A chatbot is software that replies to customers automatically; live chat is a human agent replying in real time, usually in a chat window on your website. Most consumer brands need both: software for the requests your rules and order data can settle at any hour, and human agents for exceptions, judgement calls and upset customers, joined by a handoff that doesn't make the customer start again.
So the useful question isn't which one wins. It's which requests go to which, what the human agent receives when a conversation moves across, and how long the channel gives them to reply.
In short
- The difference is who answers (software or a human agent), not the channel. Both work in a website chat widget, WhatsApp, Instagram DMs and Messenger.
- "Chatbot" covers three different tools: a menu bot, an AI chatbot that answers from your content, and an AI agent that checks order data and makes changes under your rules. Compare the one you would actually use.
- Split by request. Software takes whatever a written rule and your systems can settle; a human agent takes exceptions, goodwill, suspected abuse, safety issues and strong emotion.
- The handoff decides how the split feels to the customer: what the human agent receives, and whether they can still reply on that channel.
Chatbot vs live chat at a glance
| Chatbot | Live chat (human agents) | |
|---|---|---|
| Who replies | Software, scripted or AI | A human agent, in-house or outsourced |
| When it's available | Any hour | The hours you staff |
| First reply | Seconds | Depends on the queue |
| Conversations at once | No practical limit | A few per agent |
| Consistency | The same rule every time, and wrong in the same way when the rule or data is wrong | Varies with the agent, training and workload |
| Exceptions, judgement, upset customers | Weak; should hand over | Strong |
| Cancelling an order or starting a refund | Only when connected to your order and payment systems, within rules you set | Yes, by opening those systems |
| What drives cost as volume grows | The software plan, set-up, and keeping rules and content current | Agent hours, including evenings, weekends and sale peaks |
| How it usually fails | Loops, or confident wrong answers | Long waits, and nobody there after hours |
The two products have largely merged. Many live chat software tools now let a bot answer first and pass the conversation to a human agent: Tidio's live chat comes with its Lyro AI, Freshchat has its Freddy chatbot, and LiveChat is sold alongside ChatBot, a separate product from the same company that runs in the LiveChat widget. So choosing between them is really a routing decision: which conversations the bot takes, which go straight to a human agent, and what happens in between.
"Chatbot" means three different things
Most comparisons put a human agent against a scripted bot. That isn't the choice a support team faces now. Three kinds of software answer chats, and they differ in whether they can finish the request (our guide to chatbot vs conversational AI explains the technology).
Here is one message, "Can I change my order to a size M?", on a clothing order that hasn't shipped:
| What it does with the message | What it needs | |
|---|---|---|
| Menu (rule-based) bot | Shows buttons; "Returns and exchanges" links to the policy page | A decision tree someone maintains |
| AI chatbot | Understands the question, explains that sizes can change before dispatch, and asks the customer to email support | Your policies and FAQs |
| AI agent | Finds the order, checks it hasn't shipped and that size M is in stock, makes the exchange and confirms it | Your rules, and access to the order and inventory systems |
| Human agent on live chat | The same as the AI agent, by opening the store admin, during staffed hours | Training and the same access |
The first two answer; the AI agent and the human agent complete. That difference is the subject of our guide to customer service automation, and it moves the line between bot and human. With a menu bot, almost every order change ends with a human agent. With an AI agent, the human agent mainly sees the exceptions.
Flowcall's AI Agent is the third kind, because the customer asking for a size M wants the new size, not an explanation of the policy. Each request type becomes a workflow built from your written procedure, with ready-made ones for returns, refunds, exchanges, cancellations, cash on delivery and damaged products. Your rules decide what it may do: the checks, eligibility conditions and approvals that must pass before a refund, cancellation or replacement happens. When they pass, it checks the order in Shopify or your order system and makes the change; when they don't, an agent takes over.
Live chat isn't only on your website
"Live chat" usually means the chat window on a website. For many consumer brands, that isn't the busiest chat channel: order and delivery questions also come in on WhatsApp, Instagram DMs and Messenger. The same bot-or-human decision applies on each.
