AI & Automation

Chatbot vs Live Chat: Which Is Better for Your Business?

Chatbot or live chat? Verified 2026 pricing, speed and satisfaction data, a break-even cost model, and the hybrid bot-plus-human setup most stores end up with.

Chatbot vs live chat featured graphic comparing AI chatbot cost per resolution with human live chat cost and the hybrid handoff model
On this page
  1. Chatbot vs live chat: the short answer
  2. What each option costs in 2026
  3. Worked example: when a chatbot actually saves money
  4. Speed: where chatbots win outright
  5. Customer satisfaction: what the data says
  6. Which conversations belong to which
  7. The hybrid model most stores end up with
  8. Liability and disclosure rules you can’t skip
  9. Frequently asked questions
  10. What we’d do first

For most online stores, chatbot vs live chat ends in a split, not a winner: an AI chatbot takes the repetitive questions instantly, at roughly $0.58 to $1 per conversation on November 2026 price lists, and trained people take anything involving money, judgment or an upset customer. Pick only one and you either pay people to paste tracking links all day or leave your angriest customers arguing with software.

The useful question is where the line sits for your store. We re-checked the published pricing for Intercom Fin, Gorgias AI Agent and Tidio Lyro this month, pulled the best benchmark data we could find on speed and satisfaction, and built a hypothetical cost model. That model turned up the number most comparisons skip: below a certain resolution rate, adding a bot costs more than it saves. In our example that line sits at about 40%.

Key takeaways

  • AI agents are priced per outcome now: $0.99 per Fin outcome on Intercom, $0.90 per resolved conversation on most Gorgias plans, and about $0.58 per Lyro conversation on Tidio’s annual pricing.
  • Tidio bills when Lyro replies, Gorgias and Intercom bill when the bot resolves. Compare cost per resolution, not the sticker price.
  • Deflection is not the same as labor saved. The bot takes the easy chats, so what is left for your team takes longer per conversation.
  • Customers do not hate bots. They hate being trapped. In Gartner’s survey, 64% would prefer companies did not use AI for service, and difficulty reaching a person was the top worry.
  • You are liable for what your bot says, and EU-facing chatbots must say they are AI from August 2, 2026.

Chatbot vs live chat: the short answer

A chatbot is better for speed, after-hours coverage and high-volume repeat questions. Live chat is better for anything that needs judgment, an exception or empathy. Most stores should run both, with the bot answering first and a clear route to a person.

By “chatbot” we mean an AI agent grounded in your help center, policies and order data, not a button-tree bot. Those failed on anything off-script. Today’s fail differently: they answer confidently and are occasionally wrong, which is worse if nobody is watching.

AI chatbot Live chat (human agents)
How you pay Per resolution or per conversation, plus a platform plan Agent hours, plus helpdesk seats or ticket plans
First reply Seconds, any hour Seconds to minutes while staffed, hours when not
Best at Order status, policies, sizing, product facts Refunds, exceptions, complaints, high-value sales
Scales with volume Yes, cost rises per conversation Only by hiring or longer queues
Consistency Same answer every time, including the same wrong one Varies by agent and by shift
Main risk Confident wrong answers, customers trapped in a loop Slow replies, cost, coverage gaps

What each option costs in 2026

An AI chatbot costs somewhere between about $0.58 and $2 per handled conversation at list price, depending on the vendor and on whether you pay per reply or per resolution. Live chat costs whatever an agent hour costs you, divided by the conversations an agent clears in that hour, which in most of the budgets we see is several dollars per chat.

Customer service chatbot cost per resolution

Here is what the three tools we get asked about most actually charge, checked in November 2026:

  • Intercom Fin: Intercom’s pricing page lists Fin from $0.99 per outcome, with no seats required for Fin itself. Human seats are $85 on Advanced and $132 on Expert. Essential is listed at $29 a seat, though the page was showing $19 for new customers when we looked.
  • Gorgias AI Agent: Gorgias says each resolved conversation costs $0.90 on most plans and $1 on Starter. Overage for support-only interactions runs $1 to $2 on monthly billing. The helpdesk itself is priced by ticket volume, not by agent.
  • Tidio Lyro: per Tidio’s own pricing roundup (updated October 2026), Lyro starts at $39 a month for 50 conversations, and works out to $0.58 per conversation on annual billing from 200 a month up to 1,000. The same page lists Zendesk’s automated resolutions at $1.50 committed or $2 pay-as-you-go.

