How to Measure Agent Empathy in Customer Calls

To understand how to measure agent empathy in customer calls, assess what the agent does when a customer expresses a concern: recognises the impact, demonstrates understanding and responds appropriately. Use a behaviour-based scorecard supported by call evidence—not apology counts, a friendly tone or customer sentiment alone.

What agent empathy actually means

In a customer call, empathy is the agent’s ability to recognise the customer’s perspective and respond in a way that shows understanding. You cannot reliably measure what an agent feels internally. You can measure whether their words and actions demonstrate that understanding.

Empathy is different from politeness. An agent can say “please” and “thank you” while ignoring a customer’s concern. Equally, a calm, direct response can be empathetic without sounding especially warm.

Consider a customer who says: “This is the second time I’ve called, and I still don’t know whether my booking is confirmed.”

A generic response is: “Sorry for the inconvenience. Can I have your reference?”

A more empathetic response is: “You’ve already contacted us and still don’t have a clear answer. Let me check the booking status so I can explain what happens next.”

The difference is specificity and a relevant action—not the presence of an apology.

Build a scorecard around observable behaviours

Define empathy before asking reviewers or AI to score it. Otherwise, assessments drift towards personal preferences about warmth, speaking style or accent.

A practical scorecard can use the following dimensions. Treat these as a starting framework, then adapt them to your call types and service obligations.

Use a simple scale for each dimension:

• Not demonstrated: The agent misses or dismisses a clear opportunity.
• Partly demonstrated: The response is generic, incomplete or poorly timed.
• Clearly demonstrated: The response is specific, appropriate and supported by the conversation.
• Not applicable: The call provides no meaningful opportunity to demonstrate that behaviour.

Require a timestamp or transcript excerpt for each rating. A score without evidence is difficult to challenge, calibrate or coach against.

Score opportunities, not emotional performances

A straightforward balance enquiry may not require an explicit acknowledgement of feelings. Marking that absence as a failure encourages agents to insert artificial empathy statements into routine conversations.

Identify the opportunity first: did the customer express frustration, uncertainty, urgency, disappointment or another concern that warranted acknowledgement? Then assess the response.

For calls with several opportunities, define whether the rating covers every relevant moment or an overall assessment. Do not silently switch between those methods.

Read language, delivery and sequence together

No single signal proves empathy. The strongest assessment considers what was said, how it was delivered and what happened immediately before and after it.

Language and specificity

Look for acknowledgement tied to the customer’s actual situation. “That delay means you cannot finalise your plans” demonstrates more understanding than “I understand your concern.”

However, do not require agents to repeat sensitive details unnecessarily. A concise acknowledgement may be more appropriate when a caller discusses illness, financial difficulty or bereavement.

Interruptions, pauses and pace

Repeatedly speaking over a customer’s explanation can indicate poor listening. But an interruption may also be necessary to clarify a critical detail, address a safety issue or manage an abusive exchange.

Silence is similarly ambiguous: it may reflect attentive listening, a system delay or disengagement. Review the surrounding conversation before assigning meaning. Accent, disability, language proficiency and cultural communication styles should not become proxies for empathy.

Response sequence

A useful sequence is to acknowledge the concern, clarify what matters and explain the available action. This is a guide, not a mandatory script.

In an urgent situation, taking immediate action may be the most empathetic response. Your rubric should reward appropriate judgement rather than insist on a fixed order of phrases.

Keep empathy separate from sentiment and resolution

Customer sentiment describes the customer’s expressed emotional state. Agent empathy describes the agent’s response to the customer. Resolution describes whether the issue was addressed. These measures are related, but they are not interchangeable.

A customer may remain upset after an empathetic conversation because a policy prevents the outcome they want. Conversely, a customer may leave satisfied after an agent resolves the issue while behaving dismissively.

Track these measures separately, then examine their relationship. For example:

• High empathy with unresolved issues may reveal policy or process constraints.
• Low empathy with successful resolutions may reveal coaching needs hidden by operational success.
• Repeated frustration after apparently clear explanations may warrant a closer review of understanding and next steps.

Avoid treating any of these patterns as proof of causation. Investigate the call evidence before deciding what needs to change.

Calibrate reviewers before scaling measurement

Start with a varied set of calls covering routine enquiries, repeat contacts, complaints, different languages and difficult outcomes. Have reviewers score the same calls independently before discussing them.

For each disagreement, ask:

• Did reviewers identify the same empathy opportunity?
• Which exact behaviour supports the rating?
• Is the rubric ambiguous, or was evidence missed?
• Would the rating change if the same words were spoken in a different accent?

Record agreed examples and counterexamples in a calibration guide. Repeat calibration when scripts, policies, languages or call types change.

If AI supports assessment, validate its outputs against calibrated human judgements. Review both missed opportunities and incorrectly flagged failures, and provide a route for agents to challenge assessments. An AI score should remain reviewable rather than becoming an unquestionable judgement.

Use conversation intelligence without losing context

Manual review helps establish a rubric, but reviewing only selected calls can leave gaps in your view of agent behaviour. Conversation intelligence can support broader coverage and help teams identify patterns worth investigating.

CX Insights combines customer emotion and agent empathy analysis with Auto QA against your own scorecards. It also trends sentiment by journey stage, product and team, helping you examine whether apparent performance differences coincide with different customer situations.

Before using any automated empathy assessment for performance decisions, confirm how it handles your rubric, languages and evidence requirements. The practical steps in this guide to automating call quality assurance provide a useful foundation for the wider review process.

Turn findings into specific coaching

Report the behaviour behind the score, not just an overall empathy average. Useful measures include missed acknowledgement opportunities, generic responses and unclear next steps, each linked to supporting evidence.

Compare similar call types and review repeated patterns before drawing conclusions about an individual. A complaints team and a routine enquiries team face different emotional demands.

For coaching, select a call moment, describe the missed opportunity and practise an alternative response. “Acknowledge that the customer has already called before asking them to repeat the issue” is more actionable than “sound more empathetic.”

Check subsequent calls for that behaviour. The aim is better listening and more appropriate responses—not more elaborate scripts.

Frequently asked questions

Can you measure empathy objectively?

You can make assessment more consistent by defining observable behaviours and requiring evidence. You cannot directly measure an agent’s internal feelings, and contextual judgement will still be necessary.

Does saying sorry count as empathy?

An apology can contribute to an empathetic response, but it is not sufficient by itself. Assess whether it recognises the specific concern and is followed by an appropriate explanation or action.

Can AI accurately score agent empathy?

AI can support assessment by identifying conversational signals and applying defined criteria. Its usefulness depends on validation against your calls, languages and calibrated reviewers, with human review for disputed or consequential assessments.

Should empathy scores affect agent performance reviews?

Use them only after the rubric is calibrated and agents understand how assessments work. Consider call context, repeated patterns and other quality measures rather than making consequential decisions from an isolated score.