The channels differ in one way that matters here: timing. Website chat is live. The visitor is on the page, waiting, and may leave. Messaging apps are asynchronous: the customer sends a message, puts the phone down and reads the reply later. A human agent who answers a WhatsApp message twenty minutes later has still helped; one who answers a website chat twenty minutes later may be writing to a closed tab.
So the split has to hold on every channel, and across them, which is where omnichannel customer service comes in. When a customer leaves the website chat and writes to your WhatsApp customer support number, the conversation should arrive as the same issue, not a new one.
In Flowcall, Live Chat, WhatsApp and Instagram DMs arrive in one inbox, and the AI Agent runs the same workflows on each, so the split you write once applies everywhere. Live Chat visitors who leave a phone number or email pick up their history when they come back on another channel, and when the same customer writes on a different channel while their issue is open, Flowcall keeps it on the same ticket.
Who should answer which request
Most of the chats in ecommerce customer service fall into a handful of request families. For each, ask whether a written rule and your systems can settle it. If they can, software should settle it, and a human agent should see only the cases the rule sends on.
| Request family | Examples | Software can settle it when… | A human agent takes it when… |
|---|---|---|---|
| Questions with a fixed answer | Delivery times, return window, payment options | The answer is in your policies | The customer disputes the policy |
| Order status | "Where is my order?" | Tracking data is connected and the parcel is moving | It's stuck past your limit, or marked delivered but not received |
| Changes before dispatch | Cancel, change size, change address | Your rules say when a change is allowed and the order can be changed through software | It has shipped, or the change needs a check the rule doesn't cover |
| Returns, damaged or wrong items | "It arrived broken", "wrong colour" | The window, product and photos meet your policy | Photos don't match, the product is excluded, or it's a repeat claim |
| Refunds | "Where's my refund?" | Refund status can be read and the amount is within your limit | It's above your limit, overdue at the bank, or disputed |
| Advice before buying | Sizing, ingredients, compatibility | The answer is in your catalogue and product content | It touches health, safety or allergies, or a high-value purchase needs reassurance |
| Complaints and goodwill | "Third time I'm asking", compensation, a threat to post a review | Never. Software collects the details and hands over | Always |
Each row has its own playbook: order tracking automation for status questions, and returns management for changes before dispatch, returns, broken items and refunds. Questions before purchase are a different job again: an AI shopping assistant answers them from your catalogue.
The table doesn't make live chat a fallback. The bottom rows are where human agents matter most: a customer who has been let down twice needs someone with the authority to make an exception, and that decision shouldn't be left to software.
Seven triggers that should hand a chat to a human agent
A split by request covers most conversations. The rest need triggers: conditions that move a conversation to a human agent whatever the request. Write them down, test them and review them like any other policy. Our guide to escalation management covers the team side.
- The customer asks for a human. Hand over at the first clear request. Making a customer ask three times turns a routine request into a complaint.
- The rule says no, or there is no rule. An exchange outside the window, a product your policy doesn't cover, an order in a state nobody planned for.
- Strong emotion or a threat to escalate. Anger, a third contact about the same order, a mention of reviews, social media or a formal complaint. Our guide on how to handle customer complaints covers what the human agent does next.
- Money above a limit. Refunds, replacements or compensation beyond what software may approve on its own.
- Suspected abuse. Repeated "item not received" claims, photos that don't match the order, a pattern across accounts.
- Safety, health or legal issues. An allergic reaction, an appliance that sparked, a legal notice. These go straight to a trained human agent.
- The bot is failing. Two answers that didn't help, the same question asked twice, or a loop. Hand over rather than guess.
Each trigger also needs a destination: the team that can act on it, not a general queue.
Several of these are features in Flowcall rather than rules you write. The AI Agent offers a human agent instead of guessing when it has no answer. Every incoming message is checked for frustration, aggression, threats to escalate and requests for an agent, and customers at risk of escalating can skip the queue to the right team. The money limits, exceptions and abuse checks are rules you write into the workflow, so they are applied the same way every time.
What the handoff has to carry
"Pass the context" appears on every comparison page. In practice the human agent needs seven things in the ticket they open, before they type anything:
- The customer, matched by phone number, email or order, not just a name typed into the chat.
- The order: status, items, delivery and payment, from your systems.
- What the customer asked, in one line, with the full conversation below it.
- What they have already sent: photos, an invoice, a new address.