The catch is in the billing basis. Tidio counts a Lyro conversation as any conversation where Lyro replies at least once. Intercom and Gorgias charge when the bot resolves the conversation. So if Lyro resolves half the chats it touches, its real cost per resolution is $1.16, not $0.58. Still cheap, just not the cheapest.

Bar chart comparing chatbot vs live chat cost per conversation in November 2026: Tidio Lyro, Gorgias AI Agent, Intercom Fin and Zendesk list prices against a hypothetical human live chat cost
List prices per AI-handled conversation, and what a human chat costs in our hypothetical model.

Live chat support cost is mostly people

Software is the small line for live chat. The big one is agent time, and nobody publishes your number for you. Take the fully loaded hourly cost of an agent (wage, taxes, tools, management) and divide it by the chats one agent closes in an hour.

For the rest of this post we use a hypothetical: an agent costs $25 an hour fully loaded and clears 6 chats an hour by handling two or three at once. That puts a human chat at about $4.17. Yours could be half that offshore or double it in London or Sydney.

Watch out: per-outcome pricing gets expensive in a hurry if you let the bot “resolve” things it should not. A bot that closes a refund request by quoting the returns policy counts as a resolution on your invoice and a failure in your customer’s eyes. Exclude money conversations from the bot before you go live, not after the first bill.

Worked example: when a chatbot actually saves money

A chatbot saves money once it resolves enough conversations to offset two things: its own per-resolution fee, and the fact that the chats it leaves behind are harder. In the hypothetical below that break-even sits at roughly a 40% resolution rate. Under it, the hybrid setup costs more than staying all-human.

Say your store gets 2,000 chat conversations a month. All-human, at 6 chats an hour, that is 333 agent hours, or $8,333 at $25 an hour. Now add an AI agent at $0.99 per resolution. The bot takes order-status and policy questions, which are the fast ones. What reaches your team is returns, damaged items and fiddly pre-sale questions, so assume agents now clear 4 chats an hour instead of 6.

Bot resolution rate Bot fees Human hours Human cost Monthly total vs all-human ($8,333)
30% $594 350 $8,750 $9,344 +$1,011
40% $792 300 $7,500 $8,292 about even
50% $990 250 $6,250 $7,240 −$1,093
70% $1,386 150 $3,750 $5,136 −$3,197

Look at the 50% row. Half the conversations disappear from the queue, but agent hours only fall by a quarter, from 333 to 250. Vendor ROI calculators usually miss this, because they assume the chats that remain are average ones. They are not.

Seats and ticket plans are left out because you pay them either way. So are sales lost to a bad bot, which can outweigh everything in the table.

Line chart of monthly support cost against chatbot resolution rate in a hypothetical 2,000-conversation store, showing the hybrid model breaks even with all-human live chat at about 40 percent
Hypothetical store, 2,000 chats a month: the bot pays for itself at around a 40% resolution rate.

Speed: where chatbots win outright

Chatbots win on speed whenever the question has a known answer, because they reply in seconds at 3 a.m. on a Sunday. Live chat is also fast while someone is on shift, so the real gap shows up outside staffed hours.

The Comm100 Live Chat Benchmark Report 2025, built on more than 220 million chats, put the average live chat wait in 2024 at 23.6 seconds and the average reply to each message at 44.8 seconds. An average chat lasted 8 minutes 50 seconds. Good numbers, from teams that staff their chat. They say nothing about the overnight email form most small stores fall back to.

On the bot side, Klarna said its AI assistant handled two-thirds of service chats in its first month, with resolution typically under two minutes, as CX Dive reported.

One trap: first-reply time flatters bots. A bot that answers in one second, then loops for four minutes before handing over, is slower than a human who answers in 40 seconds and fixes it. Track time to resolution instead.

Customer satisfaction: what the data says

Human live chat still scores better on satisfaction for anything beyond simple questions, and customers are openly wary of AI service. What they object to most is not the bot. It is losing the route to a person.

A Gartner survey of 5,728 customers found 64% would prefer companies did not use AI for customer service, and 53% would consider switching to a competitor over it. The top concern was that AI would make it harder to reach a human. For comparison, Comm100 puts live chat satisfaction at a steady 79.9%.

