The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →AI can help route customer-service requests, but letting a system make or carry out multi-step decisions creates risks that a simple chatbot example cannot fully measure. The Air Canada chatbot dispute shows how a wrong policy answer can harm a customer and create legal exposure; it does not establish that an autonomous triage agent caused the error. The reported award was C$812.02, not an estimate of the airline’s total costs or a typical incident.
What happened when the Air Canada chatbot got a refund policy wrong?
In Moffatt v. Air Canada, customer Jake Moffatt asked the airline’s website chatbot about bereavement fares. It told him he could apply for a partial refund after booking and travelling. Air Canada’s policy required customers to arrange the bereavement option before travel. After Moffatt relied on the chatbot and requested a refund, the airline declined it. The British Columbia Civil Resolution Tribunal held Air Canada responsible for what its website chatbot communicated and ordered compensation. CBS News’ account of the award and a 2024 analysis in AI & SOCIETY describe the dispute and the conflict between the chatbot’s answer and the published policy.
As an Amazon Associate I earn from qualifying purchases.
The practical lesson is narrow but important: a customer-facing system can give incorrect policy guidance, and a business may face consequences when a customer relies on it. The case does not establish that an LLM, an autonomous agent, or a support-triage system produced the answer. The sources do not establish the chatbot’s architecture or autonomy level.
Recommended Free Tools
Is a support chatbot the same as an agentic triage system?
No. A chatbot may answer questions without taking action. An agentic system can plan or carry out multiple steps—for example, classify a request, retrieve account information, recommend an outcome, and trigger a workflow. Those capabilities can overlap, but the labels alone do not establish what a particular system can access or do.
#1 Best Overall
The UK Department for Science, Innovation and Technology identifies customer operations and service as bounded settings where businesses may deploy agentic AI. Its report describes agents handling service requests and processing refunds, while noting that consumer-facing authority remains limited and human escalation is common. It also flags robustness, coordination and real-world performance limitations. These are descriptions of a use case and its constraints, not evidence of a particular support system’s accuracy or safety. Read the UK government’s report on agentic AI and consumers, published 9 March 2026.
For support leaders, the key distinction is less about the product label than the authority granted to it. A system that suggests a queue is different from one that decides eligibility, promises a refund, changes an account, or initiates payment. Specify which of those tasks are permitted and which require a person.
How much did the Air Canada chatbot case cost?
CBS News reported an award of C$812.02 in the individual dispute in 2024. That is a case-specific customer award—not Air Canada’s total incident cost, a typical chatbot failure cost, or an estimate of agentic-AI losses.
The available sources do not quantify the airline’s staff time, legal expenses, reputational effects, follow-up contacts or other downstream costs. Nor do they provide a reliable aggregate rate for agentic support-triage errors or a representative estimate of what organizations typically lose when one occurs. The award should not be multiplied or generalized into a company-wide figure.
What controls help reduce risk in AI support triage?
Risk management is a lifecycle, not a single prompt, filter or human-review checkbox. NIST’s AI Risk Management Framework is voluntary guidance for incorporating trustworthiness into the design, development, use and evaluation of AI systems. Its materials support defining tasks and responsibilities, providing human oversight, evaluating systems, monitoring deployment and preparing to respond when something goes wrong. These practices can help teams manage risk; they do not guarantee that failures will not happen. NIST AI Risk Management Framework.
1. Set boundaries around tasks and authority
Document what the system may classify, recommend or execute. Treat policy-sensitive outcomes—such as refunds, eligibility decisions, exceptions, financial commitments and account access—as distinct permissions rather than assuming that permission to answer a question includes permission to act on it. Define when the system must stop and refer the case to a person, and assign responsibility for approving and changing those boundaries.
2. Make human oversight usable
A human reviewer needs both a clear handoff and the authority to intervene, correct an outcome or stop a workflow. Give customers a practical way to appeal an AI-influenced decision, and make it possible for staff to override it. Plan for cases where the system cannot recognize or correct its own error. NIST’s AI Risks and Trustworthiness guidance says risk management may need human intervention when an AI system cannot detect or correct errors. NIST AI Risks and Trustworthiness.
3. Test the support work the system will actually do
Evaluate the specific tasks, policies, tools and customer harms in scope—not just a general model benchmark. NIST’s ARIA program describes model testing, red-teaming and field testing as evaluation levels. Use test cases that reflect the real policy rules and workflows, including edge cases and situations where the correct action is to escalate. NIST Assessing Risks and Impacts of AI (ARIA).
Testing should continue after launch. Monitor live performance and customer feedback because a test result does not establish how the system will behave across changing policies, tools and real interactions. NIST’s AI RMF Core includes post-deployment monitoring among its risk-management outcomes. NIST AI RMF Core.
4. Track incidents and plan recovery
Keep records that let a team reconstruct what happened: the customer request, relevant policy version, system recommendation, human intervention, and any tool action. Track incorrect answers, escalations, overrides and complaints so teams can investigate patterns rather than treating each report as isolated. Establish ownership and procedures for incident response, recovery, system changes and decommissioning. These measures help teams detect and address failures; the cited guidance does not quantify how much any individual control reduces production error rates.
5. Include security and third-party components
Limit the system’s access and actions to what its assigned tasks require, and account for the risks of connected tools and third-party components. NIST’s 18 May 2026 summary of responses to its AI-agent-security request for information says commenters widely agreed that agents introduce novel security threats and that cybersecurity practices need adaptation. That document summarizes stakeholder submissions; it is not a measured incident rate or a support-specific finding. NIST’s summary of AI-agent-security RFI responses.
When should an AI support agent escalate to a person?
Set escalation rules around the system’s scope and the consequences of error. A workable policy should send a request to a person when it falls outside approved tasks, calls for an exception, involves an outcome the system is not authorized to decide, or cannot be resolved with the information and policy available. The human route must be able to change the outcome, not simply repeat the system’s answer.
Review escalation and override records alongside customer complaints and tool actions. If those signals show that the system is making unsupported policy claims or triggering unintended changes, pause the relevant action, investigate the workflow and restore service through a defined recovery process.
What the evidence can—and cannot—show
The Air Canada case documents a customer-facing policy-answer failure and a specific award. Government and NIST materials describe agentic AI’s potential use in customer operations and offer risk-management and evaluation practices. Together, they support treating scope, oversight, testing, monitoring, recovery and security as connected responsibilities.
They do not establish how often agentic support-triage systems fail, what a typical organization-wide failure costs, or that any one control eliminates errors. Those figures should not be inferred from a single customer dispute or from general AI risk guidance.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
