Nearly every QA rubric in Customer Support has a criterion version of it: “agent demonstrated empathy”. Sometimes it gets a point scale. Sometimes a checkbox. Occasionally a grading guide that says something like “score 3 if the agent showed genuine care for the customer’s situation”, as if genuine is something that means the same every single time.
As an industry, we tend to mostly leave that criterion as is because the outputs look fine: the customer says thank you, scores positive on CSAT, and everyone concludes the rubric worked.
But that is only part of the story. It worked on the surface and quietly ignored the thing underneath.
The problem starts with the word and the category it fits in. Empathy is the kind of word we use to describe a personality trait, and trait words do exactly one useful thing in a managerial setting: they name a quality you would like someone to have.
They don’t tell you how to build it, how to score it without arguing about it, or what you are actually looking at when it’s missing. Also, they hand you a very specific false comfort: the feeling of having measured something, when in reality we’ve just given it a nicer name.
What A Trait Word Does to Your Scorecard?
Put “empathy” on a scorecard and three things happen, none of them particularly helpful.
The first is that it stakes a claim about someone’s interior state instead of their behaviour. Interior states are invisible. You cannot see empathy.
You can see specific communicative acts: the choices of language, the timing, whether the agent actually responded to what the customer said. Those acts are either there or they aren’t.
Empathy sits upstream of all of it, unobservable, inferred from cues that shift from one reviewer to the next. That’s how “genuine care” ends up with alignment reliability somewhere in the region of a coin flip.
The second is that it quietly moves accountability somewhere unhelpful. If empathy is a personality trait, the agent who lacks it is just a person without the trait, and the only fix is to hire differently. You can’t coach a trait, but you can correct a behaviour, change an approach, or build a skill.
You cannot make someone more empathetic by sending feedback that says “more empathy here”.
All that does is make them anxious and unsure, with no idea what to actually do about it. The framing decides what happens next.
The third is that it keeps the work invisible, which is the state the work has always been in. Our industry still doesn’t have a clean way to talk about the skills that make this job genuinely hard.
The sociologist Andrew Abbott called this tacit knowledge: the kind of work that is practically indispensable to the role that is never properly defined, so it never gets the appropriate recognition.
“Empathy” on a rubric is a placeholder. Look at what a skilled agent is actually required to do in a difficult interaction:
- Read the customer’s emotional state accurately, including spotting the difference between someone who is venting and someone about to lose it.
- Adjust tone in real time to match what the moment needs.
- Giving a specific acknowledgement instead of a scripted one.
- Making the cooperative goal of the conversation attractive enough that the customer re-engages.
That is cognitively demanding, sophisticated, and professional work. “Agent demonstrated empathy: 3/5” sees none of it.
What’s Actually Going On When We Say “Empathy”?
Pull the word apart in a Support context and what you find is a set of relational competencies, deployed deliberately. Actually feeling the feelings is optional; the competencies are the job. Here is what your rubric is trying to define when it includes “empathy” as a qualitative score.
Face acknowledgement. Face, in linguistic terms, is a person’s sense of their own social worth and autonomy.
Positive face is the need to feel valued.
Negative face is the need to feel free from imposition (specifically: imposing on someone else or to be imposed upon).
Both are on the table the second a customer contacts Support.
Something they relied on broke. That failure has already dinged their positive face; they are now the person whose thing stopped working, and the act of asking for help threatens their negative face, because asking means conceding they couldn’t fix it themselves.
The agent’s first job is to show they have recognised that specific damage. Not “I’m sorry to hear you’re having trouble”. That line might be accurate, but it’s communicatively empty. It tells the customer they’ve been filed under Customers Having Trouble, which is roughly the opposite of feeling seen.
Good acknowledgement is specific. It speaks to the actual situation, shows the agent was paying attention to what went wrong, and registers why it matters.
Emotional calibration.
This is the part we usually collapse into “empathy”, which is exactly how it disappears. Lisa Feldman Barrett’s work on emotion is useful here. Her work tells us that emotions aren’t simply expressed, they are constructed.
The brain predicts what’s happening based on prior experience, context, and whatever information is available.
So reading another person’s emotional state accurately has very little to do with instinct, or with being a warm person by nature – otherwise it would indeed be a personality trait.
It comes down to what Barrett calls emotional granularity: how many words you have for feelings, and how precisely you use them.
An agent with high emotional granularity can tell the difference between a customer who is frustrated and a customer who is about to flood. Those are two different states, and they need two different responses.
A frustrated customer is addressable: match the urgency, acknowledge the inconvenience, move towards a fix.
A customer whose emotional state is flooding (nervous-system overwhelm) isn’t addressable, because they are in a physiological state where productive conversation is temporarily off the table.
Treat flooding as ordinary frustration and respond with proportionate warmth, and you’ve made a calibration error. A specific, nameable one with consequences we can point to and measure. Which means it can go on a rubric with concrete criteria, and it can be coached.
Cooperative re-entry.
After the face-threat and the acknowledgement, the conversation still has some work to do.
It has to move out of repair mode and into the actual fixing things part, where the agent solves the problem. That transition doesn’t happen on its own.
