Every support team hits the same fork: keep the scripted customer service chatbot that answers the easy questions, or move to an AI agent that resolves tickets end to end. The two get sold under one label, which is how buyers end up owning the wrong one. This guide explains what each really does, where the line between them sits, what both cost, and how to roll either out without hurting the numbers you are paid to protect.
What does a customer service chatbot actually do?
A customer service chatbot is software that talks to customers in a chat window and answers routine questions without a human. Most run on scripted decision trees, keyword matching, or retrieval from your help center. They excel at high-volume, low-complexity questions, and they respond instantly at any hour, which is why most support teams start there.
Used inside that scope, a chatbot earns its keep on four jobs:
- Instant answers to documented questions. Shipping policies, store hours, plan features, how-to steps: anything your help center already explains, served in seconds instead of a queue.
- Collecting details before handoff. Order numbers, account emails, screenshots, error messages, so a human picks up a briefed conversation instead of starting from "how can we help".
- Triage and routing. Classifying the request and sending it to the right queue, team, or priority level without a human reading it first.
- After-hours coverage. Acknowledging the customer, setting expectations, and capturing the issue when nobody is online to respond.
One note on names, because search treats them as different products. The same software gets sold as a customer service chatbot, chatbot customer care, a virtual assistant, or a support bot. The label does not matter; the mechanism does. If it follows predefined flows and cannot act on your systems, it is a chatbot, whatever the pricing page calls it.
Where scripted chatbots stop being enough
Scripted chatbots stop being enough the moment a conversation needs judgment, account context, or an action in another system. A script can recite your refund policy; it cannot check the order, apply the policy, and issue the refund. That gap between answering and resolving is where deflection turns into customer frustration.
The pattern is predictable enough that you can watch for it in your own metrics:
- Rising "talk to a human" requests. Customers learn the bot's limits fast and start skipping it, which means it is adding a step instead of removing one.
- Repeat contacts. The same customer returns on another channel within days because the bot's answer did not resolve anything.
- Escalation loops. The bot asks for details, fails to act on them, then transfers to a human who asks for the same details again.
- Script sprawl. Every product change spawns new flows, and maintaining the decision tree quietly becomes somebody's part-time job.
- A CSAT split. Satisfaction on bot-handled conversations trails human-handled ones by a visible margin.
None of this is an argument against chatbots. Inside a narrow, documented question set they are cheap, fast, and reliable, and they never improvise a refund policy. The failure is a scoping failure: asking scripted software to do resolution work. If your queue is mostly "where is my order", you are fine. If it is mostly "this charge is wrong, fix it", you have outgrown the category.
What does an AI customer service agent do differently?
An AI customer service agent resolves tickets end to end instead of deflecting them. It reads the conversation and the account, decides what needs to happen, takes the action in your systems, and confirms the outcome. Gartner projects that agentic AI will autonomously resolve 80 percent of common customer service issues by 2029 (Gartner, 2025).
The practical difference shows up right after the customer explains the problem. A chatbot searches for a matching script or article and posts it. An agent checks the order, sees the double charge, applies your refund policy, issues the credit, and writes back with what it did. One conversation answered a question; the other closed a ticket.
| Dimension | Customer service chatbot | AI customer service agent |
|---|---|---|
| How it answers | Follows scripts or retrieves help-center articles | Reasons over the request and decides the steps |
| What it can do | Answer documented questions, collect details, route | Check orders, change accounts, issue refunds, close tickets |
| Context it uses | The current chat | The chat plus account, order, and ticket history |
| When it fails | Off-script questions dead-end into an escalation | Low-confidence cases escalate with context attached |
| Metric it moves | Deflection and first-response time | Resolution rate and time to resolution |
| Best for | FAQ-heavy queues with good documentation | Queues where closing a ticket requires action |
| Cost shape | Bundled with a helpdesk or per seat | Per resolution, outcome-based, or subscription |
Failure behavior separates the two just as sharply as capability. When a chatbot meets a question outside its scripts, the customer hits a wall and starts over with a human. When a well-built agent hits a low-confidence case, it escalates on its own, attaching the conversation, the account context, and what it already tried. Your team inherits a briefed ticket, not a transcript of frustration.
If you want the concept-level grounding first, our plain-English explainer on what an AI agent is covers the mechanism, and AI agent vs chatbot goes deeper on the architectural line this section summarizes. The same Gartner forecast expects the shift toward agentic resolution to cut customer service operational costs by around 30 percent, which is why the category is moving this fast.
Chatbot or AI agent: how do you decide?
Decide on one question: does resolving your most common tickets require taking action, or just giving an answer? If your queue is dominated by documented questions, a chatbot is enough. If closing tickets means touching orders, accounts, refunds, or other systems, you need an agent. Most teams discover they need both, in layers.
A chatbot is enough when:
- your top questions are documented and change rarely;
- resolution almost never requires touching another system;
- volume is high but complexity is low;
- what you mainly need is after-hours acknowledgment and clean routing.
You need an AI agent when:
- closing a ticket means checking or changing something: an order, a subscription, a refund, an account setting;
- customers repeat themselves across channels because answers do not resolve anything;
- your team spends hours on tickets that a documented policy already decides;
- you are measured on resolution rate and CSAT, not on deflection.
