Pull up your team’s CSAT data and you’ll find a mix: good scores, bad scores, and some that are genuinely hard to read. The standard playbook is to focus on the low ones, identify which agent handled the ticket, and have a coaching conversation. It’s a reasonable instinct.
Except that in a lot of those conversations, the agent wasn’t actually the problem.
Bad CSAT scores come from all kinds of places:
- A customer who couldn’t get a refund because your policy doesn’t allow it.
- A product bug that caused the same frustration three times in a week.
- A process that required the customer to explain themselves to multiple people before anything got resolved.
In each of those cases, the score lands on the agent, even though they’re not the ones to blame.
Customer satisfaction is one of the most widely tracked metrics in support and one of the most frequently misused.
In this article, we’ll cover what CSAT actually measures, why treating all bad scores the same is a mistake, how to build an attribution framework that makes CSAT genuinely useful, and how to pair it with other metrics to get the full picture.
What Is Customer Satisfaction (CSAT) and How Is It Measured?
Customer Satisfaction Score (CSAT) is a metric that captures how satisfied a customer was with a specific interaction or experience, typically expressed as a percentage of positive responses out of total responses received.
It’s one of the most common customer service metrics out there, and its most common format is a post-interaction survey. After a support interaction, a prompt asks the customer to rate their experience as good or bad.
This might be a binary thumbs up/down, or it might mean a rating on a 1-3 or 1–5 scale. Most industry CSAT benchmarks sit between 75% and 85% for support teams, with scores above 80% generally considered strong.
How and when you measure CSAT matters more than most teams realize. There are two broad approaches:
- Per-interaction CSAT: A rating is requested after every individual response or touchpoint. This gives you a granular view of how sentiment shifts within a conversation — not just where it ended up.
- End-of-conversation CSAT: A single rating is collected when the ticket closes. This captures the overall outcome, but loses the journey. A customer who started frustrated and ended satisfied looks identical to one who was happy throughout.
The per-interaction model gives you more data to work with, but also more responsibility to interpret it correctly. Either way, the number on its own tells you very little. What matters is understanding why it landed where it did.
When CSAT Scores Don’t Tell the Whole Story
Here’s a scenario most support leaders have faced:
A customer reaches out trying to get a refund. Your refund policy is restrictive — that’s a business decision, not a support one. The agent handles the conversation well. They’re empathetic, clear, and professional. But the customer doesn’t get their refund, and they rate the interaction poorly.
Was that a bad support experience? Not really. Was it a bad customer experience? Absolutely.
This is the core tension with agent-level CSAT.
The metric captures how satisfied a customer was with an outcome and outcomes are frequently shaped by factors completely outside the agent’s control. A rigid policy. A product bug. A fulfillment delay. A complicated escalation process that requires the customer to fight to get to the person who can actually help.
The honest truth is that a skilled agent can still influence customer satisfaction, even when the product or policy is working against them. Empathy, tone, transparency, and how well they navigate a difficult answer all matter.
No product or policy is perfect, that’s precisely why customer support exists. How your team handles those imperfect moments is a real measure of service quality, and agents should own that.
But there’s a limit to this.
You can handle a conversation exceptionally well within a set of constraints and still have a customer leave dissatisfied because the constraints themselves are the problem.
The agent’s job in those moments isn’t to manufacture satisfaction, it’s to make the experience as good as it can possibly be given what’s available.
Those are two different things, and conflating them is where the CSAT unfairness creeps in.
The People, Process, Policy Framework
Before you act on any bad CSAT score, there is one question worth forcing yourself to answer first: whose problem is this, really?
A simple attribution framework helps here. When you see a dissatisfied rating, assign it to one of three buckets before doing anything else:
- People: The agent’s handling of the conversation was the primary issue — communication, empathy, accuracy, follow-through. This is where coaching is warranted.
- Process: The customer had to work too hard. Too many steps, too many handoffs, too many back-and-forths to get to a resolution. This is an operational problem, not an individual one.
