On this page
Short answer
Can AI replace customer service representatives? Not as a whole job, based on the public evidence as of November 2026. AI agents in the support products cited here answer routine questions, cancel or change orders, update addresses and process refunds, within rules you set. Zendesk and Intercom document handing unresolved requests or requests for a person to a human. Gorgias documents the specific actions described below; its cited source does not establish its fallback behavior. Following up on fixes, taking payments, judging insurance coverage and handling returned goods have no cited agent doing the work.
Two pieces of public data frame this. O*NET classifies the role as customer service representatives (code 43-4051.00): people who answer routine questions about products and services and may handle general complaints. BLS reports a median pay of $44,770 a year, or $21.53 an hour, in May 2025, across about 2,666,000 jobs. BLS projects employment to decline 5% from 2025 to 2035 and expects about 289,500 openings a year from people leaving the job. BLS attributes the decline to self-service systems, social media and mobile apps that let customers do simple tasks without a representative, and expects companies to keep in-house staff for complex inquiries "such as refunding accounts or confirming insurance coverage". That is BLS's projection for the occupation, not a claim about any single business.
Task by task
The table starts from the 13 task statements O*NET lists for the role, merged into 11 tasks a support manager would recognize. Keeping records of each contact is folded into the handoff task, and completing contract forms and discontinuance orders into the order-change task. A task is marked for an agent only where a cited source documents an agent doing it. Where O*NET lists the task but no cited source shows an agent doing it, the task is marked human, and the note says so.
Agent
0of 11
Agent with review
6of 11
- Answer customer questions about products, services and accounts
- Take, cancel or change orders, subscriptions and shipping addresses
- Resolve service and billing complaints with refunds, credits or replacements
- Look into complaints and disputed charges to find the cause
- Pass unresolved issues to the right team with the case details
- Suggest additional products or services
Human
5of 11
- Check that the fix worked and tell customers the outcome of a claim or adjustment
- Work out charges, collect deposits or payments and arrange billing
- Review insurance policy terms to decide whether a loss is covered
- Inspect disputed goods against orders and invoices and prepare return invoices
- Recommend improvements to products, shipping, service or billing
11 tasks: 0 agent, 6 agent with review, 5 human
| Task | Who does it | Why |
|---|---|---|
| Answer customer questions about products, services and accounts | Agent with review | Zendesk's advanced email AI agents and Intercom's Fin answer from your connected knowledge sources and hand the conversation to a person when they cannot resolve it or the customer asks; a person owns every escalation. |
| Take, cancel or change orders, subscriptions and shipping addresses | Agent with review | Gorgias documents actions that cancel orders and update shipping addresses after the customer confirms, Zendesk's agent runs procedures such as "Cancel my order", and Intercom says Fin can cancel subscriptions once connected to a billing system. |
| Resolve service and billing complaints with refunds, credits or replacements | Agent with review | Intercom says Fin processes full or partial refunds and applies credits through data connectors, and Gorgias lists a reship-for-free action; you set the rules and the handoff for cases outside them. |
| Look into complaints and disputed charges to find the cause | Agent with review | Intercom says Fin works out why a charge appears twice and fixes it when connected to billing; this is a vendor marketing claim, and no other cited source documents agent investigations. |
| Pass unresolved issues to the right team with the case details | Agent with review | Zendesk says its agent escalates with full context and its voice agents send a transcript, summary and detected intent; a person takes over each escalated case. |
| Suggest additional products or services | Agent with review | Intercom's billing documentation counts Fin's product recommendations from a synced Shopify catalog as resolutions; this page recommends sampling those conversations. |
| Check that the fix worked and tell customers the outcome of a claim or adjustment | Human | This is an O*NET task; no cited source documents an agent following up after the fact. |
| Work out charges, collect deposits or payments and arrange billing | Human | This is an O*NET task; Zendesk lists billing and payments among suitable use cases for its voice agents but documents no payment-collection step, and no cited source shows an agent collecting payments or setting up billing. |
| Review insurance policy terms to decide whether a loss is covered | Human | This is an O*NET task; no cited source documents an agent making coverage decisions, and BLS names confirming insurance coverage as a complex inquiry companies keep in-house. |
| Inspect disputed goods against orders and invoices and prepare return invoices | Human | This is an O*NET task; no cited source documents an agent comparing returned goods with the original order. |
| Recommend improvements to products, shipping, service or billing | Human | This is an O*NET task; no cited source documents an agent proposing process changes to the business. |
On that standard, 0 of 11 tasks are ones a documented agent completes with only spot checks, 6 of 11 are ones a documented agent does with a person handling every exception, and 5 of 11 stay with a person. None is marked agent alone because each one is customer-facing, and the Zendesk and Intercom documentation cited here rests on a person taking over what the agent hands off.