Once you know which category you are buying, the vendor question gets much easier. Our roundup of the best AI agents for customer support compares the tools worth shortlisting, from enterprise CX platforms to helpdesk-native agents and voice specialists, each with an honest best-for line. Read this guide first, then that one; category before vendor is the cheaper order of operations.
What support automation costs
Support automation pricing comes in four shapes: bundled or per-seat chatbots, per-resolution pricing, outcome-based enterprise contracts, and flat subscriptions. The shape matters more than the headline rate, because each one scales differently with ticket volume. Model all four against your real monthly queue before you sign anything.
- Bundled or per-seat chatbots. Basic bots ship inside many helpdesk and messaging plans, so trying one is often close to free. Cost scales with team size rather than ticket volume, which suits FAQ deflection.
- Per resolution. You pay when the AI closes a conversation; Intercom prices its Fin agent this way (Intercom, retrieved 2026). It aligns cost with outcomes, but busy months cost more, so model it against your volume.
- Outcome-based enterprise contracts. The large CX platforms sell tailored agents on sales-led deals priced against results. A strong fit at high volume; a heavy lift for a small team.
- Flat subscriptions. General agent platforms charge a predictable monthly plan with usage included, which keeps spend legible while you learn what automation is worth to your queue.
Gravity sits in the fourth category. The free tier costs nothing and includes one agent; paid plans start at $20 a month with $20 of usage included, and you can buy extra usage when a busy month needs it. For an operator automating the work behind the queue, that predictability is the point.
Whatever shape you pick, budget for the costs that never appear on a pricing page: cleaning up your help center so the AI retrieves current answers instead of stale ones, integration work to reach your order and account systems, and a few hours a week of human review while trust builds. If budget is the binding constraint, our running comparison of the cheapest AI agent platforms shows where the floor is.
How do you roll it out without hurting CSAT?
Roll out support automation the way you would onboard a new hire: supervised first, narrow scope, clear escalation path. A Gartner survey found 64 percent of customers would prefer that companies did not use AI for customer service (Gartner, 2024), so the burden of proof sits with your rollout, not with your customers.
- Baseline before you automate. Pull last month's tickets and tag the top intents by volume. Record current CSAT, resolution rate, and reopen rate. This is the yardstick every automation claim gets measured against.
- Start in draft mode. Let the AI draft replies that your team reviews and sends. You measure answer quality on real tickets with zero customer exposure.
- Go live on a narrow slice. Two or three high-volume, documented, low-emotion intents. Not billing disputes, not cancellations, not anything regulated.
- Keep the exit visible. A customer who asks for a human gets one immediately, with full context carried over. Burying the handoff is the single fastest way automation wrecks CSAT.
- Measure automated conversations separately. Track resolution, reopens, and CSAT for AI-handled tickets against human-handled ones, and expand intent by intent only where the agent wins.
Expansion, not launch, is where these programs succeed or stall. Our playbook on AI agents for SaaS support walks the queue-side rollout in detail, including which five jobs to hand over first. And once the queue is stable, the same machinery starts working ahead of it: onboarding nudges, usage-drop alerts, renewal saves. That next chapter is covered in AI agents for customer success.
Frequently asked questions
What is a customer service chatbot?
A customer service chatbot is software that answers customer questions in a chat window using scripted flows, keyword rules, or retrieval from a help center. It handles routine, repetitive questions instantly and at any hour. It does not take actions in other systems, so anything that needs account changes or judgment still goes to a human.
What is the difference between a chatbot and an AI customer service agent?
A chatbot answers questions; an AI customer service agent resolves tickets. The agent reads the conversation and account context, decides what needs to happen, takes the action in your helpdesk and back-office tools, and confirms the outcome. Chatbots deflect contact volume; agents close tickets end to end and escalate when confidence is low.
Do chatbots hurt customer satisfaction?
They can. A Gartner survey found 64 percent of customers would prefer that companies did not use AI for customer service, and a bot that traps people in loops with no visible path to a human confirms exactly that fear. Automation protects CSAT when handoff is easy, scope is honest, and automated conversations are measured separately.
How much does a customer service chatbot cost?
Pricing shapes vary more than rates. Basic chatbots often come bundled with a helpdesk plan or priced per seat. AI agents are commonly priced per resolution or on outcome-based enterprise contracts. General agent platforms use flat subscriptions. Model each shape against your monthly ticket volume; a low headline rate can cost more at scale than a flat plan.
Can a small team use an AI customer service agent?
Yes. Small teams often see the fastest payoff, because a handful of repetitive intents dominates most queues. Start with a helpdesk-native agent or a general agent platform, automate the top two or three intents, keep the human handoff visible, and expand from evidence. You do not need an enterprise contract to stop answering the same question forty times a week.
Should I replace my chatbot or run an AI agent alongside it?
Usually alongside, at least at first. Keep the chatbot doing what it does well, which is instant answers to documented questions, and hand the resolution work to an agent behind it. Replace the chatbot outright only when the agent consistently matches its speed on simple questions and beats it on everything else.
Sources
- Gartner, "Gartner Predicts Agentic AI Will Autonomously Resolve 80% of Common Customer Service Issues Without Human Intervention by 2029" (March 2025), gartner.com
- Gartner, "Gartner Survey Finds 64% of Customers Would Prefer That Companies Didn't Use AI for Customer Service" (July 2024), gartner.com
- Intercom, "Fin AI agent", retrieved 2026, intercom.com/fin