For retailers operating at scale, understanding customer preferences and empowering store associates is often what separates the good from the great.
Vancouver-based lululemon athletica, a global retailer of technical athletic apparel, footwear, and accessories, views its educators (retail store associates) as a direct line to how guests (customers) experience its products. To bring that frontline perspective into product and operational decisions faster, lululemon is piloting a friction-free way to capture educators’ input in the moment and automatically turn that input into insights and actionable items for other internal teams.
The challenge
Educators see, hear, and experience how guests interact with lululemon products, but the path for that feedback to reach product and operational teams has historically run through email chains and ticket systems. Each submission required manual categorization and routing, which could slow the time between submission and resolution. Along the way, insights could get lost in translation, and feedback could stay disconnected from the operational workflows where teams make decisions.
The solution
The following figure shows the end-to-end architecture of lululemon’s Retail Feedback Application, “Educator-Voice,” built on Amazon Web Services (AWS). The figure focuses on how educator feedback (voice or text) is captured, processed, analyzed, and made available for insights.
“Innovation at lululemon has always started with real experiences, and no one is closer to those experiences than our educators, who connect with our guests in stores every day. With the Educator-Voice app, we’re using agentic AI to amplify their voices and surface insights that can help shape the products and experiences our guests love. Our technology team in partnership with our product team went from idea to working prototype in just weeks, a reflection of their creativity and engineering excellence. I appreciate AWS for collaborating with us to turn bold ideas into working solutions.”
— Jebin Zacharia, SVP, Global Digital and Omni Technology, lululemon
How it works
Step 1: Feedback capture (iOS app)
lululemon designed an iOS feedback capture experience that allows lululemon’s educators to submit either text or voice feedback. The app is designed to provide an intuitive interface that store associates can use with minimal training.
- Supports both text input and voice recording
- Single-page iOS experience for maximum accessibility
- Designed for immediate, real-time submission from the store floor
Step 2: Data ingestion and ETL pipeline (Amazon S3 and AWS Step Functions)
By using Amazon S3, lululemon created a consistent data ingestion workflow for both text and audio inputs. An ETL pipeline using AWS Step Functions orchestrates all downstream processing automatically.
- Unified ingestion for text and audio via Amazon S3
- Transcription of audio feedback (if audio input)
- Translation of non-English feedback (if required)
- Automated summarization, sentiment analysis, and data artifact creation
Step 3: Agentic AI processing (Amazon Bedrock AgentCore)
lululemon built an agentic AI layer using Amazon Bedrock AgentCore for LLM-driven processing. This layer enables automated analysis of incoming educator feedback.
- Dynamic model selection for summarization, sentiment analysis of educator feedback, and classification
- Automated insights generation from educator feedback
- Final output structure optimized for downstream analytics and dashboards
- Support for natural language queries against specific store feedback data
Step 4: Insight delivery
- AI-powered insights delivery: Used AWS Quick dashboards and Natural Language Processing (NLP) capabilities to transform data into actionable insights for business stakeholders.
- Automated dashboard engineering: Fully automated AWS Quick dashboard creation using Terraform and an agentic AI approach, analyzing application interfaces and mapping them to relevant business metrics.
- AI across the SDLC: Embedded AI throughout the end-to-end software development lifecycle, spanning design, development, quality assurance, automation, and dashboard/insights delivery.
AWS technology components
The following AWS services power the Educator-Voice solution end to end:
| AWS Service | Role in Solution | Key Capabilities Used |
| Amazon S3 | Unified data ingestion for text and audio | Object storage, event triggers, lifecycle management |
| AWS Step Functions | ETL pipeline orchestration | Workflow automation, conditional branching, error handling |
| Amazon Transcribe | Audio-to-text transcription | Automatic speech recognition, multi-language support |
| Amazon Translate | Multi-language translation | Real-time neural machine translation |
| Amazon Bedrock AgentCore | Agentic AI / LLM processing layer | Summarization, sentiment analysis, classification, insights generation |
| AWS Quick | Query processed feedback using natural language, visualize trends and insights | Interactive dashboards, executive summary |
Conclusion
The solution is being piloted in 15 stores across North America, and lululemon is evaluating opportunities to extend it to additional locations. Educators submitted more than 100 pieces of feedback in the first few days of the pilot. Building on this foundation, lululemon is exploring additional uses of the app, such as gathering focused feedback on newly launched products.
“With this solution, lululemon is empowering educators to share what they see, hear, and experience, creating a direct feedback loop to our product designers. This gives our teams richer, clearer, and more actionable insights to continuously improve our products and elevate the guest experience.”
— Mahesh Maheshwari, Senior Technology Director, lululemon


