Chatbot vs live chat: which one your shop should use

Use a chatbot for the questions your shop answers the same way every time, and live chat for the ones that need judgement. A chatbot answers at 3am and costs fractions of a cent per reply. Live chat costs a person's time and wins on anything unusual. Most shops need both, in that order.

The chatbot vs live chat argument is usually framed as a choice of software. It is really a choice about which questions deserve a human. Get that split right and the tooling decision makes itself. Get it wrong and you either pay somebody to retype your returns policy forty times a month, or you hand a genuinely upset customer to a machine. If you are still mapping the wider category, our ecommerce chatbot guide covers the ground around this decision.

What live chat and a chatbot each actually are

A chatbot is software that answers a visitor's typed question automatically, without a person involved. Live chat is the same chat window staffed by a human being who reads and replies in real time. The window looks identical to the customer. What differs is who is on the other end and what happens when nobody is.

That last clause carries most of the weight. Live chat has an availability problem by construction: it works when someone is logged in and does nothing when they are not, which for a one-person shop is most of the week. A chatbot has a competence problem by construction: it works on questions its source material covers and has to be honest when a question falls outside it.

Chatbot vs live chat on the axes that matter

Nothing in that table says one wins. It says they fail differently, which is exactly why the sensible configuration is a chatbot answering first and a route to a person behind it.

Worth knowing: the real competitor to both is your email inbox. Most small shops are not choosing between a chatbot and live chat, they are choosing between either of those and answering the same delivery question by email at 11pm. Both options beat that.

Which questions should a chatbot answer?

The ones where the answer is already written down and does not change per customer. If you can point at a paragraph in your policies or product notes that answers it, a chatbot should be answering it.

There is decent evidence that these questions are the expensive ones. Baymard Institute puts the average documented cart abandonment rate at 70.22 per cent, calculated across 50 studies (Source: Baymard Institute, 2026). The stated reasons are dominated by missing information rather than missing charm: 40 per cent abandon because extra costs were too high, 20 per cent because delivery was too slow, 13 per cent because the returns policy was not satisfactory, and 12 per cent because they could not see or calculate the total order cost up front (Source: Baymard Institute, 2026).

A visitor who cannot find out what delivery to their postcode costs, or whether they can send a jumper back, does not open live chat and wait. They leave. So the chatbot list looks like this:

1. Delivery. Costs, timescales, which countries, what happens to a Saturday order.
2. Returns and exchanges. The window, who pays postage, the condition rules.
3. Sizing and fit. Measurements, what runs small, the size chart in words rather than a table image.
4. Materials, care and ingredients. The things printed on the label and nowhere else.
5. Stock and restock policy. What you do about sold out items and waiting lists.
6. Payment and checkout. Which methods you take, whether you do instalments.
7. Warranty, guarantees and aftercare. Whatever your written terms actually say.
8. Where you are and when you are open. Including the bank holiday nobody updated last year.

Every one of those is a document you already have. That is the whole trick: an assistant that answers from your uploaded policies, price lists and product guides is not being clever, it is being a search engine with manners.

Which questions still need a person?

Four categories, and they are worth writing on a sticky note:

• Anything about a specific order. Refund status, a lost parcel, a wrong item. This needs the order system and, usually, permission to make a decision.
• Anything with an emotional charge. A complaint, a gift that arrived broken, a customer who has already been let down once. A correct answer delivered by software reads as a brush-off.
• Anything worth real money. Bulk enquiries, trade accounts, corporate gifting. Someone waving a big order wants a human to acknowledge it.
• Anything your documents do not cover. New product nobody has written up yet, an edge case in your terms, a question about your business rather than your products.

That last one is where most chatbot deployments go wrong, and it is a design question rather than a staffing one. A chatbot that invents a returns window because it could not find yours has cost you more than it saved. chatfor.site answers only from material the owner uploaded and is instructed to say it does not have that to hand when the documents do not cover the question. Ask it who the chief executive is with nothing about that in the uploads and it says it does not know, which is the correct answer and an unusual one for this category of software.

Worth knowing: "I do not have that to hand, would you like to email us" is a good chatbot answer. It is a handover, not a failure. The failure mode to fear is a confident wrong answer, because you will not find out about it until the refund request arrives.

What does each one cost to run?

Live chat costs a person's attention, which is the most expensive thing a small shop owns. Three hundred questions a month at three minutes each is 15 hours of someone reading and typing, and those hours are not evenly distributed, they cluster around the moments you are trying to pack orders.

