based on risk and client impact, as well as grouping clients by inventory similarity and user demographics, instilled further confidence among stakeholders.

Learnings

Plan Strategically: Developing a well-thought-out launch plan is crucial for minimizing risks and ensuring smooth execution. By carefully considering client attributes, user demographics, and potential challenges, we can set ourselves up for success from the outset.
Embrace Machine Learning: Leveraging machine learning algorithms has proven to be a game-changer in automating processes and enhancing user experiences. Investing in ML technology opens up new possibilities for innovation and efficiency.
Scalability is Key: I've learned the importance of designing systems that can adapt and scale with the growing needs of the business. By implementing scalable solutions, we can future-proof our projects and avoid bottlenecks down the line.
Always Room for Improvement: Even successful projects offer opportunities for learning and growth. By continuously seeking out ways to enhance our solutions and processes, we can stay ahead of the curve and deliver even greater value to our clients and users.