It takes communicative work: making the shared goal visible, signaling that the agent is working for the customer’s outcome, and creating enough relational safety that the customer is willing to get back to the practical task instead of staying in their frustration.
In low-tractability interactions, where repair moves slow or not at all, full re-entry might not be possible at all.
The agent’s job then is to get as close as the conditions allow without the customer experiencing the shortfall as a second failure. That too is a skill, and it has distinguishable levels of execution.
What the Feedback Looks Like
The vague version:
“Try to show more empathy when the customer is upset. The response felt a bit cold.”
Accurate from the reviewer’s perspective, useless for the person receiving the feedback. The agent already knows the customer was upset. What they don’t know is what they specifically failed to do.
They can’t fix “a bit cold”, because cold isn’t a behaviour. They’ll likely add a couple more sorry-to-hear-that sentences in future interactions, without changing anything structural.
The specific version:
“The customer opened with “I’ve been trying to reach you for two days and nobody has helped me”. That’s a face-threat with a history attached. Your first line acknowledged the technical issue but not the experience of being left without help.
Try something that reflects the particular frustration of repeated failed contact before you move into the fix. Compare lines 3 and 11: once you had more context, you calibrated to the customer well. I’d like to see if you can get there sooner next time.”
Same agent. Same interaction. Completely different feedback.
The second version names the competency that slipped, anchors it in the transcript, points at what a corrected version looks like, and credits what the agent got right elsewhere.
A manager can track it over time. A reviewer can take the same lens to the next hundred interactions and land on consistent scores.
Why This Matters Beyond the Rubric
The rubric argument is a small one compared to the larger, more structural one.
Customer Support has spent decades treating its core skills as personality traits, and the consequences run a lot deeper than QA. When a skill is treated as a personality trait, it can’t be developed, only spotted.
Hiring becomes the only lever you have to acquire the skill within your team.
A team with the wrong traits can’t be built up, it has to be swapped out, and the answer to “how do we get better at this?” becomes “hire different people“. That answer is usually unattainable, never scalable, and not even true.
There is a third cost. When skill reads as a trait, it stops registering as professional knowledge at all. Abbott’s theory of how professions earn recognition within their field is straightforward: by turning what they do into formal knowledge other people can learn.
Look at the Law profession, for example: it rests on more than arguing well. Medicine relies on more than being good with sick people. Both have bodies of formal knowledge that turn what practitioners do into something teachable, learnable, and assessable.
CX has never done this for its relational competencies.
Instead, it calls them ‘soft skills’: a label that says nothing about how difficult those skills are to do and acquire, and everything about how little we have valued them. It has left practitioners holding genuinely demanding knowledge with no formal recognition attached.
A QA rubric that names face acknowledgement, emotional calibration, and cooperative re-entry is a better scoring instrument. It is also a statement: these competencies are real, specific, and professionally meaningful.
The senior agent who built high emotional granularity over years of practice acquired something. They did not simply turn out to be “good with people”.
The Payoff
Done correctly, this still looks like empathy. Customers experience a conversation where they feel specifically heard rather than generically acknowledged when the agent’s warmth and urgency match the actual emotional register of their situation and the shift into problem-solving feels like a natural next step.
The difference shows up most clearly at scale. When “empathy” is the definition, every agent is working from their own private interpretation of what good looks like. Some of those interpretations are excellent. Some are not.
And when you are hiring fast, onboarding quickly, and running a team across multiple shifts and time zones, that variance compounds.
Customers get wildly inconsistent experiences depending on who picks up the conversation, and without a shared language to diagnose why it becomes a costly endeavour to figure out the root cause.
When the competencies are defined specifically, that changes. Face acknowledgment, emotional calibration, cooperative re-entry – these are things you can define once, coach consistently, and calibrate across a team.
A new agent and a senior agent can look at the same interaction and use the same framework to evaluate it. Quality stops being a vibe and starts being something you can actually maintain as the team grows.
That is what customers feel: not just that one agent was kind. That the whole team is reliably, consistently good at this, and that when something goes wrong, the team knows how to fix it.
Naming the skill is one thing. Catching it across every ticket is another.
The hard part isn’t agreeing that face acknowledgement, emotional calibration, and cooperative re-entry are real skills. It’s applying that same lens consistently, across thousands of conversations, without it collapsing back into “felt a bit cold.”
That is where ResolveLoop comes in. It reads your closed Zendesk tickets and evaluates each one against the competencies that actually make support hard. Where the agent acknowledged the technical issue but missed the experience underneath it. Where the tone matched ordinary frustration while the customer was already past it. Where the conversation never made it out of repair mode and into the fix. The vague trait becomes a specific, coachable observation, anchored in the transcript, ready for feedback the agent can act on.
And when the breakdown wasn’t the agent at all, it says so. Sometimes the agent did everything the rubric asks and the customer still left unhappy, because the real cause was the product, the policy, or the process around the conversation. ResolveLoop names that too, so you coach what is coachable and fix what isn’t.
Quality stops being a vibe. It becomes something you can see, one ticket at a time.
Written by Ines van Dijk
Ines van Dijk is the founder of Customer Support Excellence and the author of The Customer Support QA Playbook. Her research on the Conversational Integrity Model is indexed on SSRN.