Klarna is the cautionary tale everyone in support now knows. After a year of leaning hard on its assistant, its CEO said cost had been “too predominant” a factor and quality dropped. The company started recruiting human agents again and now promises customers can always reach a person. It did not switch the bot off, though: per CX Dive, the assistant still handles two-thirds of inquiries.

In the support flows we review, satisfaction drops in three places: the bot will not admit it does not know, the “talk to a person” option is hidden, or the human who picks up asks for the order number the customer already gave. All three are design problems, not model problems.

From our team: put a visible “Talk to a person” option in the chat from the first message, not after three failed answers. It feels like it will hurt your resolution rate. In our experience, customers who can see the exit are more willing to try the bot, and those who leave at once were never going to be resolved by it.

Which conversations belong to which

Route by what the conversation is about, not by channel. Anything with one correct answer that lives in your data goes to the bot. Anything that needs a decision, an exception or reassurance goes to a person.

Conversation type Who should handle it Why
Where is my order? Chatbot, with order lookup One right answer, high volume, needs a sign-in or order number check
Shipping times, returns policy, sizing Chatbot, grounded in policy pages Repetitive, and the source text already exists
Product questions under your average order value Chatbot first Speed matters more than a personal touch
Considered purchases (furniture, e-bikes, B2B) Chatbot qualifies, human closes One good conversation is worth a lot of revenue
Refunds, damaged items, exceptions Human; bot collects photos and order ID Needs judgment and authority to spend money
Complaints, chargeback threats, legal words Human, immediately Tone and liability

AI chatbot vs live agent for pre-sale questions

Pre-sale is where the AI chatbot vs live agent choice really comes down to order value. For a $30 product, a fast answer about materials or fit is all anyone wants, and an AI shopping assistant handles it well (setup and metrics are in our guide to AI shopping assistants on Shopify). For a $3,000 purchase, a person who can compare models, check stock and offer a call is worth the cost. A good pattern for high-ticket stores: let the bot answer the first factual question, then offer a human the moment the customer compares two products or asks about delivery to a specific date.

The hybrid model most stores end up with

The hybrid model is an AI chatbot answering first, with rules that pass specific conversations to a person with full context. Most stores land here after trying either extreme. It only works if the handoff is designed as carefully as the bot.

Gorgias, Intercom, Tidio and Zendesk all run AI and human agents in one inbox. The software is rarely the problem. The rules are.

Chatbot to human handoff rules that work

These are the triggers we set on almost every build, roughly in order of priority:

  1. The customer asks for a person. In any wording, including “agent”, “real person” and swearing. No “are you sure?” step.
  2. Money words appear. Refund, chargeback, cancel subscription, charged twice, discount not applied.
  3. Two answers in a row miss. The customer rephrases or says “that’s not what I asked”. A third attempt rarely lands.
  4. Sentiment drops. Capitals, repeated punctuation, words like “ridiculous”. Most platforms flag this for you.
  5. The order or customer is high value. A cart above a set amount, a VIP or wholesale tag in Shopify, or a repeat buyer with a problem.

The handoff itself needs three things: the transcript, the order the customer is asking about, and what the bot already tried. If your agent has to ask “can I have your order number?”, the bot conversation was wasted and the customer knows it. Outside staffed hours, the bot should say plainly that no one is online, collect the details and give a real reply time.

Flow diagram of the hybrid chatbot vs live chat model: AI agent answers first, five handoff triggers route to a human agent with full context, and after-hours conversations become tickets
The hybrid flow we design most often: bot first, five triggers to a human, and an honest after-hours path.

Getting these rules, tone and fallbacks right is design work as much as engineering, which is why our conversational ui design projects start with real transcripts rather than a bot builder. If you are building the bot yourself, our guide on how to build an AI chatbot customers don’t hate covers grounding and testing, and conversational UI design principles covers message patterns and error states.

Not sure where your bot-to-human line should sit?

Send us a month of chat or ticket exports and we will map which conversations a bot can safely take, what it would cost, and where the handoffs belong.

Get a support flow review

From our team: once a week, open your own chat in a private window and play an annoyed customer with a damaged order. Ask for a person in a slightly odd way. Then check what the agent saw. Ten minutes of this catches more handoff bugs than any dashboard.

