Customer service chatbot: what it does and how to choose

A customer service chatbot is an assistant on your website that answers customer questions day and night, using the information you give it. It is strong on the questions that come back every day, such as delivery times, sizes, stock and terms, and it should pass everything else to a person.

This article covers which customer questions you can safely hand to a chatbot and which you should not, how a button bot differs from an AI chatbot, which businesses it pays off for, and how to choose the right one in six steps. It ends with how to make sure a customer still reaches a person when needed, even if your team is one or two people.

What is a customer service chatbot?

A customer service chatbot is a chat window on your website that answers customer questions by itself. A modern AI chatbot reads the question the way the customer types it, looks for the answer in your product pages, terms and FAQ, and writes the reply in plain language.

Take an online shop selling garden furniture. On a sunny Saturday evening, dozens of questions come in: is this lounge set still in stock, does the cover fit last year's model, what does delivery to the islands cost. Without a chatbot those questions sit in the inbox until Monday morning, and some of those customers will have bought elsewhere by then. With a chatbot they get an answer straight away, from the same information that is already on your site.

Which customer questions can a chatbot handle, and which not?

The rule of thumb: if the answer is written down somewhere, on your site, in your catalogue or in your terms, a chatbot can give it. If the answer depends on a decision, an exception or someone's feelings, it belongs with a person.

A wholesaler of installation materials, for example, gets asked every day which fitting suits a given pipe diameter. That answer is in the catalogue, so a chatbot can give it, even at ten in the evening when the installer is preparing tomorrow's order. Whether a large order earns an extra discount goes to the sales desk.

What is the difference between a button bot and an AI chatbot?

When people say chatbots do not work, they usually mean the button bot: a decision tree with fixed choices. An AI chatbot works differently. You notice the difference most with the question that does not quite fit the menu.

A button bot can still work for one fixed process, such as registering a return. For the broad flow of questions an online shop or wholesaler gets, an AI chatbot is the logical choice, because customers simply do not phrase their questions in your menu structure.

When does a customer service chatbot make sense for your business?

A chatbot pays for itself fastest in businesses where fixed rules or a fixed catalogue decide the answer and the same questions keep coming back. If three or more of these sound familiar, it is worth trying:

It makes less sense when nearly every question is bespoke, for example when you sell advice and every answer depends on the customer's situation. Then a chatbot can at most help collect details for the first conversation.

How do you choose the right customer service chatbot?

A demo always looks good, because it is built on sample questions. So choose on your own questions, in six steps:

1. Make your list. Take last month's emails, chats and phone notes and write down the twenty questions that come back most. This is your test set.
2. Try it on your own information. Pick a vendor with a trial, add your product pages and terms, and ask your twenty questions. Check whether the answers are right and whether they sound like your business.
3. Test what it does when it does not know. Ask something whose answer is nowhere in your sources. A good chatbot says so honestly and offers to pass the question on. A bad one makes something up.
4. Look at the handover. What happens when the customer wants a person? Does the conversation reach your team with the question and contact details attached, so nobody has to start over?
5. Ask where the data goes. Customers type names, addresses and order numbers into a chat. Ask where that is stored and whether the vendor signs a data processing agreement, as the GDPR requires.
6. Price it on your volume. Ask what you pay per month at last month's number of conversations, including VAT, what happens in a busy month, and how soon you can cancel.

In the first weeks, read along with the conversations yourself and fix whatever an answer gets wrong. That is when you learn most about what your customers really ask, and often about what is missing from your website too.

How do you make sure a customer can still reach a person?

A chatbot that keeps customers talking when they want a person costs you more than it brings in. So agree up front when the chatbot hands over and how. Three moments are enough: when the customer asks for it, when the answer is not in the sources, and with a complaint or an exception.

How you set up the handover depends on the size of your team. With one or two people you do not need to sit behind a chat screen all day. A workable setup:

• During the day, when someone is in, the customer can continue with a colleague via live chat. They see the conversation with the chatbot and do not have to ask anything again.
• Outside office hours the chatbot asks for a name and email address and is honest about when the customer will hear back. The next morning the full question is in your inbox.
• With a complaint the chatbot does not say it is solved, but that a colleague will look into it. That way it never promises something you still have to decide.

A trailer rental company with two people at the counter works well this way. The chatbot answers questions about driving licence categories, the deposit and pick-up times. If a customer wants to keep a trailer one more day, the chatbot passes the request on with the booking attached, and the staff member decides.