Empathy training for customer service: language and practice

Short answer

Empathy training for unhappy customers is not about memorising "we understand your frustration". Empathy is four observable behaviours: restating what the customer said in your own words, naming how the problem affects them, taking ownership, and moving straight to a concrete action. These can be practised and assessed in simulation, because each is visible in the conversation. Assessing tone and empathy calls for care, though: tone reads differently in chat and on a call, and AI can misread subtle nuance.

Why scripted phrases don't work

Many support teams hand agents ready-made phrases: "We understand your concern", "We apologise for the inconvenience". Customers have heard them many times and read them as procedure, not empathy. Especially when nothing changes after the sentence — the same wait, the same generic answer — a scripted line raises anger rather than lowering it.

Empathy lives in specifics, not in the sentence itself. "You've been waiting for the courier for three days and took time off work again today" sees the customer's situation; "we understand your concern" sees nobody.

The four behaviours of empathy

1. 1AcknowledgeRestate the problem briefly in your own words: "The order was due on Monday and still hasn't arrived."
2. 2Name the impactSay what the problem cost the customer: "So you had to wait at home again today."
3. 3Take ownership"I'm checking this now" — not "that's how the system works".
4. 4ActA concrete next step and time: "I'll speak to the courier and write back within 10 minutes."

Phrases that break empathy

• "Calm down" — telling the customer what to feel.
• "It's not our fault" — defence instead of a solution.
• "That's the rule" — a rule cited without explanation.
• "Other customers don't complain" — dismissing the customer's experience.
• A long apology followed by no action.

Handing these out as a banned list is not enough — agents must also practise what to say instead.

How to set up the practice

Empathy practice runs on an unhappy-customer scenario. The AI customer starts out irritated and reacts according to its profile: it calms down when the agent gives a concrete acknowledgement and action, and sharpens at a scripted phrase or "please wait". How to write the scenario card is covered in AI customer simulation for support.

1. 1Round 1The agent runs the scenario as they normally would.
2. 2FeedbackWhich of the four behaviours were there and which were missing — with concrete sentences.
3. 3Round 2The same scenario, focusing only on the weak behaviour.
4. 4Round 3A different scenario — to check whether the skill transfers.

Empathy in chat versus on a call

On a call, much of empathy is in the voice: pace, pauses, not interrupting. In chat there is no voice and everything travels in words: short but not cold sentences, using the customer's name, careful use of emoji. What kills empathy most in chat is delay — even the best sentence has no effect 40 minutes later. So practice should be run separately for each channel.

How to assess tone and empathy

A criterion like "was empathetic" is subjective, and two assessors will score it differently. A more reliable approach is to check the four behaviours separately: is each present or not, and if so, in which sentence. Both people and AI apply such a criterion more consistently, because the evidence — a concrete sentence — can be shown.

• Acknowledgement: did the agent restate the problem in their own words?
• Impact: did they name the customer's situation?
• Ownership: did they say "I'm doing it", or point to the system or another department?
• Action: did they give a concrete step and time?
• Breaking phrase: was any phrase from the list used?

The limits of assessment

Tone and empathy are the hardest skills to assess. In chat, irony, short replies and mixed languages can confuse AI; on a call, tone of voice is lost in the transcript. So in the first weeks, an empathy score should always be checked by a person on samples. It must never be the sole basis for an agent's pay or dismissal. The line between tone and text in audio analysis is covered in audio and transcript analysis.

An illustrative example

This is an illustrative example. An electronics retailer's support team starts empathy practice. In round one, most agents say "we understand your concern" and jump straight to the rule: "returns are within 14 days". The feedback shows that "impact" and "action" are almost entirely missing.

In round two, agents practise naming the customer's situation instead of a scripted line and ending the reply with a concrete step. A few weeks later the lead also reads real chats to see whether the change shows up in real work.

Common mistakes

• Turning empathy into a list of ready-made sentences.
• Empathy without action: a long apology, no step.
• Assessing chat and calls with the same criteria.
• A general criterion like "was empathetic".
• Accepting an AI empathy score without checking.

Limitations

In simulation, the AI customer does not reproduce every emotional reaction of a real person and sometimes calms down too quickly. Empathy also depends on the agent's fatigue and workload: the empathy of someone handling 150 unhappy customers a day is protected by the workload plan, not by training. Training does not replace that.

In Vexvon AI Training

In Vexvon AI Training, the unhappy-customer profile is built by the company itself — with behaviour, objections and notes. The company can set its own criteria, for example the four behaviours above as separate criteria, and assigns their weights itself. The evaluation shows an explanation for each criterion, a mistake tied to the agent's own sentence and a better sample answer. Practice can run in chat (Telegram) and in an AI test call. More on AI Training.

Next step

Open 10 chats with unhappy customers from last week and mark the four behaviours in each. The behaviour missing most often is the topic of your first practice. How de-escalation is assessed on real calls is covered in the de-escalation scorecard. More articles are in the customer service training section, and we can build the first practice together during a demo.