Key takeaways on detecting customer incivility in calls
- Incivility targets the person, not the situation. A dissatisfied customer, even a heated one, is exercising their right to complain. Insults, threats, discriminatory remarks and personal disparagement cross a different line: they are aimed at the advisor themselves.
- It remains largely invisible. The advisor moves on to the next call without always reporting it, and listening to a sample misses almost all of it.
- Identify: a criterion written in terms of observable remarks, set with no impact on the score, checks 100% of calls and quotes the passage concerned.
- Alert: an alert to handle, assigned to the team's manager, with a short delay, actions to take and an escalation if the delay is exceeded.
- Support: the advisor first, the follow-up with the customer next, then the causes, by cross-referencing incivility with call reasons.
- It is neither an emotion measurement nor a customer profile. It is a remark that was made, dated, within a conversation, and that can be verified by listening.
An angry customer is not an uncivil customer
A customer service team, and a complaints department even more so, deals with dissatisfied customers: that is what it is there for. The customer who says "this is unacceptable, it's the third time I've called about the same bill" is expressing disagreement about a situation. They may raise their voice, repeat themselves, ask for a manager. It is up to the advisor to handle the request, and that is a skill that can be learned.
Incivility is of a different nature. It is no longer about the case: it targets the person handling it. The INRS, France's national institute for occupational risk prevention, classes incivility, verbal aggression and threats as external violence, the violence an employee suffers in the workplace at the hands of people outside the company, customers and service users included.
This distinction is not a matter of careful wording. It determines what you look for in calls. Looking for a customer's "anger" or "aggressiveness" amounts to flagging a large share of complaints, and burying managers under alerts that say nothing about what the advisor went through. Looking for specific remarks, addressed to the advisor, brings up few calls, and the right ones.
Five forms can be spotted reliably in what is said:
| Form | Example remarks | What does not count |
|---|---|---|
| Rudeness aimed at the advisor | "You're taking the piss" | A swear word let slip about the situation: "this is a bloody shambles" |
| Insult | "You useless idiot", an abusive term aimed at the person | A criticism of the service: "your service is pathetic" |
| Threat against the person | "I know where to find you", "I'll be waiting for you outside" | Announcing a cancellation, a referral to the ombudsman or a complaint against the company |
| Discriminatory remark | A comment on the advisor's origin, accent, sex or age | No exception: the remark alone is enough |
| Personal disparagement | "You're too stupid to understand", repeated mockery | Asking to speak to a manager |
The last column matters as much as the other two. It is what separates a demanding customer from an uncivil one, and it is what will determine the quality of the detection.
Why incivility remains invisible
The advisor does not always report it. The next call is waiting, and reporting a difficult exchange takes time. Some consider that "it comes with the job", others fear being blamed for the way the call went. Unreported incivility exists for no one but the person who received it.
The sample misses it. Take an assumption: incivility in 0.3% of calls. A contact centre handling 20,000 calls a month then receives around sixty incidents. A quality review of 400 calls a month hears, on average, just over one. Listening to more changes almost nothing; only analysing every call brings them to light. The calculation is detailed in our article on analysing 100% of calls.
The usual indicators do not see it. Incivility moves neither average handling time nor the resolution rate. It can even lower the call's quality score: an exchange that turns sour prevents the advisor from going through the planned steps. With no other information, the manager sees a badly handled call where an advisor in fact came under attack.
1. Identify: a criterion that does not penalise the advisor
In Raisetalk, incivility is detected like any other fact in the conversation: through a criterion, an assertion that the analysis checks on every call. Three choices make the difference.
Write down what is said, and what does not count. The assertion follows the rule of effective criteria: who does what. The actor is the customer, the target is the advisor, and the assertion states explicitly what stays out of scope:
The customer directs at the advisor an insult, an obscenity or a discriminatory remark aimed at them personally. A complaint, even one made heatedly, criticism of the service or the company, and a request to speak to a manager do not count.
Separate threats. A threat against the person does not call for the same follow-up as an insult: it may justify a security measure or a report to the police. A second criterion gives it its own alert and its own process:
The customer threatens the advisor personally: to go after them, to track them down or to harm them. Announcing a cancellation, a referral to the ombudsman, a complaint against the company, a complaint to management or a negative review does not count.
