The employee can see what went wrong.
The automated system rejected a reasonable request, the customer has already repeated the story three times, and the only button that could fix the problem is greyed out. The employee writes an incident note, watches the handling-time clock continue to run, and decides not to say what everyone involved already knows: the system caused this.
That decision is not always cowardice. Sometimes it is a rational response to the price of honesty.
The Quiet Behaviour Behind “Good Service”
When speaking plainly becomes risky, people rarely announce that they are self-censoring. Their behaviour simply changes.
A customer-service employee avoids describing a broken process in direct language and writes that the interaction was “challenging.” A manager records an individual performance issue rather than admitting that the employee had no authority to approve the only workable solution. A platform worker saves screenshots of an unfair rating but does not appeal because the process takes unpaid time and may lead nowhere.
The official record becomes cleaner as the underlying reality becomes messier.
This is visible across several countries and service systems. The evidence does not support a simple claim that customers everywhere are becoming more aggressive. It supports something more specific: frontline workers are repeatedly placed between frustrated people and systems they do not control.
British retail estimates cited by the UK government recorded roughly 590,000 incidents of violence and abuse in 2024–25. That was about 20 percent lower than the previous year, but still approximately 1,600 incidents a day. Both facts matter. The problem is extensive, and the strongest current estimate does not show an uninterrupted rise.
The distribution is also uneven. A British survey covering 1,669 retail employees found reported abuse on roughly 20 to 27 percent of working days in shops, customer service and call-centre roles. In back-office functions, the figure was below 1.5 percent.
The organisation may describe itself as one workplace. Its employees are not living in the same workplace.
Why People Stop Saying What Happened
The first reason is practical risk.
When a rating can affect shifts, bonuses, access to tasks or continued employment, the customer is no longer providing only feedback. The rating becomes a hidden management decision.
Human Rights Watch documented platform workers across several countries being governed through ratings, acceptance rates, cancellations and automated decisions, often with limited explanation or meaningful appeal. A 2026 Chinese study involving 209 gig workers found that perceived unfair negative ratings were associated with greater job insecurity and lower trust in the fairness of the rating system.
The study does not prove that ratings alone caused the insecurity. It does show why a worker might tolerate more, document privately and challenge less.
The second reason is unclear authority.
Employees are frequently held responsible for the customer experience while lacking the power to change the rule, reverse the automated decision, remove an unfair score or end an abusive contact safely. They must calm the customer, protect the company’s reputation and keep the interaction short. They may need approval from someone who is unavailable.
Responsibility without authority creates learned silence. The employee stops reporting the contradiction because the contradiction has become the job.
The third reason is normalisation.
The Institute of Customer Service reported in January 2026 that 42 percent of surveyed frontline workers had experienced hostile customer behaviour within the previous six months. Among those exposed to verbal hostility, 22 percent had also faced physical threats. More than a quarter had taken time off, and 35 percent had considered leaving their job.
A separate UK–US survey published in July found that 39 percent of frontline respondents believed customer abuse was treated as “part of the job.” The survey was funded by a company selling body-camera technology, so it should not be treated as a neutral prevalence measure. Its description of workplace culture still fits the broader evidence: people report less when they expect little to happen.
A reporting procedure is not protection if the employee expects the incident to remain on the record while the rating remains on the employee.
When Automation Leaves Humans With the Conflict
Automation can remove repetitive work. It can also remove the easy work.
A large randomised field experiment in Taobao customer service found that an AI system could reduce average conversation time. It also found weaker customer ratings in some AI-suitable chats and different results depending on why a human intervened. Human involvement worked better in technical escalations than in emotionally escalated cases, particularly when the intervention came late.
This does not prove that AI makes customer service worse. Nubank has reported substantial improvements in customer satisfaction and self-service completion in selected production uses of generative AI.
The difference is design.
An effective system solves the problem before frustration grows. A badly designed system filters out simple contacts, sends the unresolved exceptions to a human and then measures that human against colleagues who received easier cases.
The employee becomes responsible for the final five minutes of a failure that began twenty minutes earlier.
That changes how honestly people speak inside organisations. They stop writing, “The automation failed and the customer had no usable exit.” They write, “The customer resisted the available solution.” The first description questions the system. The second places the problem safely inside the interaction.
