Using a Chatbot for Business: An Implementation Guide

By Umair Ali Sadaqat · Updated September 26, 2026 · Editorial policy

How to decide whether your business needs a chatbot, which type to choose, and how to implement one in six steps, including writing the knowledge base, setting up human handoff and measuring whether it helps.

Using a chatbot for business makes sense when you get the same questions over and over, or lose enquiries that arrive outside working hours. Implementing one well comes down to three things: a narrow job for the bot, a clean source of answers it is allowed to use, and a fast route to a real person when it cannot help. Skip any of those and the bot frustrates customers more than it helps them.

This is a vendor-neutral guide. It walks through whether you need one, which kind, and how to roll it out.

Step 0: check whether you actually need one

Before buying anything, look at the last month of customer emails, messages and calls. Sort them into piles.

If most are the same ten questions (opening hours, pricing, delivery times, "do you serve my area?"), a chatbot can help, but first ask why your website is not answering them. Often a better FAQ page solves it for free.
If most are unique, complex or emotional (complaints, custom quotes, account problems), a chatbot will mostly get in the way. Better human response times help more.
If you are losing leads after hours, a bot that captures contact details and books a callback may be worth it.

A chatbot is a tool for a specific volume problem. It is not a sign of a modern business in itself.

The main types

Type	How it works	Good for	Watch out for
Rule-based (buttons, decision trees)	Customer clicks options you designed	Routing, simple bookings, lead capture	Frustrating when the question is not on the menu
AI knowledge-base bot	Answers free-text questions using content you supply	FAQs, product questions, policy lookups	Wrong or invented answers if not constrained
Live chat with bot triage	Bot greets and collects details, a human takes over	Sales and support with real staff	Needs people available to answer
Messaging-channel bots	Runs in WhatsApp, Messenger or Instagram DMs	Businesses whose customers already message there	Each platform has its own rules and approval steps

Many products combine these. Common platforms include Intercom, Zendesk, Tidio, HubSpot and Crisp, among others. Features and pricing change often, so check current pricing and trial before committing.

Implementation in six steps

1. Give the bot one clear job

Write a single sentence: "The bot answers questions about X and collects details for Y. Everything else goes to a person." Examples:

A dental clinic: answer questions about hours, location, insurance accepted and new patient forms; collect name and number for booking requests.
An online store: answer shipping, returns and order-status questions; hand anything about a damaged item to staff.
A B2B service: qualify leads with three questions and book a call.

A narrow job is easier to test and far less likely to go wrong.

2. Write the knowledge base before you configure anything

An AI chatbot is only as good as the content it answers from. Create a single, maintained document or help centre with:

Every recurring question, answered in plain language.
Prices, policies and service areas stated exactly, with the date they were last checked.
What you do not do (a surprising number of enquiries are about services you do not offer).

Keep this as the one source of truth. When a price changes, it changes here, and the bot, the website FAQ and staff all use it.

3. Set guardrails

Configure the bot so it:

answers only from your approved content,
says clearly when it does not know, instead of guessing,
never makes commitments on refunds, legal matters, medical advice or custom pricing,
tells users they are talking to an automated assistant.

Transparency matters. Laws and regulations in some regions require disclosure that a user is interacting with AI, and customers generally react better when they know.

4. Design the human handoff

This is where most business chatbots fail. Decide:

Triggers: the customer asks for a person, uses words like "complaint" or "cancel", or the bot fails to answer twice.
During hours: the chat passes to a named team inbox, with the conversation history attached so the customer does not repeat themselves.
After hours: the bot says when someone will reply, collects name, email or phone and the question, and creates a ticket.

Never trap a customer in a loop. A visible "talk to a person" option on every screen is worth more than any clever automation.

5. Handle data carefully

Chat conversations often contain personal data: names, phone numbers, addresses, sometimes health or payment details.

Tell people what you collect and link your privacy policy in the chat window.
Check where the provider stores data and whether it meets the rules you are subject to (for example GDPR in the UK and EU).
Do not ask for card numbers or passwords in chat.
Set retention: delete old transcripts you do not need.

6. Test, launch small, then review weekly

Test with 30 to 50 real questions from your past emails, including awkward ones. Note every wrong or vague answer and fix the source content.
Test on a phone. The chat bubble should not cover your main call-to-action button or the cookie banner.
Launch on a few pages first, such as the contact and pricing pages, before putting it everywhere.
Every week, read a sample of conversations. Look for unanswered questions and add them to the knowledge base.

Adding it to your website

Almost every provider gives you a small JavaScript snippet. The typical process:

Create the bot in the provider's dashboard.
Copy the embed snippet.
Paste it into your site builder's custom code or header/footer area (or ask your developer).
Publish, then check it loads on desktop and mobile.

Chat widgets add scripts to every page, so check page speed afterwards. If load time gets noticeably worse, configure the widget to load after the page, or only on key pages. Our guide on how to add live chat to your website covers the embed process in more detail.

What to measure

Metric	What it tells you
Resolution without handoff	How many questions the bot fully answered
Handoff rate	How often people needed a human; very high means the bot's job is too broad
Unanswered questions	Gaps in your knowledge base
Satisfaction rating (if available)	Whether people found it helpful
Leads or bookings captured	Direct business value, especially after hours
Total support volume	Whether staff time actually went down

Compare against the month before launch. If support volume did not fall and satisfaction is poor, simplify the bot's job or switch it to a lead-capture form.

Common mistakes

Launching a bot that tries to answer everything.
No human handoff, or one that loses the conversation history.
Letting the bot quote prices from outdated content.
Hiding the fact that it is a bot.
Never reading the transcripts after launch.
A widget that covers the "Book now" button on mobile.

A pre-launch checklist

[ ] One-sentence job written down
[ ] Knowledge base complete and dated
[ ] Guardrails: only approved content, says "I don't know"
[ ] Discloses it is automated
[ ] Human handoff tested in and out of hours
[ ] Privacy notice linked, retention set
[ ] Tested on 30+ real questions and on mobile
[ ] Weekly transcript review scheduled

A chatbot works best on top of a website that already answers the basics clearly. If yours does not, fix that first: a strong FAQ and a clear contact page do a lot of the same job. We.Inc builds websites and landing pages from a chat description, and you can paste a third-party chat widget's snippet into your site like any other embed.

Start free

Frequently asked questions

Does my small business need a chatbot?

Only if you get a steady volume of repetitive questions that a well-written FAQ page is not already answering, or you receive enquiries outside working hours that you lose. If you get a handful of messages a week, a clear contact form and a good FAQ page usually serve customers better.

What is the difference between a rule-based and an AI chatbot?

A rule-based bot follows buttons and decision trees you design, so it is predictable but limited. An AI chatbot uses a language model to understand free-text questions and answers from content you provide, which is more flexible but needs guardrails, testing and a clear handoff to a human.

How do I stop an AI chatbot from making things up?

Restrict it to answering from your own approved content, tell it to say it does not know and offer a human when the answer is not in that content, keep prices and policies in one maintained source, and review real conversations weekly to fix gaps.

How do I add a chatbot to my website?

Most chatbot and live chat providers give you a short script snippet. You paste it into your site's header or footer settings, or into a custom code area, and the chat bubble appears on every page. Test it on mobile before going live.

How do I measure whether a chatbot is working?

Track how many conversations are resolved without a human, how many hand off to staff, customer satisfaction ratings if the tool offers them, leads or bookings captured, and the questions it failed to answer. Compare support volume before and after.