CSAT vs NPS: which to use, and when each one lies

CSAT and NPS get treated as interchangeable "happiness numbers", which is how companies end up praising support for a score the pricing page just tanked, or blaming a product team for a metric that only measures how polite the last email was. They answer different questions on different timescales. Use the right one for the right question and both become useful; mix them up and both lie.

The one-paragraph version

CSAT asks "how was that interaction?" right after a specific event, usually on a 1 to 5 scale, reported as the percentage of positive responses. NPS asks "how likely are you to recommend us?" on a 0 to 10 scale, sampled periodically regardless of any interaction, reported as promoters (9 to 10) minus detractors (0 to 6), landing between -100 and +100. CSAT is a close-up of a moment; NPS is a wide shot of the relationship.

What CSAT is actually good at

• Judging conversations, not companies. Sent after each closed support conversation, it isolates the interaction from feelings about pricing or the product.
• Fast feedback loops. Scores arrive within hours of the conversations they describe, so a bad week is visible while it is still fixable.
• Segmentation. Per agent, per channel, per issue category. "CSAT is 82% overall but 61% on billing disputes" is an actionable sentence.

Where it lies: response bias (the furious and the delighted click; the middle does not), survey fatigue if you ask after every tiny chat, and agent gaming when only happy customers get asked. Report it with its response rate, always.

What NPS is actually good at

• Trend over quarters. A relationship number moving steadily up or down says something real about the whole experience.
• Segment comparison. NPS of new customers vs long-time customers, or plan A vs plan B, surfaces where the relationship weakens.
• The follow-up question. The score is honestly the least valuable part; the open-text "why?" is where the insight lives.

Where it lies: tiny samples (a 20-response NPS swing is noise), cultural scoring differences, timing effects (surveying right after a price increase measures the price increase), and the eternal temptation to chase the number instead of the reasons behind it.

The mistake that causes the most damage

Using NPS as a support-team KPI. Support influences NPS but does not control it: product gaps, pricing, and brand all push the same number. Grading a support team on NPS is grading the weather; grading them on conversation CSAT, first response time, and reopen rate is grading their actual work. (Those three, plus a few more, are covered in our guide to the customer service metrics that matter.)

What about CES, the effort score?

Customer Effort Score asks "how easy was it to get your issue resolved?" It correlates well with loyalty for support interactions specifically, because effort is what people remember about support. If your CSAT sits high but customers still churn after contacting support, run CES for a quarter; high satisfaction with high effort ("they were lovely, it took five emails") is a process problem CSAT cannot see.

A sane setup for a small team

1. CSAT on every closed conversation, automated, with response rate reported next to the score. Read every free-text comment weekly.
2. NPS quarterly, sampled across the customer base, owned by the company rather than the support team. Treat the "why" answers as the deliverable.
3. Do not merge them. One chart per metric, each answering its own question.
4. Act visibly. The fastest way to kill survey response rates is asking without ever changing anything; the fastest way to raise them is telling customers what changed because they spoke up.

Chattypie sends conversation CSAT automatically and reports it with response rates and per-category breakdowns in the inbox analytics. But whichever tool you use, the principle holds: CSAT judges the conversation, NPS judges the relationship, and neither substitutes for reading what customers actually wrote.