Blog
The chatbot that answers instantly, and gets the same thing wrong every time
Most e-commerce service problems are not about response speed. They are about answers with no real data behind them.
Published
The assumption that sinks most chatbot projects is that the problem in customer service is how fast you reply.
Look at what an online store is actually asked. Where is my order. When will it arrive. Can I change the size. When does the refund land.
Every one of those can only be answered if the system knows what this customer ordered, where the parcel is, and what the store's policy is for this case.
A chatbot not connected to any of that will answer very quickly and miss every time — which is worse than waiting ten minutes for a real answer.
The order to do this in
Connect the data first. Before automation, the system needs the order status, the carrier's tracking, and this customer's contact history. With those in place, even simple rules answer most questions.
Answer only what is certain. The system should reply when confident and hand over when not. The wrong design is to have it attempt everything, which produces answers that sound confident and are wrong — and a wrong answer about a refund or a delivery date does more damage than no answer.
Hand over with context. When it goes to a person, the conversation and the order details go with it. A customer made to explain everything again to a second person is a customer angrier than when they started.
What modern language models genuinely changed
A few years ago chatbots relied on keyword matching, which collapsed the moment a customer made a typo, wrote colloquially, or mixed Thai and English in one sentence.
Current models handle Thai as people actually type it far better, typos and abbreviations included. That is real progress and worth using.
But language ability is not knowledge of facts. A model that understands the question and has no access to the order will invent an answer confidently, which is more dangerous than saying it does not know.
The right design has the model understand the question and phrase the reply, while every fact in that reply comes from the system rather than from the model.
Metrics worth using
Automation rate looks good in a report and says very little.
What means something: the share of conversations that end without the customer asking again, the time until the problem is actually resolved, and how often the same customer comes back about the same thing.
A system that closes many conversations while customers return the next day is one that improves the numbers and damages the business.
What often beats a chatbot
In many stores, reducing the questions beats answering them faster.
A tracking page customers can check themselves, a proactive message when a parcel is delayed, and a returns policy written to be understood, cut contact volume substantially with no automation at all.
What GIPSIC does here
We build websites and e-commerce systems, including integration with back-office systems, carriers and payment gateways. On one, health package sales rose over the first three months after the relaunch.
We will say plainly if we think the first thing to build is not a chatbot but a way for customers to see their own status without asking anyone.
If you would like to talk about your store, get in touch.
About the author
Film — Wisit. A businessman who still does his own BA work more often than he probably should, and writes a fair bit of code, front and back. Runs two or three small businesses. Follows technology and business obsessively, in Thailand and everywhere else. Off the clock: physics, astronomy, DIY, and anything to do with networks. Music always on, though he cannot sing. Plays instruments anyway, badly. Plays a lot of sport, racket sports above all. Not much of a traveller by himself, but happy to take Mint anywhere in the world. A man who fears — sorry, loves — his wife. One flaw: he barely touches video games.
Written with Claude Opus 5