Skip to content

What is a chatbot, how does it work, and how do you build one customers don't hate?

Everyone has called a helpline and heard “press 1 for…, press 2 for…”, then pressed 0 again and again, hoping for a person. A chatbot can be that menu. It can also be the most helpful member of your team at midnight. The difference is entirely in how it is designed.

Customer conversations4 min readBy Reoncore Labs

Short answer

A chatbot is software that holds a conversation in a chat window, on Messenger, WhatsApp or a website. Rule-based chatbots follow fixed buttons and keywords; AI chatbots understand free-form messages using a language model and answer from information you provide. A good business chatbot answers common questions accurately from your real data, collects what your team needs and hands over to a person whenever it should.

01Two kinds of chatbot, and the hybrid

Kind
Rule-based
How it works
Follows buttons, menus and keywords set in advance
Good for
Limited, predictable questions: opening hours, location, how to order
Kind
AI chatbot
How it works
Understands free-form language and answers from the business's own information
Good for
Varied questions, natural language, mixed Bangla and English
Kind
Hybrid
How it works
Buttons for the common paths, AI for free text, a person for the rest
Good for
Most real businesses

AI chatbots are more capable and carry more risk. Without the right information they can guess. That is why a good one answers only from what you have given it, and passes the conversation on when it does not know.

02How an AI chatbot answers a message

  1. 01A customer writes, in a mix of Bangla and English letters: “Is this shoe available in 42? Cash on delivery possible?”
  2. 02The chatbot works out what is being asked: a stock question and a payment question.
  3. 03It looks the answers up in your information: the stock list and the payment policy.
  4. 04It writes a short reply in the customer's style, using only what it found.
  5. 05If something is missing or sensitive, it says so and passes the conversation to your team with a summary.

Steps three and five decide whether the chatbot helps or harms. Without real data, it guesses. Without a hand-over, it traps people.

03Where chatbots help a business

  • Sales: instant answers on price, stock, size and delivery, and collecting the details of interested customers.
  • Support: order status, return rules and common problems, any hour of the day.
  • Bookings: appointments for clinics, salons, consultants and service businesses.
  • Qualification: asking two or three questions so the sales team knows who to call first.
  • Inside the company: staff questions answered from policies and documents.

04Why people hate chatbots, and how to avoid it

  • Loops: the bot asks the same question again and again. Fix: after two failed attempts, offer a person.
  • Pretending to be human. Fix: say clearly that it is an automated assistant.
  • No way out. Fix: “talk to a person” works at any point.
  • Confident wrong answers. Fix: answer only from approved information, and allow “I don't know”.
  • A form pretending to be a chat: ten questions before any help. Fix: help first, ask only what is needed.
  • Forgetting the conversation: asking for the order number twice. Fix: keep the details the customer already gave.

05Six decisions before you build one

  • The job: start with one clear job, such as answering price and stock questions, not everything at once.
  • The channel: Messenger, WhatsApp, your website, or several.
  • The source of truth: product list, prices, policies, FAQs, and who keeps them current.
  • The hand-over: which questions always go to a person (complaints, payments, anything uncertain) and how fast someone responds.
  • The tone: friendly and short, in the languages your customers actually use.
  • The measure of success: reply time, conversations settled without a person, hand-overs, orders, and complaints about the bot itself.

06Test a chatbot idea for free first

Before building anything, test what the chatbot should say:

  1. 01Copy your last 100 customer messages, with names and numbers removed.
  2. 02Sort them into groups: price, stock, delivery, ordering, complaints, other.
  3. 03For the three biggest groups, write the ideal answer and the information it needs.
  4. 04Set those three up as FAQ buttons or keyword replies in Meta Business Suite and watch for a week.
  5. 05Count how many conversations they settle on their own.

That week tells you, with your own data, whether a full AI chatbot is worth it and what it must know. It is also the best brief you can hand to anyone who builds one.

Questions

Is a chatbot the same as AI?

No. Many chatbots are simple rule-based menus with no AI at all. AI chatbots use a language model to understand messages written freely.

Can a chatbot work in Bangla and Banglish?

Rule-based bots handle Bangla buttons easily. AI chatbots can understand Bangla and Bangla typed in English letters, but quality should be tested with your real customers' messages.

How much does a chatbot cost?

A simple rule-based bot on Messenger can cost nothing using Meta's built-in tools. An AI chatbot connected to your stock, orders and WhatsApp has running costs (AI usage, WhatsApp template fees, hosting) plus the work to build and maintain it. Compare it with the staff time and lost sales it saves.

Will a chatbot annoy my customers?

A badly designed one will. A good one answers quickly and correctly, is honest about being automated, and lets people reach a person at any time.

Last updated

Tell us what is not working.

Describe the problem in your own words. We will tell you honestly whether we can help, and how we would approach it.