- What the software checked and did, such as "return window: within", "replacement stock: available", "refund: not started".
- Why it handed over: the trigger, such as "second damage claim in 30 days; policy requires approval".
- What's left to do, and by when: the decision needed and the reply deadline on that channel.
When any of these is missing, the symptom is the same: the human agent asks the customer for something she has already given.
One late-night message, two handoffs
At 11:40 pm, a customer messages a skincare brand on WhatsApp. Her serum arrived with a broken seal and leaked, and she wants a refund rather than a replacement. It's her second damage claim this month, and the brand's policy says a second claim needs a human agent's approval.
A poor handoff. The bot asks for her order number and sends the returns policy. When she replies "I want a refund", it says an agent will get back to her. At 9:30 the next morning, a human agent opens a ticket titled "Refund request", asks for the order number and a photo of the damage, and waits. She replies at lunchtime, annoyed, and the approval slips to the next day.
A good handoff. The AI agent recognises her number and finds the order. It asks for a photo in the chat, checks that it shows the damage, and applies the policy: the refund is within the window, but a second claim needs approval. It tells her the team is back at 9 am and that she won't need to send anything again. At 9 am, the human agent opens a ticket with her order, the photo, the checks, the reason for the handoff and a reply deadline well inside WhatsApp's window. They approve the refund and reply by 9:15.
In both versions the rule sent the case to a human agent. Only the second made the handoff worth having. Our live chat examples write out the same kind of handoff in a website chat.
The second version is how a Flowcall handoff works. The AI Agent reads the photo the customer sends and checks it for damage. When a human agent is needed, it creates a ticket with the conversation, a summary, the customer and order context and what's left to do, and routes it to the right team; outside business hours, it tells the customer when the team is back. In the AI Helpdesk, the ticket shows her orders, delivery, payments and past conversations, and the human agent approves the refund from the ticket instead of opening Shopify and the payment gateway.
The reply clock: how long each channel gives the human agent
After-hours handoffs have a deadline that many teams don't plan for: the channel's own messaging rules. The WhatsApp, Instagram and Messenger rows below come from Meta's developer documentation, checked on 28 September 2026:
| Channel | How long you can reply freely | After that |
|---|---|---|
| Website live chat | While the visitor is on the page | Usually only by email, if the chat collected one |
| 24 hours from the customer's last message (the "customer service window") | Only pre-approved template messages | |
| Instagram DMs and Messenger | 24 hours from the customer's last message | A human agent can still reply manually for up to 7 days, using Meta's Human Agent tag |
| No limit | — |
Three things follow for the split:
- Night-time WhatsApp handoffs need a morning deadline. If the human agent picks the ticket up after the window closes, their first message has to be an approved template, which is a poor way to answer a complaint.
- Website chats that hand over after hours should collect a contact route before the visitor leaves, or move to WhatsApp or email as the same issue.
- The bot should say what happens next. Before handing over at night, it should tell the customer when the team is back and that they won't need to repeat anything. That is the difference between a customer who waits and one who chases.
If you promise 24/7 customer support, this is where the promise is kept or broken.
In Flowcall, response and resolution targets are set for each team or workflow and follow your business hours and holidays or run around the clock. Set a first-response target shorter than WhatsApp's 24-hour window, and the people you choose are alerted before a handed-over ticket breaches it; breached tickets move up the queue.
How to measure the split
A bot's reply in seconds makes per-channel averages such as first response time look healthy, even when the conversation then waits all night for a human agent. Measure the split itself:
| Measure | How to count it | What it tells you |
|---|---|---|
| Settled by software, per request type | Conversations of that type closed without a human agent and with no repeat contact about the same order within 7 days, divided by all conversations of that type | Whether software finishes the work, not just the chat |
| Handoff rate by trigger | Handoffs for each of the seven triggers, divided by conversations | Which rules, content or integrations to add next |
| Wait after handoff | Time from handoff to the human agent's first reply; separately, WhatsApp handoffs first answered after the 24-hour window | Whether staffing matches the hours your bot hands over |
| Handoff restarts | Handoffs where the human agent asked for something the customer had already given, divided by all handoffs | Handoff quality |
| CSAT by path | Ratings for software only, software then human agent, and human agent only | Whether customers accept software for each request type |
Be careful with "containment" or "deflection". A customer who gives up on the bot and leaves counts as contained, and the 7-day repeat-contact check is what catches the ones who come back on another channel. For the wider set of customer service metrics, see our guide. Numbers also miss tone and near-misses, so read a sample of handed-over conversations every week.