That is how narrative capture happens without a conspiracy. People learn which description produces consequences and which one passes quietly through the organisation.
The Strongest Difference Is Not Between Men and Women
The clearest documented divide in this research is between frontline and back-office work.
Gender still affects the type of pressure people can face. ILO pilot surveys in Burkina Faso, Côte d’Ivoire and Senegal found higher reported levels of some forms of psychological and physical work-related violence among men, while women more often reported sexual violence and harassment. The surveys covered broader workplace violence and harassment, not customer abuse alone, so they cannot support a universal gender ranking.
They do show why equal rules do not always produce equal experiences.
Men in retail, transport and security may be expected to take over physical enforcement, confront a suspected thief or absorb an escalating customer because they are assumed to be able to handle it. The practical response may be to say less, treat the situation as a task and document the event only after the danger has passed.
Women in customer-facing work may face sexualised comments, gendered degradation or pressure to remain warm and accommodating while setting a boundary. A direct refusal can be recoded as poor service. The practical response may be indirect language, delayed reporting or checking with colleagues before naming what happened.
Neither response proves that men are naturally emotionally closed or that women are naturally indirect. Behaviour changes when the expected penalty changes.
The shared mechanism is self-protection. The form varies because physical risk, sexualised abuse, employment status, authority and access to support are not distributed evenly.
The Uncomfortable Truth
Many organisations do not merely have difficult customers. They have system failures that customers and employees have been trained to fight each other over.
The customer cannot reach the rule-maker, software owner or executive who approved the staffing model. The customer reaches the person wearing the headset.
The employee did not create the delay, the app error or the inflexible policy. Yet that employee carries the conversation time, the emotional labour, the complaint and sometimes the rating.
Customer anger does not excuse threats, discrimination or violence. But calling every conflict a customer-behaviour problem allows the organisation to disappear from its own service chain.
The polished narrative says that the employee owns the customer experience. The operational reality is that the employee often owns only its consequences.
The Positive Truth
This pattern is not inevitable.
Tokyo has developed consultation, training and expert support around customer-harassment prevention. Japan is preparing a national employer duty scheduled to take effect on 1 October 2026. South Korean labour guidance already describes employer obligations to protect customer-facing workers and allows measures such as temporary work suspension or reassignment when harm occurs or is likely.
Australia and the United Kingdom are developing more formal consequences for repeated or serious attacks on retail workers. The ILO’s 2026 platform-work convention moves international standards toward greater transparency, human review and protection against unjustified deactivation, although it will require national ratification and implementation before becoming an enforceable right.
These measures are incomplete. Laws can exist without useful enforcement. Reporting can create unpaid paperwork. Barring systems can be misused. A manual can tell an employee to stop a conversation while the rating system quietly punishes the stop.
Still, they demonstrate a different principle: safety, escalation and appeal can be designed as organisational responsibilities rather than tests of personal toughness.
How to Make Honesty Cheaper
A useful safeguard begins by separating three things that organisations often collapse into one category:
What failed?
Name the original service, policy or technical problem.
What did the person do?
Separate legitimate complaint, disagreement, personal degradation, discrimination, threat and physical danger.
What could the employee actually change?
Record the available authority, requested backup and any blocked solution.
This can fit into a short incident note or a manager conversation. It does not require a ten-step framework.
The point is to prevent a system failure from becoming a character judgment. It also prevents customer frustration from being used to excuse abuse.
The same rule should apply to ratings. A customer score can be useful evidence, but it should not become an automatic sanction when the employee did not control the price, delay, software failure or policy. Serious consequences require context and human review.
Documentation should also count as work. Reports, rating appeals, safety notes and follow-up calls are not personal housekeeping. When organisations require them, organisations should pay for the time.
Finally, employees need to know the boundary before the contact begins: when they may stop, who takes over, what happens to the rating, whether their pay or shifts are protected and how a system error will be recorded.
“Use your judgement” is not a safety procedure. It is responsibility transferred without support.
Silence can make a workplace look calm. It can protect a customer score, shorten a meeting and keep an uncomfortable fact out of the report. But the unresolved problem remains, and the people closest to it learn that accuracy is more dangerous than adaptation.
What would the people handling the problem say if telling the truth could no longer cost them their safety, income or standing?
