CSAT vs NPS vs CES: Choosing the Right CX Metric

Most CX dashboards track all three of these, which usually means none of them drives a decision. They answer genuinely different questions, and picking the right one for a given decision matters more than collecting all of them.

CSAT — was this specific interaction good?

Customer Satisfaction measures a moment. Asked right after an interaction, it tells you whether *that* conversation went well. It's the right metric for evaluating a channel, a team, an article, or an AI agent's answers.

• Strength: immediate, specific, easy to act on.
• Weakness: heavily influenced by outcome. A customer told “no” politely and correctly still rates it poorly.
• Use it for: interaction quality, agent coaching, AI answer quality.

NPS — how do they feel about us overall?

Net Promoter Score asks about willingness to recommend. It's a relationship metric, not a transaction one, and it moves slowly. Its most common misuse is asking it after a support interaction, where it mostly measures whether the customer likes your product — not how support performed.

• Strength: tracks overall sentiment; comparable over time.
• Weakness: insensitive to specific fixes, and easily distorted by when you ask.
• Use it for: relationship health at the account or company level.

CES — how hard was it?

Customer Effort Score asks how much work resolving the issue took. For support specifically, it's often the most actionable of the three, because effort is something you directly control and reducing it reliably improves loyalty.

• Strength: points at fixable friction — repeat explanations, transfers, hunting for information.
• Weakness: less meaningful outside a support or task context.
• Use it for: self-service design, process friction, channel comparison.

The free-text box matters more than the number

Whichever you choose, the score tells you *that* something is wrong; the comment tells you *what*. Teams that read comments weekly consistently outperform teams that watch the aggregate — the number is a trigger for reading, not a substitute for it.

“A score is a smoke alarm. It tells you to go look. Nobody ever put out a fire by staring at the alarm.”— Knowledge Agents

Measuring AI agents specifically

Segment AI-handled conversations from human-handled ones. Aggregated together, a strong human CSAT masks a struggling agent and vice versa. Track thumbs up/down at the answer level as well as CSAT at the conversation level — the first tells you which content is failing, the second whether the overall experience worked.

For turning those signals into fixes, see the knowledge gaps AI exposes in CX.

Frequently asked questions