Sentiment analysis: what it measures and what it misses

Sentiment helps with brand and product questions. Did the launch land well? Are complaints about onboarding rising? Is a competitor's price change making people angry? If you track the share of negative posts over time, it can warn you about a problem before it shows up in churn numbers.

For sales, sentiment on its own is a weak signal, because tone and intent are different things. A furious post can come from someone who will never buy anything. A flat, neutral post can come from a buyer with budget who just writes plainly. Rank posts by negativity and you get the angriest people, not the ones readiest to buy.

These sample sentences show the gap between sentiment and buying intent. Each one has a sentiment label and a 0 to 100 buying intent score, the way HearBuyers scores posts:

• Sample, neutral, score 88. "We're moving off our current help desk next quarter. Looking for something with SLA reporting for a 15-person team." There's no emotion in it, and it's a strong buyer.
• Sample, very negative, score 41. "This help desk tool is garbage. Third outage this month." Angry, and maybe a future switcher, but not asking for anything yet.
• Sample, positive, score 3. "Loving the new dark mode on our help desk!" Happy, and staying put.

A sales team watching sentiment would chase the second post. A team watching intent would answer the first one today.

• Trusting labels on sarcasm. Many models read "Great, another outage" as positive. Spot-check samples before you trust a chart.
• Averaging everything. One sentiment score for a brand hides the one thread that matters.
• Using sentiment as a lead filter. Sentiment measures feeling. Buying intent measures whether someone wants to act.
• Ignoring context. "Killer feature" and "this is killing us" share a word and mean opposite things.

• Social listening: watching public conversation, where sentiment is one lens.
• Share of voice: how much of the conversation is about you.
• Voice of customer: what customers say they need, in their words.
• Intent data: signals that someone is likely to buy.

To see the difference in your own market, paste a neutral-sounding request into the buying intent scorer and compare it with a complaint.