Review & Sentiment Analysis

The Voice of 10,000 Customers

A D2C skincare brand on Amazon India has 8,500 reviews across 12 products. The founder reads the first page of reviews for each product — maybe 50 total. She sees some complaints about packaging and a few comments about the fragrance being too strong. She adjusts the fragrance in the next batch.

But buried on page 47 is a pattern she never sees: 230 customers across three products mention that the pump dispenser breaks within 2 weeks. This is not a fragrance problem — it is a supplier quality problem that is silently driving returns and killing repeat purchases. AI reads all 8,500 reviews in 30 seconds and surfaces this pattern immediately.

This is the power of review and sentiment analysis. Your customers are already telling you what is wrong, what is right, and what they want next. The challenge is not collecting feedback — Indians are enthusiastic reviewers — but processing it at scale. This chapter teaches you how.

Why Reviews Are Gold

Reviews serve three business functions that most retailers underestimate:

1. Product Intelligence

Reviews tell you what your product team should fix, improve, or create next. No focus group or survey gives you this quality of unfiltered feedback at this volume.

2. Trust Building

In India, 85% of online shoppers read reviews before purchasing. Products with 100+ reviews and 4+ stars convert 2-3x better than products with few or no reviews. Reviews are not just feedback — they are your sales team.

3. SEO and Discovery

On platforms like Amazon and Flipkart, products with more reviews rank higher in search results. Reviews contain natural language that matches how real customers search ("lightweight cotton kurti for office" is more useful than your catalogue description "Ladies Regular Fit Kurta").

Net Promoter Score (NPS): The One-Number Summary

NPS is the simplest way to measure overall customer satisfaction. It asks one question: "On a scale of 0-10, how likely are you to recommend us to a friend?"

NPS = % Promoters - % Detractors

Example:
100 responses: 45 Promoters, 30 Passives, 25 Detractors
NPS = 45% - 25% = +20

Indian e-commerce benchmarks:
  Excellent: > +50
  Good: +30 to +50
  Average: +10 to +30
  Poor: < +10

Open data/nps-survey.csv to see NPS survey responses from 500 customers of an online home goods retailer — including the score, open-ended comment, product purchased, city, and purchase channel.

Sentiment Analysis: Beyond Stars

A 3-star review can mean very different things:

AI-powered sentiment analysis goes beyond the star rating to understand the emotional content of the text. It classifies sentiment at the aspect level:

This aspect-level analysis tells you exactly what to fix. A product with great quality sentiment but terrible packaging sentiment needs a packaging supplier change — not a product redesign.

Common Indian E-Commerce Complaints

Analyzing millions of reviews across Indian platforms reveals consistent pain points that AI can automatically categorize:

Delivery Issues (35% of negative reviews)

Product Mismatch (25% of negative reviews)

COD-Specific Issues (15% of negative reviews)

Packaging Problems (12% of negative reviews)

Return and Refund (13% of negative reviews)

Open data/product-reviews.json in the code panel. This file contains 2,000 product reviews from an Indian e-commerce platform with star ratings, review text, product category, verified purchase flag, and timestamps. This is the dataset you will practice sentiment analysis on.

Using AI to Analyze Reviews

The Basic Prompt

Here are 50 customer reviews for our [product]:
[paste reviews]

Please:
1. Classify each review as Positive, Negative, or Mixed
2. Extract the top 5 complaints (with frequency count)
3. Extract the top 5 praises (with frequency count)
4. Identify any safety or quality issues that need immediate
   attention
5. Suggest 3 specific product or service improvements based
   on these reviews

The Advanced Prompt (for larger datasets)

I have exported 500 reviews for our product line. Here is the data:
[paste or reference data file]

Perform aspect-level sentiment analysis:
- Aspects: quality, delivery, packaging, pricing, service
- For each aspect, give me:
  a) Sentiment score (-1 to +1)
  b) Top 3 positive phrases customers use
  c) Top 3 negative phrases customers use
  d) Trend: Is sentiment improving or declining over the
     last 3 months?

Also flag any reviews that mention:
- Health or safety concerns
- Legal threats or regulatory complaints
- Competitor comparisons (what are they comparing us to?)

Acting on Sentiment Data

Analysis without action is a waste. Here is a framework for turning sentiment insights into business decisions:

Severity Matrix

The Feedback Loop

Key Takeaways