Set them with no impact on the score. The customer's behaviour must cost the advisor nothing. Both criteria take the orientation "If the assertion is verified, the criterion is: Not acquired" and 0 points in both cases: the criterion page shows them as "no impact on the score". The verdict remains visible on the evaluation page, filterable in reports and able to trigger an alert, but the compliance score does not take it into account. The same library criterion can appear in several scorecards: add it to each of your active scorecards.
Each verdict comes with its justification and the quoted passage, timestamped: one click starts playback at the moment the remark was made. That is what makes the detection verifiable, and it is the basis of everything that follows.
Test on difficult calls before activating. Re-evaluate around fifty past calls with these criteria: a few known incidents of incivility and, above all, heated complaints that are not uncivil. The latter are the real test: if they trigger, tighten the assertion. The method is described in our article on testing a scorecard before rolling it out.
2. Alert: the manager informed, and in charge of acting
An information alert is not enough here: incivility calls for a human response, and someone has to be in charge of it. The rule is created in the Automations menu, in five steps.
The trigger. The Score creation event, that is every new analysis, with a Criterion (acquired / not acquired) condition: the incivility criterion is not acquired.
The recipients. A notification in Raisetalk for Manager (same team), plus an email if your managers live in their inbox more than in the application. Only people who can see the conversation are notified: the alert does not leave the team's scope.
The handling. Turn on Require handling and set it up as follows:
- Handler: the Manager (same team) role rather than a named person, so that the alert is picked up during holidays and absences;
- Delay: short, 4 h for example, counted in business hours if needed;
- If the delay is exceeded: notify Manager (parent teams), so that no alert is left pending.
Actions to take, copied into each alert, come down to a few steps that each start with a verb:
- Listen to the flagged passage (mandatory)
- Talk to the advisor about the call (mandatory)
- Offer a break or a change of queue
- Decide on the follow-up with the customer
For the threats criterion, a second rule can add a mandatory action, "Pass on to the security or legal department", and a shorter delay.
Closure. The handler closes the alert with a comment and an outcome: Founded alert, False positive or No further action. False positives measure the accuracy of the criterion: if they multiply, the assertion needs tightening. How alerts work is detailed in our article on quality alerts.
3. Support: what the manager does with the alert
The alert protects no one by itself. It guarantees that a difficult call does not go without follow-up, and it puts in the manager's hands what is needed to act: the conversation, the advisor, the time, the exact words and the recording to hear them.
The advisor first. The first step is to go and see them or call them, tell them the call has been heard and ask how they are. This is not an evaluation debrief: there is no question of a score, nor of what they could have done better. A break, a change of queue for the rest of the day or a referral to the contact the company has designated, such as occupational health or an employee assistance programme, may follow.
The customer next. The follow-up with them falls under the company's procedure: a callback from a manager, a letter reminding them of the rules of the exchange, a referral to mediation and, for a threat, reporting the incident and filing a complaint with the police. Raisetalk decides nothing at this stage. It provides the conversation number, the date and the quoted passage, frozen in the alert record at the moment the alert is raised.
A trail of what was done. Each ticked action records who carried it out and when, and the handling comment sums up the follow-up given. The company thus has a history of the incivility received and how it was handled, which feeds into its occupational risk assessment. The comment is read by the alert's recipients: it records what was done, not the advisor's state of health.
Later, skills development. A few days later, and if the advisor wishes, the call can serve as material for a debrief devoted to de-escalation techniques. The most useful role-plays start from real situations: with the incivility identified across all calls, the training team has genuine cases, selected and transcribed with the names removed, rather than invented scenarios.
Steering: where, when and for which call reasons
An alert deals with a call; steering deals with the causes. After a few weeks, the reports give the company a figure it did not have: the number of incidents of incivility its advisors receive, and how it changes over time. All it takes is to filter a report on the incivility criterion, result Not acquired, to keep only the conversations concerned.
- The trend: in the Team report filtered this way, the Conversation volume tile counts incidents of incivility day by day, week by week or month by month, compared with the previous period.
- The call reasons: the Attribute report, broken down by the call reason attribute, shows on which topics exchanges tip over. If incivility is concentrated on a delayed refund or a service outage, the cause lies upstream of the contact centre, and that is where it has to be dealt with. Cutting a processing time sometimes protects advisors better than a training course.
