What an AI chatbot is, and how it differs from an auto-reply
An auto-reply is a fixed script. Someone writes the rules in advance: if the message contains "price", send the price list; if it arrives after closing, send "we are closed, we will reply tomorrow". The system matches words against those rules and nothing else. It is predictable and cheap, and it falls over the moment a customer phrases something in a way nobody planned for.
An AI chatbot reads the whole message, works out what the person is actually asking, and writes a reply from the information the business has given it: opening hours, prices, delivery areas, the return policy, the product list. Nobody has to predict the wording. "Ada stok tak size M?", "is the black one still available" and a half-typed "M ada lagi?" sent at midnight are three different sentences with one answer, and the chatbot finds it in the same facts each time.
That matters because of how Malaysian customers really write. Messages are short, mixed-language and lean on context: "boleh COD area Shah Alam?", "harga sama macam hari tu ke?", "slot petang ada?". A keyword list never covers that range. A model that reads the sentence can. The trade-off is control: a script only ever says what you wrote, while a model composes its own sentence, so how well it is kept inside your information becomes the question that decides whether it is any good.
Chatbot or AI agent: where the line is
The two words are used loosely, and sellers tend to pick whichever sounds newer. A workable line: a chatbot talks, an agent talks and also does things. An agent can check a booking diary and put a customer into a free slot, save a name and phone number as a lead, label a conversation as a complaint, or stop and bring a person in when the question is beyond it.
In practice they overlap. Many products sold as chatbots can take at least one action, and an agent spends most of its day simply answering questions. So do not pay for the label. Ask what it can actually do in a conversation, and what information it uses to decide.
When aiBalas says "agent" it means the second kind: one assistant for your business, or for one branch or brand, that answers customers from what you told it and, on the Agent plan, can book, record details and hand a conversation to your team. This guide still says "chatbot" most of the time, because that is the word most people search for.
What a Malaysian business should expect it to do
Handle the languages your customers use. A customer may open in Malay, switch to English halfway, and add "boleh tak", "lah" or "can or not" without thinking. A decent AI chatbot replies in the language the customer wrote in and copes with Manglish without asking them to rephrase.
Cover the hours you are not there. For many small shops and clinics, the messages that sit unanswered are the ones at eleven at night and on Sunday afternoon: "esok buka pukul berapa?", "ada slot Sabtu?", "masih ada stok?". Those routine questions are exactly what a chatbot should clear, so the morning starts with orders and bookings to confirm instead of a pile of the same five questions.
Stay inside what you told it. When someone asks about something your information does not cover, such as a discount you never offered, an item you do not stock or a delivery date nobody promised, the right behaviour is to say it does not have that detail and offer to get a person. Inventing a helpful-sounding answer is the failure that costs you a customer.
Hand over cleanly. Complaints, refunds, big or custom orders and anyone who is clearly upset should reach a human quickly. You should be able to read every conversation, step in at any point, and have the chatbot stay quiet while you are talking.
Meet customers where they already write. For most Malaysian businesses that is WhatsApp first, then a chat box on the website for people browsing, and Telegram for some audiences. It should be one setup answering on all of them, not a separate chatbot to maintain per channel; an aiBalas agent, for example, answers on WhatsApp, Telegram and a website chat from the same information.
How to choose: the checklist
Before comparing prices, put every option through the same eight questions. If a seller cannot give you a clear answer to one of them, that tells you what you need to know.
- How does it connect? For WhatsApp, ask whether it links your existing number by scanning a QR code or goes through Meta's official Cloud API, and what each route costs and risks. Ask which other channels it covers and whether one setup serves them all.
- How is it billed? Per message, per conversation, per staff seat, per channel, or a monthly allowance. Work out what your busiest month would cost, not the quietest one.
- Where do its answers come from? You should be able to see and edit the information it uses: hours, prices, policies, products. If you cannot see what it knows, you cannot tell why it said something.
- What does it do when it does not know? Ask for a demo using a question your information does not cover. It should admit the gap and offer a person, not guess.
- Which languages does it handle well? Test it with your own real messages, in Malay, English and the mixed way your customers actually type.
- Who can take over, and how? Check that your staff can read every conversation, reply themselves, and that the chatbot stays out of the way while they do.
- What happens to your customers' data? Ask where conversations are stored, who can read them, whether you can export or delete them, and how that fits your duties under Malaysia's personal data protection law.
- How do you cap the cost? There should be a limit you control, a warning before you reach it, and a plain answer to what happens when it runs out.
What it cannot do
It only knows what you give it. If your price list is out of date, its answers will be out of date too, and it will say them confidently. The quality of the replies is mostly the quality of the information behind them, which is why the real setup work is writing, not configuring.
It can be wrong. Each reply is written fresh, and now and then it will misread a question or put something badly. Read its conversations regularly, especially in the first few weeks, and when you find a gap, fix the information rather than hoping it will not come up again.
Sensitive cases need a person every time: a health question beyond the services you list, a payment dispute, anything legal, a customer who is angry. Set it up to pass those on rather than handle them, and make sure someone on your side actually picks them up.
It is not a substitute for a phone call when something has gone wrong. A customer whose order arrived broken wants to hear that a person is sorting it out. The chatbot can collect the order number and the photos and pass them along; the apology and the fix should come from you.
Getting started without a big project
You do not need a consultant or a month of meetings. One ordinary week is enough to find out whether it works for your business.
First, write down what you answer every day: opening hours and public holiday closures, prices or price ranges, delivery or service areas, how customers pay, your return or cancellation policy, and the questions you are tired of typing. Plain sentences are fine. This is what the chatbot will answer from, so the hour you spend here matters more than any setting.
Next, test it privately before any customer sees it. Ask it what your customers ask, the way they ask it, including the awkward questions and the ones with typos. When it answers badly, the fix is almost always in your information, not in the settings.
Then switch on one channel, usually the one where most enquiries already arrive. Leave the others until that one runs smoothly.
For the rest of the week, read every conversation it has. You will find questions you never thought to write down; add the answers. Decide which kinds of conversation should come to a person, and check that they really do.