Case Study 06 · Text Analytics & Consumer Insights
What 1.3 lakh app reviews tell you about what Indian investors actually want
The brief
Everyone talks about customer centricity. Very few actually go into the data and read what customers are saying at scale.
I ran a full sentiment analysis and topic modelling project on Google Play Store reviews from three of India's biggest investment apps, 1.3 lakh+ reviews, real user feedback, and the findings told a story that most fintech marketing teams would find uncomfortable.
At a glance
Groww, Zerodha & Upstox
1.3L+ Google Play reviews analysed. Tools: Python, VADER sentiment analyser, LDA topic modelling (Gensim), aspect-based sentiment analysis, correspondence analysis. Purpose: uncover real consumer insight for fintech marketing strategy.
The challenge
Survey-based research tells you what people say when they know they're being asked. App reviews tell you what they say when they're not. The dataset was 1.3 lakh reviews. The challenge was extracting signal from that volume in a structured, defensible way.
Volume and noise
1.3 lakh reviews across three platforms. Without structured NLP methodology, it's unreadable noise. Needed a framework that produced defensible, actionable insights.
Aspect-level specificity
Overall sentiment scores hide where the real problems are. Needed to break sentiment down by aspect: trading execution, app performance, customer support, multilingual support, settlement time.
Marketing translation
Technical findings are useless unless translated into marketing decisions. The output needed to tell a brand exactly where to invest their messaging and why.
The approach
VADER sentiment classification across all 1.3L reviews.
Classified each review as positive, negative, or neutral with a polarity score. Established the baseline emotional landscape across the three platforms.
LDA topic modelling to find hidden clusters.
Latent Dirichlet Allocation via Gensim to identify 7–8 topic clusters (coherence score optimised). Correspondence analysis mapped topic clusters against star ratings to validate findings.
Aspect-based sentiment scoring.
Sentiment scored specifically across: trading execution, app performance, settlement time, customer support, multilingual support, and overall product benefits. This is where the real actionable signal lives.
Translated data into marketing strategy.
Where to win, where to defend, what messaging angle builds trust, and what the brand that dominates this space in the next three years will communicate differently.
What the data said
Key takeaways
01
Reviews are your most honest focus group. They cost nothing and they're already there. The brand that reads them at scale and acts on the signal will build trust faster than any campaign.
02
Positive sentiment about features doesn't equal loyalty. Negative sentiment about service will override it every time. Customer support isn't just a product problem, it's a marketing one.
03
Sentiment analysis at scale tells you where to invest marketing messaging. Not what people say they want, what they actually respond to emotionally.
Want research
that drives strategy?
Consumer insight only matters when it changes what you do next.
Connect on LinkedIn ↗