- The exposed teams: the same filter gives the number of conversations concerned for each team. This is a reading of exposure, not of performance: a debt collection team inevitably receives more than an information team. It helps to spread the workload, organise rotations on the most exposed queues and target training.
- Open questions: on the Understand page, Discovery queries the conversations filtered on these criteria, for example "What comes before the insults in these calls?" The answer comes with the extracts it is based on. The principle is described in our article on querying calls in natural language.
What detection does not do
It does not intervene during the call. The alert is raised when the call is analysed, after the exchange has ended, as calls come in with a direct connector. It does not cut the line and does not prompt the advisor. The instruction that allows them to end a call after a warning, where the company has provided for one, remains the first line of protection.
It does not measure emotion. Raisetalk seeks to know neither whether the customer is angry nor what state the advisor is in. It identifies remarks that were made, tied to a precise passage, which anyone can listen to again and contest.
It does not keep files on customers. Each alert concerns one conversation. Raisetalk does not keep a list of uncivil customers: the follow-up given to a customer falls under the company's procedure and its legal framework.
It does not replace prevention. Training, de-escalation techniques, the right to end the call, management support: detection makes these measures fairer, because they rest on what really happens on the phone.
Finally, a word on the legal framework. The employer takes the measures needed to ensure the safety and protect the physical and mental health of workers (article L4121-1 of the French Labour Code): identifying incivility and responding to it is part of that. Call analysis nevertheless remains a system that touches on employees' activity: they are informed beforehand (article L1222-4), and the comité social et économique, the French works council, is consulted before it is introduced (article L2312-38). Also tell your teams clearly that these criteria exist, what they pick up and that they do not count towards their score. An advisor who knows that an insult will be seen and followed up is no longer alone in knowing it.
Key terms
- Incivility: a customer remark aimed at the advisor as a person rather than at the situation being handled, such as an obscenity, an insult, a discriminatory remark or personal disparagement
- External violence: violence suffered by an employee in the workplace at the hands of a person outside the company, a customer or service user; the notion covers incivility, verbal aggression, threats and physical assault
- Criterion with no impact on the score: a criterion set to 0 points in both cases, whose verdict remains visible, filterable and able to trigger an alert
- Alert to handle: an alert assigned to a handler, with a delay, actions to take and an outcome chosen at closure
- Escalation: a notification sent to a higher level when the handling delay of an alert is exceeded
- False positive: an alert triggered wrongly, flagged as such at closure; their share measures the accuracy of the criterion
Frequently asked questions
How can customer incivility be detected in calls?
By analysing every call with a criterion that describes specific remarks addressed to the advisor (insult, obscenity, discriminatory remark, threat) and that explicitly excludes complaints, even heated ones. Each detection quotes the passage concerned, which can be played back in one click, and can trigger an alert assigned to the team's manager.
Should you detect the customer's anger or aggressiveness?
No. A customer who is angry about a situation is not uncivil, and a complaints department hears from such customers every day. A criterion based on anger would flag a large share of complaints and drown out genuine incivility. A remark that was made can be identified, verified by listening and discussed; an assumed emotional state cannot.
Does a customer's incivility lower the advisor's score?
Not if the criterion is set with no impact on the score, at 0 points in both cases. The verdict appears on the evaluation page and feeds alerts and reports, without counting towards the compliance score. The manager sees the incivility alongside the score, which avoids reading a call where the advisor came under attack as a badly handled call.
Who should receive the alert, and how quickly?
The advisor's line manager, designated by role so that the alert is still handled during absences, with a short delay of a few hours and an escalation to the level above if it is exceeded. An information alert is not enough: incivility calls for a human response that someone must be in charge of.
Can the analysis intervene during the call?
No. The alert is raised after the call, once it has been analysed. It replaces neither the instruction that allows the advisor to end an abusive exchange, nor management support in the moment. It guarantees that no call of this kind goes unnoticed, and that the advisor is not alone in knowing it.
Getting started
Incivility already happens in your calls. The question is who hears it, and what happens next. To see what a well-written criterion picks up in your own conversations, try Raisetalk for free.





