When an AI chatbot takes load off customer support and when it is better to wait: scenarios, the knowledge base, handing the conversation to a manager, privacy and common mistakes.
About us
An AI chatbot makes sense on a website where the same questions come up every day: order status, delivery, payment, returns, availability, opening hours. It answers at any time and takes load off managers. Complex, disputed and sensitive conversations stay with a person, so the bot needs a knowledge base and a clear handoff of the conversation to a manager.
A regular bot works from a script: the customer presses buttons, and the bot leads them through prepared branches. A bot based on a large language model understands a freely worded question, looks for the answer in your materials and writes a reply in clear language. Between these options there is an intermediate one: search over ready-made questions and answers.
| Bot type | How it works | What it suits | Limitations |
|---|---|---|---|
| Button bot | Leads the customer through a script with ready-made choices | Booking a service, collecting contacts, simple routes | Requires choosing from ready-made options |
| Search over questions and answers | Finds the closest ready-made question and shows the stored answer | Short answers to typical questions | Answers only what is already written |
| Language model bot | Understands free text and writes an answer based on your knowledge base | Support where questions are worded in different ways | Needs a knowledge base, boundaries and review of the answers |
A language model can be confidently wrong and name a delivery condition or a price that does not exist. That is why the answers are tied to your knowledge base, and the bot is allowed to say that it has no information and hand the conversation to a person.
A bot brings value where there is something to automate. The signs are:
Order status by number (requires integration with the accounting system), size selection from a size chart, return terms, availability at a pickup point, answers about warranty and product care. The customer gets an answer right away, and only the conversations that need a person reach the manager.
Booking a consultation, collecting the initial data for a quote, answers about timelines, the work process and the required documents. The bot records the contacts and the essence of the request, and the manager receives a prepared inquiry.
Review the inquiries from the past few weeks and mark the recurring ones. Calculate how much time managers spend on such answers and compare it with the cost of the service and the setup. If there are few recurring questions, answer templates and an FAQ page are enough.
There are situations in which a bot creates more trouble than value:
In such cases it helps to put things in order first: update the delivery and payment pages, add an FAQ section, set up answer templates in the messenger. These steps bring value even without a bot.
The knowledge base matters more than the choice of service. The bot quotes what is written in it, so the texts must be accurate and up to date.
The website pages are a useful source: they answer the same questions for people and for search engines. A bot that relies on these pages gives answers that match what the customer sees on the website.
A few weeks are usually enough for a pilot. During this time, review the conversations where the customer was left without an answer every day, and correct the knowledge base.
Useful metrics:
Compare the metrics with the period before the launch and assess the quality of the answers together with the number of conversations.
The bot talks to people and collects data, so the same rules apply to it as to the forms on the website. Tell customers that they are talking to a bot, and keep the manager's contact in plain view. In some countries the law requires such a notice; the rules change, so check the current requirements with a lawyer.
Do not ask for card numbers, passwords or copies of documents in the chat. Find out from the service provider where the conversations are stored, for how long, and whether they are used to train models. Describe this in the privacy policy.
The company is responsible for what the bot says. If the bot promised a wrong price or timeline, the customer perceives it as a promise from the store. That is why the bot takes prices and availability from the accounting system and hands disputed topics to a person.
The bot calls itself a bot and does not pretend to be a live employee. This approach is more honest toward the customer and lowers the risk of complaints.
The amount depends on the service, the volume of conversations and the complexity of the setup. It usually consists of a one-time implementation and a monthly fee for the platform or for the number of conversations. To estimate the payback, compare it with the time managers spend on typical answers.
The bot takes over the recurring questions, and managers spend their time on complex cases, complaints and sales. The bot hands the conversations it did not resolve to a person together with the conversation history.
Yes, if the store or the CRM provides access to the data through an integration. Then the bot checks the order status or availability by number. Without an integration, it answers with general information from the knowledge base.
Limit the answers to your knowledge base, allow the bot to say that it has no information, and set up a handoff to a person. Regularly review a sample of conversations and correct the sources from which the bot took wrong data.
The text of conversations in the widget usually does not become part of the page for search engines, so the questions and answers are still needed on the pages themselves. The widget script adds load, so compare the speed before and after connecting it.
An AI chatbot is useful where questions repeat, there is an up-to-date knowledge base and a manager who picks up complex conversations. Start with a narrow scenario, keep a conversation log, review the answers and gradually expand the topics. Separately make sure that the bot's answers match the terms on the website pages and that the widget does not slow down loading. The result depends on the niche, the volume of inquiries and the quality of preparation, so assess it on your own data. LightSEO is a digital agency from Mukachevo that has been building websites and online stores since 2015, so it can help with a turnkey corporate website or the launch of an online store, where the chat will work on a clear page structure and honest delivery and payment terms.