- Policy: The outcome the customer wanted wasn’t available to give them. Refund terms, SLA commitments, eligibility criteria — business decisions that no amount of excellent service can fully overcome.
Once you’ve attributed the score, you know who needs to act. People issues go to managers for coaching. Process issues go to support operations. Policy issues go to the people who own those policies.
The failure mode most teams fall into is skipping this step entirely. (This is exactly the gap we’re building ResolveLoop to close — more on that further down.)
They see a bad score and defaulting to an agent conversation regardless of root cause. That doesn’t just create an unfair culture. It actively misdirects your improvement efforts toward the wrong part of the business.
Should CSAT Be an Individual Agent KPI?
This is the debate the support community keeps returning to. The honest answer is: it depends on how you use it.
If CSAT is treated as a key performance indicator (KPI) — a hard target that agents are held accountable for hitting — it becomes a blunt instrument.
Individual scores are volatile, heavily influenced by factors outside the agent’s control, and vary significantly depending on ticket type, customer segment, and channel.
An agent who handles primarily billing complaints will always have a harder road than one managing onboarding queries.
Measuring them against the same CSAT threshold isn’t fair performance management. It’s noise.
There’s also a growing complication that doesn’t get talked about enough: the AI-to-human handoff.
More support journeys now begin with an automated layer — a bot, an AI deflection attempt, a self-service article that didn’t resolve the issue.
By the time that customer reaches a human, they may have already spent several frustrating minutes trying to get past a chatbot. They arrive carrying that frustration, and it shows up in the agent’s CSAT score.
Research suggests 72% of consumers have abandoned a purchase after a poor automated service experience — the emotional residue of that frustration doesn’t disappear when they finally reach a person.
What CSAT genuinely is useful for as an individual metric is consistency analysis.
Is this agent consistently low across a range of ticket types, including ones where peers are scoring well? That’s a coaching signal worth exploring. Is this agent only low on ticket types where everyone on the team scores low? That’s a process or policy signal — escalate it, don’t coach it.
Agents should absolutely be held accountable for their CSAT scores, but only for the parts of it they actually control.
That means how they handle the conversation, how they communicate bad news, and how they advocate for the customer within the boundaries they’ve been given.
What they shouldn’t be held accountable for is the outcome of a policy they didn’t write or a product issue they can’t fix.
This cuts both ways.
If a product issue is consistently driving low scores, your support team has a responsibility to do more than absorb those ratings. They should be quantifying the pattern and feeding it back to the product team: how many conversations, what segment of customers, what specific friction point?
If the signals are there and they’re not being raised, that’s a gap in accountability too. CSAT, used well, becomes part of the product feedback loop, not just a reflection of individual agent performance.
Factor this in to who should follow up on negative scores: the first review of CSAT responses should be manager-led rather than agent-driven.
A manager who has already run the attribution step — People, Process, or Policy — can have a far more grounded and useful conversation than an agent reviewing their own dissatisfied ratings cold.
Use CSAT as a discovery tool. Use it to ask better questions, not to deliver verdicts.
Pair CSAT with IQS to Get the Full Picture
Customer satisfaction tells you what the customer felt. It doesn’t tell you whether the agent actually did their job well. That’s where Internal Quality Score (IQS) becomes essential.
IQS is an internal evaluation of the interaction against your own quality rubric — did the agent follow the right process, use the right tone, resolve accurately, escalate when needed?
It’s assessed by a reviewer, manager or AI system, not by the customer.
Critically, a well-functioning quality framework runs independently of CSAT. Coaching on how to handle difficult conversations, communicate policy clearly, or navigate a frustrated customer doesn’t wait for a bad score — it happens continuously as part of how your team develops.
CSAT flags patterns from the customer’s point-of-view. IQS flags them against your internal standards. Your quality assurance process should address both.
The combination of CSAT and IQS gives you four diagnostic quadrants worth understanding:
- High CSAT + High IQS: The interaction went well and was handled well. Celebrate it and look for what’s replicable.
