Generative AI-Powered Customer Sentiment Analysis for a Global Apparel Retailer
The Client
A global apparel manufacturer and retailer with an extensive omnichannel presence across regions, offering a wide array of clothing and fashion products. The company receives thousands of customer reviews across its e-commerce platforms and relies heavily on customer feedback to guide product enhancements and merchandising strategies.
The Challenge
Despite the wealth of customer feedback available, the retailer struggled to convert this unstructured review data into actionable insights.
Some of the key challenges included:
- Lack of visibility into attribute-level sentiments (e.g., fit, fabric, color), which limited product design improvements.
- Difficulty in identifying the root causes of high return rates, especially for new designs.
- Manual review of thousands of comments was time-consuming and inconsistent across teams.
- Product pages displayed disorganized and lengthy reviews, leading to decision fatigue for online shoppers.
Our Solution
Affine built and deployed a Generative AI-powered review summarization and sentiment engine, tailored to the apparel retail industry.
Key components of the solution included:
- A data ingestion pipeline using Databricks to consolidate review, rating, and attribute data across all SKUs.
- Prompt engineering and vector embeddings using Azure OpenAI (GPT-3.5 and GPT-4) to extract attribute-level sentiment (e.g., 45% negative on “Fit”, 95% positive on “Fabric”).
- A dashboard and consumption layer providing business users with concise summaries of top 10 relevant reviews, sorted by sentiment and topic.
- A human-in-the-loop validation framework to ensure reliability and relevance of outputs.
- The system was integrated into monthly product analytics workflows and exposed to key design and merchandising stakeholders.
Business Impact
- 25–27% reduction in return rates, especially for poorly rated fit-related products.
- $8 million annual savings in COGS due to lower returns and better sizing consistency.
- 7–12% uplift in customer satisfaction, as product issues were addressed more rapidly.
- 5% increase in click-through rates (CTR) on product detail pages with clearer review summaries.
- Empowered product and design teams to make data-driven decisions, replacing subjective analysis with AI-backed insights.
Enabling Technology:
