AI chatbot E-commerce Practical guide
AI chatbot, live chat or FAQ page: what each one solves
The three support tools side by side: what each one solves, what it costs and when it frustrates customers, with numbers from 100 real conversations.
The question is almost always framed as a choice between options. An online store owner notices that customers keep asking the same things, wants to answer them faster, and ends up staring at three options assuming only one of them can be picked: write a page of frequently asked questions, install live chat, or install an AI chatbot.
It is not that kind of choice. The three tools solve three different problems, and the mistake that costs money is not picking the “worst” one: it is putting one where the other was needed. This article puts them side by side, with numbers from 100 real conversations we analysed in a Cypriot online store.
The short answer
- The FAQ page solves the questions you already know you get. It costs your time to write and to maintain. Its limit: nobody reads it when they are in a hurry.
- The AI chatbot solves the questions you did not anticipate: above all, product search in natural language. Predictable cost, answers around the clock.
- Live chat solves the questions that need a decision: a price for custom work, an exception to a policy, a complaint. Its cost is not a subscription, it is availability.
- 83 out of 100 conversations end in three questions or fewer. That is why live chat is expensive per conversation: you are paying for the waiting, not for the answering.
- 14 out of 100 needed a human. That is why handoff is not an extra feature: it is a precondition.
What each one actually is
Before the comparison, one distinction clears up most of the confusion: live chat is a channel, a chatbot is a responder, an FAQ page is knowledge.
Live chat answers nothing by itself. It is a window that opens onto a human on your team. The quality of the answer is the quality of that person, and its speed is their availability.
An AI chatbot is the responder itself. It reads your catalogue, your stock and your policies and composes an answer. It has no opening hours, but it does have a limit, and that limit is yours to define.
An FAQ page is not a channel at all. It is written content that stands on its own: the customer reads it, Google reads it, and AI tools read it when somebody asks them about you. And, crucially, your bot reads it too. Without written policies there is no chatbot that knows them.
The three side by side
| Criterion | FAQ page | AI chatbot | Live chat |
|---|---|---|---|
| What it solves | Predictable questions, written once | Product search, questions against your own data, unpredictable phrasing | Decisions, exceptions, problems, complaints |
| What it does not solve | Anything you did not anticipate; it does not search the catalogue | Anything requiring human judgement or a commitment | Anything outside opening hours; it does not scale with volume |
| Cost | The time to write and maintain it | A predictable subscription scaled to volume | A human being available for every hour you cover |
| Response time | Instant, if the customer finds it | Seconds | Minutes to never |
| Availability | 24/7 | 24/7 | Only while somebody is there |
| With volume | One page, unlimited readers | Linear and cheap | Every extra conversation needs extra attention |
| When it frustrates | When it is outdated or vague | When it insists on answering where it should have handed over | When it says “we’re offline, leave your email” |
That last row deserves attention, because it is where customers are lost in all three columns. All three tools disappoint in exactly the same way: when the customer realises they are talking to something that is not going to help them.
What each one really costs
The chatbot is the easy one, because it is a subscription. At Pythia Works the packages start at €29 per month (2,500 credits), the typical small-to-medium store sits on Pro at €99 (11,000 credits, with live handoff to a human and PrestaShop integration), and Enterprise is €349 (50,000 credits). Pricing per model, with the credits per reply, is on the AI chatbot page.
The FAQ page has no price tag. It has a time cost: a good e-commerce FAQ is two to four hours of writing plus half an hour every quarter so that it does not start lying. Its real cost shows up when you forget about it: an outdated FAQ is worse than none, because the customer believes it.
Live chat is the interesting one. Its cost is not the subscription to the tool, whatever you pay for that: it is the obligation to have somebody there. And this is where the number changes the picture. In the store we analysed, those 100 conversations were spread over 28 days, roughly a dozen visitor messages a day, and 83 out of 100 closed in three questions or fewer.