A chatbot's cost is a plan price plus a fraction of a cent per answer. chatfor.site measures 1,794 input and 87 output tokens per answer, which at Claude Sonnet 5's $2 per million input and $10 per million output tokens works out at $0.00446 per answer (Source: Anthropic documentation, verified 4 September 2026). The plans run from free for 100 answers a month to $199 for 20,000. Our full breakdown of what an answer costs to serve shows the arithmetic and the margins.

The comparison people get wrong is treating these as substitutes on cost. They are not. A chatbot removes volume from the queue; it does not remove the queue. The saving is that the fifteen hours becomes two, and the two are spent on the questions where a human answer is worth something.

Does a chatbot annoy customers?

It can, and the research on why is old enough to be settled. Nielsen Norman Group's usability work on chatbots found that "as soon as users deviated from the prescribed script, problems occurred", that bots "were generally not able to take advantage of previously entered information when a new task was started", and that customers perceived customer-service bots as "generally less helpful than human representatives", while still appreciating the speed (Source: Nielsen Norman Group, 2018).

Read that as a specification rather than a verdict. The findings condemn a particular design: a scripted decision tree that cannot handle an unexpected sentence and forgets what you just told it. They do not condemn a plain text box that reads your actual policies and answers in prose. The same research concluded that "improving the UX of your website or app will bring you higher return on investment than creating a chatbot" (Source: Nielsen Norman Group, 2018), which is a fair warning: if your delivery information is genuinely findable in one click, a chatbot is a smaller win than fixing the page.

Three things keep it on the right side of that line:

• No scripts and no menus of buttons. A text box, and answers in sentences.
• A visible route to a person. Optional pre-chat contact capture means a question the assistant cannot answer becomes an email address rather than a dead end.
• Owner review. Being able to open any past conversation, see which passages an answer was drawn from, and overrule it turns complaints into a fifteen minute fix.

How to run both without hiring anyone

The practical configuration for a shop with no support team:

1. Write the answers down once. Upload your delivery, returns, sizing and care information. Paste in anything that only exists in your head. Add curated question-and-answer pairs for the twenty things you are asked most.
2. Put the assistant on every page. One script tag, the same shape as adding Google Analytics.
3. Turn on contact capture. When a question falls outside the documents, you get the email address instead of losing the visitor.
4. Read the transcripts weekly. Fifteen minutes. Every question the assistant could not answer is a document you have not written yet.
5. Keep live chat for the hours you are actually there, if you want it at all. Many shops find that the residue after automation is small enough to handle by email.

That order matters. Automating first and staffing the remainder is cheaper than staffing everything and automating the leftovers, because the leftovers are the part that needs a person anyway.

For tool-by-tool detail see our ranked look at the tools available, and if you sell in a physical shop as well as online, the retail-specific version of this decision covers opening hours, click and collect and stock questions.

Frequently asked questions

Is a chatbot better than live chat?

Neither is better; they fail differently. A chatbot answers written-down questions instantly at any hour for a fraction of a cent. Live chat answers judgement calls, complaints and order-specific problems properly. The right setup is a chatbot answering first with a clear route to a person behind it.

Should I add a chatbot to my ecommerce site if I already have live chat?

Usually yes, because the two cover different hours and different questions. A chatbot handles the repeat questions arriving outside your working day, which is most of the week for a small shop. Live chat then deals with a smaller queue of enquiries that genuinely need a decision.

Can a chatbot hand a conversation over to a human?

Yes, and it should. chatfor.site supports optional pre-chat contact capture with a custom note, so a visitor whose question falls outside your uploaded documents leaves an email address rather than leaving the site. You pick the conversation up when you are next at a keyboard.

Do customers know they are talking to a chatbot?

They generally work it out quickly. Nielsen Norman Group found users drop politeness markers and simplify their sentences once they realise (Source: Nielsen Norman Group, 2018). That is fine. Trying to pass a chatbot off as a named human is the version that damages trust when it goes wrong.

What happens when the chatbot does not know the answer?

That depends entirely on the tool, and it is the most important thing to test before you buy. chatfor.site answers only from material you uploaded and is instructed to say it does not have that to hand rather than guess. Ask any trial assistant something deliberately uncovered and watch what it does.

Sources

• Baymard Institute, Cart Abandonment Rate Statistics (70.22 per cent average documented rate, calculated from 50 studies; stated reasons for abandonment), 2026: https://baymard.com/lists/cart-abandonment-rate
• Raluca Budiu, "The User Experience of Chatbots", Nielsen Norman Group, 25 November 2018: https://www.nngroup.com/articles/chatbots/
• Anthropic, Claude API pricing documentation (Claude Sonnet 5 at $2 per million input tokens and $10 per million output tokens), verified 4 September 2026: https://platform.claude.com/docs/en/docs/about-claude/pricing