You reduce DSAT by treating dissatisfaction as a set of repeating causes rather than a run of unlucky tickets. The method has five steps: find every dissatisfied conversation, categorize the drivers instead of reading anecdotes one at a time, separate the causes you control (process, knowledge, tone, handling) from the ones you do not (price, policy, product), fix the systemic controllable ones at the root, and re-measure the same category to confirm the fix moved it. The step teams most often skip is the first, because a survey only captures the customers who chose to answer, and most dissatisfaction never produces a survey response at all.
In short
- A DSAT score records that a customer was unhappy. It does not record why, so the score alone is not something you can act on.
- Survey response rates are typically low, so the dissatisfied responses you can see are a self-selected slice of the dissatisfaction that actually happened.
- Reading negative tickets one by one produces anecdotes. Categorizing them produces a ranked list of causes, which is what you can actually fix.
- Separate causes inside your control (process, knowledge, tone, handling) from causes outside it (price, policy, product), because the second group is a routing job, not a coaching job.
- Most controllable dissatisfaction is systemic rather than individual: a confusing macro, a missing knowledge base article, a broken handoff between queues.
- Verification matters more than volume. A fix is only proven when the specific driver category you targeted goes down, not when the overall number happens to move.
- Full conversation coverage is the precondition for this method, not the headline. It is what makes the categorization representative instead of a sample of whoever complained.
Step 1: Find the dissatisfied conversations, including the ones with no survey
Most DSAT reduction programs start by pulling the negative survey responses. That is the right instinct and an incomplete data set, for a reason worth stating plainly: a survey only measures customers who chose to answer it. Response rates on post-interaction surveys are typically low, and the customers who reply are not a random sample. Someone with a strong reaction is more likely to respond than someone who was quietly let down and simply left.
So there are two populations to find, and they need different methods.
The dissatisfaction that has a rating
Pull every response in your dissatisfied range for a meaningful period, along with the full transcript for each one. The rating is the index, not the evidence. Anything you conclude has to come from the conversation attached to it, which means the export has to include the conversation.
The dissatisfaction that never got a survey
This is the larger group, and you find it by looking for the behavioral fingerprints of an unhappy customer rather than a rating. The reliable signals are all in the conversation record: negative sentiment in the customer’s own language, an explicit complaint or escalation request, a reopened ticket, multiple contacts about the same issue, a transfer chain, or a conversation that ended without a resolution being confirmed. Each of these is a customer who was dissatisfied and never told you on a form.
Combining the two gives you a working set that reflects your operation rather than your survey respondents. This is the point at which scoring every conversation stops being a coverage statistic and starts being a practical requirement: if you only review a sample, your categorization in step 2 inherits whatever bias picked the sample.
Step 2: Categorize the drivers instead of reading anecdotes
The failure mode here is universal. A manager reads twenty angry tickets, remembers the three most vivid, and builds a plan around them. Twenty tickets read individually produce twenty stories. The same twenty tickets tagged against a fixed set of driver categories produce a ranked list, and a ranked list is the only thing you can prioritize against.
Build the category list before you start reading, so you are classifying rather than narrating. Keep it short enough that categories are unambiguous, usually eight to twelve. Then tag every conversation in your working set, allow more than one tag where a conversation genuinely had two causes, and count.
What you are looking for is concentration. Dissatisfaction is almost never spread evenly. A flat-looking rate across the operation usually resolves into one or two categories carrying most of the volume, and those categories usually cluster further into one queue, one channel, one issue type, or one workflow. That concentration is the finding. Everything after this step is acting on it.