SOURCES
Source 1
Source: “Crime and Policing Act 2026: retail crime factsheet” — UK Home Office and Ministry of Justice
What the source found:
The factsheet describes the new offence concerning assault of retail workers and cites an estimated 590,000 incidents of violence and abuse in British retail during 2024–25, down from approximately 737,000 the previous year.
How it appears in the article:
It supports the argument that retail abuse remains extensive while correcting the unsupported narrative of an uninterrupted increase.
Everyday impact:
Retail workers may face repeated verbal or physical incidents while also spending time reporting, documenting and managing known repeat offenders.
Limitations:
The statistics combine material from government, industry and union sources using different definitions and sampling methods. The estimates must not be added together or treated as a harmonised national count.
Source 2
Source: “Customer abuse in retail: What the latest survey tells us and what employers should do now” — Foot Anstey, discussing Retail Trust data
What the source found:
A survey of 1,669 British retail employees reported abuse on approximately 20–27 percent of working days in frontline, customer-service and call-centre roles, compared with below 1.5 percent in back-office functions.
How it appears in the article:
It supports the article’s central distinction between employees who meet customers directly and colleagues who work inside the same organisation without comparable exposure.
Everyday impact:
Workers in high-contact roles may spend a meaningful part of the working week anticipating, absorbing or recovering from conflict.
Limitations:
The source is a secondary legal analysis of survey data. Some subgroups were small, and exposure measured by working days cannot be compared directly with annual incident totals.
Source 3
Source: “An all-encompassing issue — tackling the abuse of public-facing workers” — Institute of Customer Service
Link:
https://www.instituteofcustomerservice.com/all-encompassing-issue-abuse/
What the source found:
The organisation reported that 42 percent of surveyed frontline workers had experienced hostile customer behaviour within six months. The findings also linked exposure to physical threats, absence and consideration of leaving the job.
How it appears in the article:
It supports the discussion of normalisation, under-reporting, absence and employee withdrawal.
Everyday impact:
Workers may lower their threshold for what they report, take time away from work or quietly look for another job.
Limitations:
The full methodology is not presented on the public summary page, and the organisation has an institutional interest in stronger protection for service workers.
Source 4
Source: “For frontline staff, abuse by customers is now ‘just part of the job’” — HR Executive, reporting a HALOS-funded survey
Link:
https://hrexecutive.com/for-frontline-staff-abuse-by-customers-is-now-just-part-of-the-job/
What the source found:
A survey of 2,500 frontline workers in the United Kingdom and United States found substantial normalisation of customer abuse, job-leaving considerations and absence after incidents.
How it appears in the article:
It supports the description of a workplace culture in which employees may believe that reporting will not lead to meaningful protection.
Everyday impact:
Workers may remain silent, shorten interactions or avoid particular shifts rather than entering a reporting process they do not trust.
Limitations:
The survey was commissioned by a company that sells body-camera technology. It is used as a current cultural signal, not as an independent prevalence estimate.
Source 5
Source: “Tackling work-related violence and harassment: what data from three pilot surveys in West Africa reveal” — ILOSTAT / International Labour Organization
What the source found:
Pilot surveys in Burkina Faso, Côte d’Ivoire and Senegal documented significant work-related violence and harassment. Customers, patients, students and relatives were among the identified perpetrators. The data also showed differences in the types of harm reported by men and women.
How it appears in the article:
It supports the gender section and the claim that customer- or user-related harm is not restricted to Western retail systems.
Everyday impact:
Workers may avoid certain people, take absence, leave work or adapt their communication to reduce physical, psychological or sexual risk.
Limitations:
The surveys cover broader workplace violence and harassment, not customer abuse alone. Results from three countries cannot be generalised to Africa as a whole or directly compared with British retail statistics.
Source 6
Source: “Algorithms of Exploitation: Rights Abuses in the Gig Economy and the Global Fight” — Human Rights Watch
What the source found:
The investigation describes how ratings, acceptance rates, cancellations and automated systems can influence access to work and benefits while offering workers limited information or effective appeal.
How it appears in the article:
It supports the argument that a customer rating can become an indirect management decision rather than simple feedback.
Everyday impact:
Workers may accept undesirable tasks, tolerate abusive behaviour or spend unpaid time collecting screenshots and trying to challenge an automated decision.
