How Nykaa Turns 100K+ Customer Reviews into Smart Retail Decisions

About Customer

Nykaa is an Indian e-commerce company headquartered in Mumbai. Founded in 2012 by Falguni Nayar, a former investment banker, Nykaa has grown from an online-only platform into an omnichannel powerhouse with over 100 physical stores across India. In 2020, it became the first Indian unicorn startup headed by a female entrepreneur, Falguni Nayar. The company offers a diverse portfolio of products, including cosmetics, skincare, haircare, fragrances, and fashion items, catering to a wide range of consumer needs. 

‍

Business Challenge

Nykaa faced a significant challenge in extracting actionable insights from the vast volume of customer reviews generated daily on its platform. While rich in sentiment, these reviews are particularly useful for high-traffic categories, which are unstructured and difficult to analyze at scale. Understanding the issues of the customers and which features needed attention became increasingly complex.

‍

Manual review analysis proved to be both time-consuming and inconsistent, falling short of keeping pace with the growing data. In the absence of a scalable solution, valuable product-level insights remained buried within the noise of subjective, free-form feedback.

‍

Without a scalable system to analyse and act upon the reviews, this valuable feedback remained unused, limiting Nykaa’s ability to respond to customer needs.

‍

Shellkode solution

The project began with the collection and processing of over 100,000 customer reviews. From this dataset, we extracted a diverse range of potential aspects. These were narrowed down to 30 to 40 predefined aspects to ensure the model generated only relevant, consistent results.

‍

We initially experimented with small language models, which were cost-effective but required multiple token-limited batches to process large volumes of data, which eventually added to the cost. To overcome this limitation, we adopted Claude 3.5 via Amazon Bedrock to enable aspect-based sentiment analysis at scale. Leveraging its extended token capacity, we developed a batch-processing architecture capable of analyzing up to 30 reviews per API call, significantly reducing the cost per Batch, generating structured JSON outputs that mapped predefined aspects to corresponding sentiments - positive, negative, or neutral.

‍

Beyond model integration, ShellKode automated the end-to-end processing workflow using AWS Lambda. When a review file was uploaded to an S3 bucket, the system triggered a multi-threaded pipeline that split the data into optimized batches, invoked the LLM model, processed the results, and returned a structured output in the desired format. 

‍

Nykaa leverages Amazon Redshift to efficiently analyze large volumes of structured review data at scale. The output from the aspect-based sentiment analysis is directly pushed into Redshift, enabling seamless assessment of customer sentiment for each generated aspect across SKUs, without the overhead of managing infrastructure or query performance.

‍

Results and Benefits:

• Analyzed 100,000+ reviews to understand what customers like or dislike, helping Nykaa suggest the right products.

• Found repeated issues like “bottle leakage” using Claude 3.5 to extract key points and sentiments, cutting manual work by over 80%.

• Structured insights help teams quickly identify top pain points and respond more proactively to improve customer satisfaction.

‍