The evidence comes from product documentation and vendor marketing from three support software companies.
- Zendesk. Zendesk's help center announced on April 20, 2026 that advanced email AI agents can answer one or more questions from connected knowledge sources, run procedures such as "Cancel my order", ask clarifying questions and escalate to a human agent "when the customer requests it, when a procedure specifies it, or when the AI agent cannot confidently resolve the request in one response". Zendesk says its voice AI agents send escalated calls to a person with a transcript, summary and detected intent.
- Intercom Fin. Intercom's billing documentation charges $0.99 per resolution, defined as the customer confirming the answer helped or leaving without asking for more help, and does not bill escalations a customer asks for. Intercom's marketing page says Fin processes refunds, cancels subscriptions and updates account details through data connectors to systems such as Shopify and Stripe, and reports an average 76% resolution rate. That rate is Intercom's own figure, measured on its own definition.
- Gorgias. Gorgias's blog describes AI Agent Actions that cancel orders, update shipping addresses, pause or cancel subscriptions and reship lost orders for free in connected ecommerce apps. It says actions that change customer or order data require the customer's confirmation, and that when several actions match one ticket the agent performs none.
No cited source reports an independent accuracy or resolution rate. A resolution in Intercom's definition can mean the customer simply left, so an agent's resolution count is not the same as satisfied customers. Zendesk's own best practices for voice agents include periodically listening to call recordings or reviewing transcripts. Our recommendation: read a weekly sample of conversations the agent closed, not only the ones it escalated.
A worked example
This is an illustration, not a reported case. Assume, for this example, that you run an online store with an AI agent connected to your order system, with cancellation allowed for unshipped orders and refunds over $100 sent to a person. A customer emails at 11 p.m. asking to cancel an order placed an hour earlier and asking whether a product they bought last month is covered by its warranty after a fall. The agent checks the first order, confirms it has not shipped, asks the customer to confirm and cancels it. It cannot decide the warranty question under its rules, so it hands the case to the queue with the order details and the customer's description. Your rep reads it in the morning, asks for a photo and decides. The agent did the routine change overnight. The person made the judgment call.
Telling customers they are talking to an AI
Disclosure rules vary by jurisdiction. In the EU, Article 50(1) of the AI Act requires providers of AI systems intended to interact directly with people to design them so the people concerned are informed they are interacting with an AI system, unless that is obvious to a reasonably well-informed, observant and circumspect person given the circumstances. Article 50(5) says this information must be given clearly and at the latest at the first interaction. Ask counsel which obligations apply to your role and jurisdiction.
Jobs you can hand to an agent today
The agent-with-review tasks above group into separate jobs, each with its own page covering what to give the agent, a brief you can copy, risks and how to run a trial.
- JobSupervisedAI support agent for tier-1 customer support
An AI support agent can answer routine chat and email questions from your help content and hand the rest to your team. A person still owns refunds, policy exceptions and every escalation.
- JobSupervisedAI phone agent for customer service: scope, cost, checks
An AI phone agent can answer routine inbound calls, such as FAQs, bookings and status questions, and route the rest. A person still owns escalations, account changes that move money and the call scripts it follows.
- JobSupervisedAI ticket triage: what an agent sorts, routes and misses
An AI agent can read each new ticket, tag its topic, language and urgency, and route it to the right queue. A person still owns the categories, the tickets it cannot classify and anything flagged as a safety, legal or security issue.
- JobSupervisedAI agent for knowledge base upkeep: scope, checks, cost
An AI agent can find questions your help center does not answer, draft new articles and propose edits to stale, duplicate or conflicting ones. A person still decides what is true and approves every change before it goes live.