Step 3: Separate what you control from what you do not
Once the drivers are counted, split them into two piles. This is the step that stops a DSAT program from becoming demoralizing, because roughly half of customer dissatisfaction in most support operations is caused by decisions the support team did not make.
| Driver category | Example | Within support’s control? | Where the fix lives |
|---|---|---|---|
| Process | Handoffs between queues, repeated verification, long transfer chains | Yes | Workflow and routing design |
| Knowledge | Wrong or outdated answer given, agent could not find the policy | Yes | Knowledge base and enablement |
| Tone and communication | Curt reply, no acknowledgment, jargon, unclear next step | Yes | Coaching and QA scorecard |
| Handling and ownership | Ticket left idle, promised follow-up never happened | Yes | Coaching and workload management |
| Expectation setting | Timeline given that the team could not meet | Yes | Macros, templates and training |
| Policy | Refund window, warranty terms, eligibility rules | No, but route it | Policy owner, with evidence attached |
| Price | Cost of the plan, a fee the customer disputes | No, but route it | Commercial and pricing owners |
| Product | Bug, missing capability, confusing interface | No, but route it | Product and engineering backlog |
Step 4: Fix the systemic causes, not the individual tickets
The controllable pile is now yours to work. The temptation is to convert it into individual feedback: forward each bad conversation to the agent who handled it and ask them to do better. That produces defensiveness and very little movement, because most controllable dissatisfaction is not an individual failure. It is a system that made the failure likely.
Fix the system first
If eleven agents all gave the same wrong answer, the problem is the knowledge base article, not eleven agents. If negative ratings cluster on tickets that were transferred twice, the problem is the routing rule. If a macro sets an expectation the team cannot meet, the problem is the macro. Ask of every category: what would have to be true for this to stop happening regardless of who picked up the ticket? Then change that.
Coach the pattern, not the incident
Where the cause genuinely is individual handling, coach on the repeated pattern rather than the single worst conversation. One bad ticket is a bad day and coaching on it reads as punishment. The same behavior across fifteen conversations is a habit, and it is coachable precisely because the evidence is undeniable. Ground every coaching conversation in the specific lines from the transcripts, so the discussion is about what was said rather than about whether the score was fair. The mechanics of that handoff are covered in turning QA data into coaching.
Route the uncontrollable pile with evidence, not opinion
The policy, price and product categories are not a dead end. They are your strongest input into other teams, provided you send a count and a set of quoted conversations rather than a complaint. “Fourteen percent of our dissatisfied contacts last quarter were the refund window, here are thirty transcripts” is a business case. “Customers hate the refund policy” is not. This is often the highest-value output of a DSAT analysis, because those drivers are usually invisible to the teams that own them.
Step 5: Measure whether the fix actually moved it
Track the category you fixed, not just the headline number. Overall DSAT moves for many reasons at once, including seasonality, volume mix and changes in who responds to your survey, so an overall improvement after a fix is suggestive rather than evidence. The claim you want to be able to make is narrower and much stronger: the driver you targeted fell as a share of dissatisfied conversations, and it fell in the queue where you made the change.
Watch the leading indicators, not just the survey
Survey results lag, and they only ever cover respondents. The conversation-level signals from step 1 move sooner and cover everything: the rate of negative sentiment, escalation requests, reopened tickets and repeat contacts in the affected queue. If those are falling and the survey has not caught up yet, the fix is working. If those are flat, the fix did not land, whatever the monthly percentage does. Read them alongside the operational metrics that tend to move with dissatisfaction, particularly first contact resolution and first response time.
Re-run the categorization on a schedule
Driver mix changes. A category you fixed six months ago can quietly return through a product change or a new macro, and a new category can appear that was not in your original list. Re-running the tagging quarterly keeps the list honest and turns DSAT reduction into an operating rhythm rather than a project with an end date.
Why the survey is not enough on its own
Everything above depends on one thing: being able to see what happened inside the conversations, at enough scale that the categorization is representative. That is exactly what a survey cannot give you. It returns a number from the customers who chose to answer, and it returns no reasoning with it.
Reviewing conversations manually solves the reasoning problem but not the scale one. A team that reads two percent of conversations gets a genuine explanation of two percent of the operation, and the two percent is chosen by whoever pulled the sample. Systemic drivers that appear in a small fraction of contacts, which is what most of them are, are easy to miss entirely at that sample size.