Flowcall reports CSAT for the AI Agent and human agents separately, by channel, team and agent, next to first response time, resolution and the AI automation rate. Reading a sample only catches what you happen to read, so Flowcall's AI Quality Assurance checks every handed-over conversation for missed or late handoffs and loops, and quotes the conversation in each finding.
Chatbot, live chat or both: how to choose
| Choose… | If… |
|---|---|
| Live chat with human agents only | Volume is low, most conversations are one-off advice or high-value purchases, and you can staff the hours customers write |
| A basic chatbot in front of live chat | Most questions have fixed answers (delivery times, policies) and order changes are rare enough for human agents to handle |
| An AI agent with human agents behind it | You get a high volume of order, delivery, return and refund requests, your orders sit in Shopify or another system with an API, and customers write on WhatsApp and Instagram at all hours |
| Human agents first, with AI assisting them | Most conversations need judgement, or your category is regulated (health, finance), and a human agent should own every conversation from the first message |
Some limits are worth saying plainly. If you get a few dozen chats a day and most are bespoke, a chatbot may not repay the set-up, and a well-run team following good live chat best practices will serve those customers better. An AI agent is only as good as the rules you write and the systems it can reach: if your order data lives in spreadsheets, start with answering and routing, not acting. And if you run a small Shopify store and mainly need a chat widget, start with our comparison of each Shopify live chat app worth considering.
How Flowcall handles chatbot and live chat work
Flowcall puts the software and the human agents on the same conversations: the AI Agent settles what your rules allow on WhatsApp, Instagram, email and Live Chat, collecting photos or a new address in a form inside WhatsApp or Live Chat, and hands the rest to the AI Helpdesk with the work done so far. Your team sets the split, the triggers and the reply targets; Flowcall applies them the same way on every channel and shows you, through CSAT for the AI Agent and human agents and QA findings, where to move the line next.
If you'd like to see an AI Agent hand a WhatsApp or live chat conversation to a human agent with the order, the photos and the reason attached, book a demo.
Frequently Asked Questions
Is LiveChat a chatbot?
Not on its own. LiveChat is live chat software from Text, Inc.: an inbox where human agents answer customers, with AI agents alongside them. Its chatbot, ChatBot, is a separate product from the same company that runs in the LiveChat widget and passes conversations to human agents. More generally, "live chat" means a human agent replying, though many live chat tools now let a bot answer first.
How can you tell if a chat is a bot?
Ask it. A well-run brand says when a customer is talking to AI and offers a human agent. In the EU, disclosure has been a legal duty under the AI Act since 2 August 2026: providers of AI systems that talk directly with people must design them so people are told they are interacting with AI, unless that's obvious.
What are the four types of chatbots?
There's no single agreed list. For customer service, a useful split is menu or button bots that follow a decision tree, keyword bots that match words to set answers, AI chatbots that understand free text and answer from your content, and AI agents that also check your systems and make changes under your rules. For the basics, see what is a chatbot; to compare tools, see our guide to AI chatbots for customer service.
Is ChatGPT a chatbot?
Yes, ChatGPT is a general-purpose AI chatbot. A customer service chatbot has a narrower job: it should answer from your policies and order data, act only within your rules, and hand over when it can't help. Our guide to ChatGPT for customer support covers the work it can do for your agents.
Is a chatbot cheaper than live chat?
At high volume, software usually costs less per conversation than agent time, but compare the right costs. Live chat costs grow with agent hours, including evenings, weekends and sale peaks. Chatbot costs are the plan (often priced per conversation or per resolution), the set-up, and the time spent keeping rules and content current. Count failures too: a bot that hands over late or wrongly adds work rather than removing it.
Can a chatbot replace live chat agents?
Not for a consumer brand with real exceptions. Software can take over the requests your rules settle, which changes what your human agents spend their day on rather than removing the need for them. We look at the wider question in will AI replace customer service.