Read that as a shift schedule. The actual typing is a few minutes a day; the availability required to answer within minutes is every hour the window stays open. For a store with normal daytime hours, that is more than forty hours a week during which somebody has to be interruptible. Live chat is not paid per answer. It is paid per hour of waiting.
There is a second gap live chat structurally cannot cover: in the same data, more than half of all conversations started outside opening hours, with Sunday, the one day the store is fully closed, coming out as the busiest day of the week. However good the person behind the window is, half the questions never reach them.
Four scenarios with a clear recommendation
A 2,000-product catalogue and a site search that cannot match part numbers. The problem here is not support, it is findability: the customer types a code or describes something in their own words and finds nothing. No FAQ page fixes that, and no human keeps up at volume. An AI chatbot with access to the catalogue, and then read the logs, because they also tell you what customers were looking for and you did not stock.
Five products, but every order is a quote. Cut to size, quantities, custom work. The price comes from a person, not a table. Here live chat is the right channel, and the chatbot, if there is one, works as a spec collector: it gathers dimensions, quantity and deadline and hands them over ready, so the human answers once instead of five times.
Half the questions are “how much is shipping to Limassol?” That is a missing-information problem on the site, not a channel problem. Write the FAQ page first, and put the shipping cost somewhere visible before checkout. After that, whichever tool you layer on top will at least have something to answer with.
You work alone and the catalogue changes every week. Live chat is not an option: you cannot be in the warehouse and in the chat window at the same time. What you need is coverage without presence: an FAQ for the policies, a chatbot for everything else, and a notification to you only for the few that need a decision.
The bot’s limit, and why escalation is not optional
Out of those 100 conversations, 14 needed a human: custom work and dimensions where the price is a human decision, order or payment problems, and complaints.
One of them is worth telling, because it is the most useful negative example we have. A conversation about a shipping charge escalated until the customer wrote that we were frauds. The bot carried on explaining the pricing: correctly, politely and completely beside the point. At that moment the thing that was needed was not accurate information. It was a person.
The conclusion is not “the bot failed”. It is that handoff has to be a written rule, not a hope. Three thresholds work in practice:
- Topical: a custom price, an exception to a policy, anything that commits the store.
- Emotional: once the tone turns into a complaint, the next reply is not information.
- Repetitive: when the customer asks the same thing for the third time, the bot stops answering and hands over.
If the solution you are evaluating does not let you define those three, you do not have a chatbot with live chat behind it. You have a chatbot with no exit.
The combination that works
The three tools are not competitors. They stack.
The FAQ page is the knowledge. You write your shipping, returns, payment and warranty policies once, in plain language, on a page that exists publicly. You win twice: customers who prefer to look things up read it, and AI tools read it when somebody asks them about your store. If you do not have such a page yet, it is part of the work we do on the site itself.
The chatbot is the front line. It feeds on that same knowledge, serves it in natural language, searches the catalogue and works on a Sunday night. It also covers the questions you never anticipated, and tells you what they were, so they can go into the FAQ next month.
The human is the escalation. Not the front line, not in front of every conversation: only where there is a decision, money or displeasure involved. With 83 out of 100 resolved in three questions, what is left is a small, clear set that genuinely deserves your time.
That order has a practical budget consequence too: each layer filters the next. A good FAQ reduces the volume reaching the bot, and a well-bounded bot reduces the conversations reaching a human down to the few that could not have been solved any other way.
What to do tomorrow
Open the last fifty questions you received (emails, phone calls, social messages) and put one of three letters next to each: W if the answer is already written somewhere, C if it needs the catalogue or stock levels, D if it needs a decision from you.
The share of Ws tells you how much an FAQ page will save you. The share of Cs is the reason chatbots exist. The share of Ds is the load that is never coming off your plate, and the only part where it is worth it being you.
If you want to see where your own mix falls before paying for anything, look at what the AI chatbot does and does not do, or send us your ten most frequent questions and we will tell you which of the three solves them.