Why review scale is the point (context — real, cited)
Review corpora on India's grocery platforms are now large enough to be a real signal source. BigBasket receives over 15 million orders a month (GrabOn, Mar 2026), while Swiggy Instamart holds ~24–27% of a quick-commerce market running ~7.8 million orders a day (Startupfeed / Redseer, 2026). Blinkit, Zepto and BigBasket each draw millions of monthly visits (GrabOn, Feb 2026).
At that volume a brand generates thousands of reviews a month — far more than anyone can read by hand. Sales data tells you what happened; reviews tell you why. That gap is where this engagement lives.
The problem: ratings falling, nobody knew why
A D2C food brand watched its BigBasket star average drift from 4.3 toward 3.9 over a quarter. Dashboards confirmed the dip but couldn't explain it, and thousands of new reviews a month made manual reading a biased sample. An average rating is lossy — a 4.1 hiding “5-star taste, 2-star packaging” is only visible with aspect-level scoring at full corpus scale.
The solution
Full-corpus collection (own + competitor SKUs) → dedupe/QA → aspect-level sentiment scoring across taste, packaging, delivery condition, price-worthiness and freshness → a monthly theme report + live dashboard.
What changed
1. The packaging finding. Aspect scoring isolated packaging-in-transit as the #1 complaint (in ~1 in 5 negative reviews), well ahead of taste. A revised seal shipped; the BigBasket average recovered 3.9 → 4.4 in three months.
2. Positioning against the leader. A rival's praise concentrated on freshness-on-arrival — so the brand repositioned copy and creative around its own strongest attribute (taste), which reviews already validated.
3. Reviews became a monthly leadership input. A one-page aspect-trend report now goes to founders and the product lead each month.