- High CSAT + Low IQS: The customer was satisfied but the quality of execution was below standard. Worth addressing before it becomes a habit.
- Low CSAT + High IQS: The agent did everything right. The problem is elsewhere — product, policy, or process. This is precisely the pattern the attribution framework is designed to catch.
- Low CSAT + Low IQS: Both the experience and the execution fell short. This is where individual coaching conversations are genuinely warranted.
In my experience at the Academy to Innovate HR (AIHR), we measure CSAT on every individual interaction — not just at ticket close — which gives us visibility into how sentiment shifts throughout a conversation.
We started specifically analyzing mixed conversations: tickets where the customer rated poorly early and then improved, or rated well and then dropped off. Those variance patterns were far more informative than CSAT scores alone.
In one analysis, we noticed a cluster of conversations that started well but deteriorated at a particular point in the resolution process. When we looked closely, we found that a specific policy we were enforcing was creating friction. It wasn’t the agents handling it.
We didn’t open a coaching conversation. We opened a policy review. That’s what CSAT is supposed to do.
Using Zendesk Apps to Track Attribution at Scale
The People/Process/Policy framework is straightforward to apply to a handful of tickets. At volume, you need your tooling to carry some of the load. If you use Zendesk, Zendesk apps can be a great help for this:
- Ticket Classification automatically tags incoming tickets by intent — a product question, a billing issue, a policy request, a complaint. That tagging doesn’t just help with routing. It provides a layer of context when you go back and analyse your CSAT data. If every low-rated ticket in a given period is tagged as a refund request, you have your attribution before you’ve read a single conversation.
- Ticket Parser Autofill extracts and populates structured data from ticket content automatically, keeping your custom fields accurate and your reporting clean. When you’re segmenting CSAT across ticket types, clean field data is the difference between useful analysis and noise.
- If your CSAT scores are actually low because of how agents are handling them — particularly on sensitive or policy-adjacent conversations where tone and accuracy matter most — Agent Co-writer helps agents draft clearer, more accurate responses in the moment.
- The three apps above help you tag, structure, and respond better, but they still leave the attribution step to a human reviewer. That’s the gap we’re building ResolveLoop to close — a post-resolution intelligence layer that runs the People / Process / Policy attribution on every closed ticket automatically, so by the time a manager sits down to review CSAT, the whose problem is this question is already answered with evidence from the conversation itself. It’s currently in pilot with a handful of Zendesk teams; if that’s a problem you recognise, come talk to us.
The goal isn’t to automate away accountability. It’s to give your team the context and support they need so that when you do look at CSAT, you’re working with the right signal.
CSAT Is a Signal, Not a Verdict
Customer satisfaction matters. It’s worth measuring, tracking, and acting on. But it is an outcome shaped by your product, your policies, your processes, and your people — and treating it as a pure measure of individual agent performance misses most of what it’s actually telling you.
The customer service teams that get the most out of CSAT are the ones who use it to ask better questions. Is this consistent? Is this pattern pointing somewhere systemic? It’s the agent, or it’s the policy? What can we actually change, and who owns that change?
Before you act on a bad score, run it through the People, Process, Policy lens. Pair it with IQS and a continuous quality framework that coaches independent of ratings. Look for patterns across conversations rather than reacting to individual scores.
Hold your agents accountable for how they handle difficult conversations — and expect them to surface product and process issues quantitatively when those issues are the real problem.
CSAT is too reactive and too blunt to carry the full weight of quality accountability on its own.
Use CSAT as one signal in a bigger picture, and it will tell you a lot. Use it in isolation, and it will mostly mislead you.
If you’re curious on how Zendesk apps like Swifteq’s might be able to help you improve your customer service and your team’s capabilities, you can schedule a demo and we’ll show you how our apps best fit your workflows. You can also check out all of our apps here — many of which are free or include 14-day free trials (no credit card required).
Written by Neal Travis
Curious learner and builder of customer experiences in scale-ups. Neal is the Head of Customer Experience at the Academy to Innovate HR (AIHR) and Host of Growth Support.