Ready
0of 4
Supervised
4of 4
Assist only
0of 4
Not yet
0of 4
- Tier-1 support covers answering routine questions and making order and account changes in chat and email.
- Phone support covers the same work on inbound calls, where recording and disclosure rules add their own steps.
- Ticket triage covers classifying incoming requests and routing them to the right person with the details attached.
- Knowledge-base maintenance covers keeping the help content current, which the Zendesk and Intercom agents cited here answer from.
If you are deciding where to start, start with tier-1 support on email or chat for one request type, such as cancelling unshipped orders. Zendesk and Gorgias both document order cancellation, and you can check each outcome against your order system.
What stays human
Five things stay with a person in the sources cited here.
Every escalated case. Zendesk and Intercom document human escalation; the Gorgias source cited here does not describe fallback behavior. Zendesk escalates when the customer asks or the agent is not confident; Intercom does not bill escalations the customer asks for. Our recommendation: staff that queue so escalated customers get a reply on the same day.
Coverage and judgment calls. O*NET lists reviewing insurance policy terms to decide whether a loss is covered. BLS names confirming insurance coverage and refunding accounts as complex inquiries companies keep in-house. No cited source shows an agent making those decisions.
Payments and billing setup. Working out charges, collecting deposits and arranging billing are O*NET tasks with no cited agent doing them. The refunds the products above issue run inside rules you configure.
Follow-ups and returned goods. Checking that a fix worked, telling customers the result of a claim and comparing returned goods with the original order have no cited agent evidence.
Improving the service. Spotting a recurring problem and recommending a change to products, shipping or billing stays with your team.
A setup the cited documentation supports is an agent handling routine questions and rule-based changes, with representatives owning escalations, judgment calls, payments and follow-ups. If you want to hire for that when the job board opens, sign up below.
Sources
- 43-4051.00 - Customer Service Representatives
O*NET OnLine (U.S. Department of Labor)PrimaryAccessed Nov 14, 2026
- Customer Service Representatives: Occupational Outlook Handbook
U.S. Bureau of Labor StatisticsPrimaryAccessed Nov 14, 2026
- Announcing agentic AI for advanced email AI agents
Zendesk helpPrimaryAccessed Nov 14, 2026
- About voice AI agents
Zendesk helpPrimaryAccessed Nov 14, 2026
- Fin AI Agent outcomes
Intercom HelpPrimaryAccessed Nov 14, 2026
- AI Agent That Takes Action: Refunds & Cancellations
IntercomVendorAccessed Nov 14, 2026
- 7 Customer Requests AI Agent Actions Resolve End to End
GorgiasVendorAccessed Nov 14, 2026
- Regulation (EU) 2024/1689 (Artificial Intelligence Act)
EUR-Lex (Publications Office of the European Union)PrimaryAccessed Nov 14, 2026
- Article 50: Transparency obligations for providers and deployers of certain AI systems
European Commission AI Act Service DeskPrimaryAccessed Nov 14, 2026
What we could not verify
- Independent resolution or accuracy rates for AI customer service agents. The only figures found are vendor-reported (Intercom's 76% average resolution rate for Fin and its Kajabi figures), and Intercom counts a conversation as resolved when the customer leaves without asking for more help.
- Whether these products reduce the number of representatives a given business needs. BLS projects a decline for the occupation as a whole and names self-service and automated systems as the reason; no cited source quantifies the effect of AI agents specifically.
- Klarna's widely reported 2024 claim that its assistant did the work of 700 agents, and later reports that it resumed hiring people. The original press release is older than 24 months and the follow-up reports were not confirmed from a primary source, so neither is used.
- Production evidence of agents following up to confirm a fix worked, collecting payments, judging insurance coverage, inspecting returned goods or proposing process improvements. Those tasks are marked human for that reason, not because an agent provably cannot do them.
- Which AI disclosure rules apply outside the EU AI Act. US state laws were not checked for this page.
Related
- AI agents for customer service: what to hand over first
- AI support agent for tier-1 customer support
- AI phone agent for customer service: scope, cost, checks
- AI ticket triage: what an agent sorts, routes and misses
- AI agent for knowledge base upkeep: scope, checks, cost
- How to hire an AI agent: scope, brief, trial, supervision