Liability and disclosure rules you can’t skip

You are responsible for what your chatbot tells customers, and in the EU you must tell people they are talking to AI. Neither rule depends on which vendor you use.

The case support teams cite is Moffatt v. Air Canada. In February 2024 a British Columbia tribunal held the airline liable after its website chatbot described a bereavement fare refund that its actual policy did not allow. Air Canada argued the chatbot was responsible for its own actions. The tribunal disagreed and treated the bot as part of the website, as the American Bar Association’s summary explains. The damages were small, around CA$650. The precedent is not.

In Europe, Article 50 of the AI Act requires that people are told they are dealing with an AI system unless that is already obvious, no later than the first interaction. The Digital Omnibus pushed back the high-risk rules, but as Jones Walker’s July 2026 analysis sets out, the chatbot disclosure duty still applies from August 2, 2026. If you sell into the EU, name the bot as an AI assistant in its first message. It costs nothing and is good practice anywhere.

Watch out: do not let the bot make promises it cannot keep. Keep refunds, discount codes and delivery guarantees behind approved actions or a human. A bot that improvises a 20% apology discount is creating an obligation for you, one conversation at a time.

Frequently asked questions

Is a chatbot better than live chat?

A chatbot is better for fast, repetitive questions such as order status and returns policy, while live chat is better for refunds, complaints and high-value sales. Most stores get the best result from both: the bot answers first and hands off to a person with full context. In our hypothetical model, the bot only saves money once it resolves about 40% of conversations.

Can a chatbot replace customer support agents?

No, a chatbot can reduce the number of agents you need but should not replace them entirely. Customers still need a person for exceptions, money and complaints, and Gartner found difficulty reaching a human was their top worry about AI service. Klarna leaned heavily on its AI assistant, then began hiring human agents again and promised customers could always reach a person.

How much does a customer service chatbot cost?

A customer service chatbot typically costs about $0.58 to $2 per handled conversation at list price in November 2026. Intercom Fin charges $0.99 per outcome, Gorgias $0.90 per resolved conversation on most plans, and Tidio Lyro about $0.58 per conversation on annual billing. Check whether you pay per reply or per resolution, because that changes the real cost.

Do customers prefer live chat or chatbots?

Customers generally prefer a human for anything complicated, but accept chatbots for quick answers when they can still reach a person. Gartner’s survey found 64% would prefer companies did not use AI for customer service. Comm100’s benchmark puts live chat satisfaction at 79.9%. A visible route to a human matters more than whether the first reply comes from a bot.

When should a chatbot hand off to a human?

A chatbot should hand off when the customer asks for a person, when money is involved, after two missed answers, when the customer is clearly frustrated, or when the order or customer is high value. The human should receive the transcript, the order details and what the bot already tried, so the customer never has to repeat themselves.

Do I have to tell customers they are talking to a chatbot?

Yes, if you serve customers in the EU, and it is good practice everywhere else. Article 50 of the EU AI Act requires that people are told they are interacting with an AI system from August 2, 2026, unless it is already obvious. A first message that introduces the bot as an AI assistant covers it.

What we’d do first

Start with your own conversations, not a tool.

  1. Export the last 90 days of chats and tickets and tag the top ten reasons customers contact you.
  2. Add up the share that has one correct answer in your data: order status, shipping, returns policy, sizing, stock. If it is under about 40%, a bot will struggle to pay for itself, and better help pages may do more.
  3. Price it on your numbers. Use your real agent cost, chats per hour and volume in the table above, not the vendor’s ROI calculator.
  4. Launch the bot on three or four intents only, with the five handoff triggers, a visible human option and an AI disclosure in the first message.
  5. Read 50 bot transcripts a week for the first month. Fix the knowledge source, not the prompt, when answers are wrong.

Expand the bot’s scope only when the resolution rate holds and complaints stay flat. Slower than the launch decks suggest, and it works.

We design and build chat experiences on Shopify and custom stacks, and we have built storefronts for brands like Delsey, EcoFlow and Maker & Moss. If you want help drawing the line between bot and human for your store, our conversational UI and UX design team can map it with you.

Let's connect

Got a project in mind?

Tell us what you're building. We'll come back with the right stack, a realistic budget and a timeline from kick-off to launch. Free, and no obligation.

You have unsaved changes. Do you really want to close?