Updated
What is AI customer service?
AI customer service is support where an AI agent resolves customer conversations first, from sources you approve, and hands anything that needs judgment to a person in the same thread. Done right, customers get an answer in seconds and your team gets its day back. Done wrong, it is a chatbot that loops.
This guide is the map. It covers what an AI agent should and should not do, how to roll one out without burning customers, and how the pricing models compare. Shopping for the agent itself? Start with AI customer service agents. Shopping for the whole inbox? Read AI helpdesk software.
The basics
What it is. What it is not.
The words "AI customer service" cover everything from a decision-tree bot to an agent that issues a refund. The difference shows up on the customer's side of the chat.
| AI customer service is | AI customer service is not |
|---|---|
| An AI agent that resolves the conversation | A menu of buttons that sends people to an article |
| Answers from sources you approve, with a citation | A model guessing from the open internet |
| Hands over to a person in the same thread | A dead end that makes the customer start again |
| Part of the inbox your team already works | A second tool with its own setup and its own bill |
| Measured on resolved conversations | Measured on how many people it kept away from you |
Want the mechanics of how an AI finds and grounds an answer? See how AI answers customer questions.
The split
Kal takes the repeats. People take the calls.
The best line between AI and human is not a percentage. It is a list. Write down who owns each kind of conversation before you turn anything on.
| Conversation | Who resolves it | Why |
|---|---|---|
| Where is my order? How do I reset my password? | Kal | Repeat questions with a documented answer. |
| Refund inside policy, plan change, address fix | Kal, with guardrails | Actions your rules allow, confirmed by the system that ran them. |
| Refund outside policy, custom terms, a legal threat | A person | Judgment calls. Kal hands over with a summary. |
| An angry customer, a churn risk, a VIP | A person | Relationships. Kal drafts, a human decides. |
| A question no source covers | A person | Kal does not guess. The gap goes back to your sources. |
Actions like refunds and plan changes are where an AI agent stops answering and starts resolving. See how Kal runs them end to end on Auto-resolve.
Rollout
Roll it out. Don't roll the dice.
Four steps, in order. Skip one and customers find out before you do. The long version is how to roll out an AI support agent safely. For ten concrete automations to start with, see AI customer service automation strategies.
- 1
Pick the sources.
Connect the help articles, docs and past answers Kal may read. The Knowledge Store is that list of sources. If a source is missing, Kal does not make one up.
- 2
Draw the lines.
Set guardrails in plain words: never quote prices, never mention competitors, never promise SLAs. Decide which actions Kal may run and which always go to a person.
- 3
Wire the handover.
Choose who gets what. When Kal stops, the person sees the whole thread and a summary of what Kal tried. The customer never repeats themselves.
- 4
Watch, then widen.
Start with your most common questions. Read the replies, fix the sources Kal tripped on, then hand it the next set. Every reply is logged with its citations.
Pricing
Pay for answers. Not for attempts.
AI customer service software is usually priced one of three ways. Know which one you are buying before you compare numbers.
| Model | You pay for | Watch for |
|---|---|---|
| Per seat | Each person on the team | AI sold as a separate add-on on top |
| Per resolution or outcome | Each conversation the AI closes | What counts as closed, and any monthly minimum |
| Both | Seats for people, usage for AI | Whether handovers or timeouts are billed |
SupportWire One is both, kept small: $19 per seat per month, the first 50 AI resolutions each month included, then $0.49 each. Lite seats are unlimited and free. A handover is never a resolution, so it is never billed. For heavy volume, Scale prices per resolution with a minimum commitment. Intercom's Fin lists $0.99 per outcome on its own pricing page.
Full plan details on pricing. Comparing against Intercom? Run your own numbers in the Intercom pricing calculator.
The market
AI powered customer service platforms, by name.
Most helpdesks now sell an AI agent, and a few products are built around one. We will not rank them here or put words in their mouths. These are the names you will meet, each with a side-by-side we wrote:
Want a ranked list with prices? See AI customer support software, ranked, or browse every comparison.
Our take
Small team. Big punch.
We built SupportWire so a team of a few people can run support like a team many times its size. Kal answers first from your Knowledge Store and cites every source. Handover passes the rest to a person with a summary on top. Wire Through keeps open conversations intact while you switch from your old tool. Your team stays small. Your support does not.
FAQ
AI and customer service, answered.
Updated November 2026