Scoring conversations automatically closes both gaps. It applies the same rubric to every conversation rather than to a sample, which is what makes the driver counts in step 2 trustworthy, and it surfaces dissatisfaction in conversations that never generated a survey response. Coverage is the precondition for the method, not a claim about seeing all. Kaizo does this against your own scorecard and traces every score back to the specific evidence in the transcript, so a driver category is a defensible count with conversations behind it rather than an assertion. Because it runs natively in Zendesk and Salesforce, the analysis happens where the conversations already are. At UiPath, Kaizo automated 100% of QA with 200% ROI and an 8% lift in quality score.
The definitions underpinning all of this are in what is DSAT, and the reason the number is not simply the inverse of your satisfaction score is worked through in CSAT vs DSAT.
Common mistakes when trying to reduce customer dissatisfaction
- Treating the survey population as the dissatisfied population: response rates are typically low and respondents self-select, so the ratings you can see under-represent the problem and skew toward the loudest cases.
- Reading tickets instead of counting categories: anecdotes feel like insight and produce a plan built around whichever conversation was most memorable rather than most common.
- Coaching individuals for systemic failures: if many agents made the same mistake, the process or the knowledge base caused it, and coaching them individually will not stop it recurring.
- Ignoring the uncontrollable drivers: policy, price and product dissatisfaction is real dissatisfaction. Not owning the fix is not a reason to leave it uncounted and unrouted.
- Declaring victory on the headline number: overall DSAT moves for reasons unrelated to your fix. Prove the specific driver category fell in the specific queue you changed.
- Targeting the score rather than the cause: pressure to improve a dissatisfaction metric without addressing its drivers tends to change survey timing and survey wording rather than the customer experience.
Frequently asked questions
How do you reduce DSAT in a call center?
Work causes, not tickets. Collect every dissatisfied conversation including the ones with no survey response, tag them against a fixed set of driver categories, split those drivers into what support controls and what it does not, fix the controllable ones at the system level rather than coaching individuals for systemic failures, route the rest to the teams that own them with transcript evidence attached, then re-measure the specific category you targeted.
What causes customer dissatisfaction most often?
In support operations the recurring drivers are process friction such as transfers and repeated verification, knowledge failures where the answer given was wrong or outdated, tone and communication problems, poor expectation setting, and ownership gaps where a promised follow-up did not happen. Alongside those sit price, policy and product drivers, which are real causes of dissatisfaction that the support team cannot fix and should route with evidence instead.
Why does a DSAT survey not tell you why a customer was unhappy?
Because it captures a verdict, not a reason. A rating of 1 could mean the answer was wrong, the wait was long, the policy was unwelcome, or the tone was curt, and each requires a completely different fix. Even a free-text comment gives you the customer’s summary rather than the sequence of events. The reasoning is in the transcript of the conversation the rating is attached to.
How do you find dissatisfied customers who never answered the survey?
Look for the behavioral signals in the conversation record rather than for a rating: negative sentiment in the customer’s own words, an explicit complaint or escalation request, a reopened ticket, repeat contacts about the same issue, long transfer chains, and conversations that closed without a confirmed resolution. Because response rates are typically low, this group is larger than the one that rated you.
How long does it take to move a dissatisfaction score?
Longer than the fix itself, because survey data lags and only reflects respondents. Watch the conversation-level indicators instead, since negative sentiment, escalations, reopens and repeat contacts in the affected queue move first. If those fall and the survey has not caught up, the change is working. If they are flat, the fix did not land regardless of what the headline percentage does.
Should you coach agents on every negative rating?
No. Most controllable dissatisfaction is systemic, so reviewing every negative rating with the agent who handled it creates defensiveness without changing the underlying cause. Coach on repeated patterns across many conversations, grounded in the specific lines from the transcripts, and fix the process, macro or knowledge gap when the same mistake appears across multiple agents.
Related terms
Find the dissatisfaction your survey never saw
Bring a week of your real conversations and we will show you the dissatisfaction drivers ranked by volume, including the unhappy customers who never answered a survey, with every score traced back to the exact evidence in the transcript.