Limitations:
The research is qualitative and problem-focused. It documents mechanisms and cases but cannot establish how common each practice is across all platforms.
Source 7
Source: “How and when gig workers’ unfairly negative customer ratings affect knowledge sharing” — Aslib Journal of Information Management / Emerald
What the source found:
The survey of 209 Chinese gig workers found associations between perceived unfair negative ratings, job insecurity, lower trust in rating-system fairness and reduced knowledge sharing.
How it appears in the article:
It supports the discussion of ratings, self-protection and private documentation while carefully avoiding a claim of proven causation.
Everyday impact:
A worker who distrusts the system may become less willing to challenge a customer, share problems with colleagues or rely on formal appeal.
Limitations:
The study is cross-sectional, based on self-reports and limited to a relatively small Chinese sample. It cannot prove that ratings caused the reported insecurity.
Source 8
Source: “Agentic AI in customer service: randomized field evidence from Taobao” — research preprint
Link:
https://arxiv.org/abs/2605.14830
What the source found:
The randomised field study examined conversation time, ratings, escalation and human intervention. It found that the effects differed between technical and emotional escalation and that late human intervention was less effective in some emotionally escalated cases.
How it appears in the article:
It supports the argument that automation can leave humans with a concentrated share of complex or emotionally difficult contacts.
Everyday impact:
A service worker may receive fewer total conversations but a higher proportion of cases in which the customer is already frustrated and the available solution is limited.
Limitations:
The version used was a preprint, covered one production system and cannot represent every form of AI-assisted customer service.
Source 9
Source: “Generative AI for customer support at Nubank” — Nubank researchers
Link:
https://arxiv.org/abs/2606.08867
What the source found:
The company reported improved customer satisfaction and self-service completion in selected generative-AI implementations, including strong results in an A/B-tested card-delivery use case.
How it appears in the article:
It provides the counterweight to a predetermined anti-AI argument and supports the claim that outcomes depend on design, measurement and escalation.
Everyday impact:
When a system genuinely solves routine problems, customers repeat themselves less and employees can avoid unnecessary explanatory work.
Limitations:
The study was produced by the company evaluating its own selected implementations and focused more on service outcomes than on the complete employee workload.
Source 10
Source: “Customer Harassment Prevention Promotion Project” — Tokyo Metropolitan Government
Link:
https://www.customer-harassment-taisaku.metro.tokyo.lg.jp/en/
What the source found:
Tokyo’s programme provides consultation, seminars and expert assistance to help organisations develop practical responses to customer harassment.
How it appears in the article:
It supports the positive section showing that customer abuse can be treated as an organisational prevention responsibility rather than a personal resilience test.
Everyday impact:
Workers may gain clearer procedures, earlier management involvement and a more legitimate basis for ending harmful contact.
Limitations:
The programme description does not establish whether incidents, absence or employee turnover have fallen.
Source 11
Source: “2025 Labor Laws in Korea” — KOTRA / Invest Korea
Link:
https://www.investkorea.org/file/ik-en/252025Labor_Laws_in_Korea.pdf
What the source found:
The guidance describes employer obligations to protect customer-facing employees from health damage caused by customer abuse and identifies temporary work suspension or reassignment as possible protective measures.
How it appears in the article:
It supports the claim that some legal systems recognise a practical right to remove a worker from a harmful interaction.
Everyday impact:
An employee may have a formal basis for stopping work or being reassigned rather than remaining in contact until the customer chooses to stop.
Limitations:
The source describes the legal framework, not how consistently the rules are enforced in individual workplaces.
Source 12
Source: “Decent Work in the Platform Economy Convention, 2026 (No. 193)” — International Labour Organization
What the source found:
The convention establishes an international standard concerning platform work, including algorithmic management, access to information, human review and protection against unjustified deactivation.
How it appears in the article:
It supports the positive direction toward ratings and automated decisions being subject to explanation and human review.
Everyday impact:
Platform workers may eventually gain stronger national rights to understand and challenge decisions that affect their income.
Limitations:
The convention does not automatically change national law. It requires ratification and domestic implementation.
Comments are welcome, but this is not a ragebait space. Claims need evidence. Disagreement is allowed. Dehumanization, personal attacks and narrative-protection will not carry the discussion.