Sentiment analysis on reviews: what it actually tells you

The three questions worth asking

Which service line generates the most negative sentiment? Which staff member's replies raise post-reply sentiment the most? Which single word in your reviews correlates with churn?

The metric that matters — aspect-level sentiment

Overall sentiment averages away the useful signal. Aspect-level (delivery, price, staff, quality) is where the insight lives. A 4.6 average that hides a 2.9 on 'delivery' is an operations problem the star hides.

What sentiment cannot do

Predict revenue. Correlate with LTV in small samples. Distinguish sarcasm reliably. Any vendor promising these things is overselling.

• No LTV prediction under 1,000 reviews
• Sarcasm accuracy still under 80%
• Cross-language accuracy varies wildly
• Domain-specific tuning is required

How to act on it monthly

One meeting a month. Aspect scores on the left, actions on the right. Two commitments per meeting. That is the operating cadence that turns sentiment into product decisions.