LINE Chatbot Case Study: Auto-Replies That Actually Work

A retail client came to me with a problem: their LINE Official account was getting 200+ messages per day, and their team couldn't keep up. Response times were slipping, customers were getting frustrated, and they were losing sales.

Here's how we built a chatbot that handles 80% of questions automatically—and hands off the rest gracefully.

The Problem

Most messages were the same questions over and over:

• "Do you have [product] in stock?"
• "How much is shipping to [province]?"
• "What are your business hours?"
• "Can I pay with [payment method]?"
• "Where's my order?"

Staff were answering the same questions 50+ times per day. They were burned out, and complex customer issues weren't getting the attention they deserved.

The Solution

How It Works

Step 1: Intent Detection

When a message comes in, the system first tries to identify what the customer wants. Common patterns are matched immediately—"ราคา", "shipping", "สั่งซื้อ", etc.

Step 2: Knowledge Base Search

The message is searched against a database of products, FAQs, and business information. If there's a high-confidence match, that answer is used.

Step 3: AI Enhancement (if needed)

For questions that need interpretation, the AI takes the retrieved context and generates a natural response. This handles variations like "how much does the blue one cost" when the product is called "Navy Linen Shirt".

Step 4: Handoff (if needed)

If confidence is low, or the customer asks to talk to a person, the system tags the conversation for human follow-up and lets the customer know someone will respond.

The Knowledge Base

The chatbot is only as good as its data. We built a comprehensive knowledge base with:

• 200+ products: Name, price, description, stock status, categories
• 50+ FAQs: Common questions with multiple phrasings
• Business info: Hours, locations, shipping rates, payment methods
• Order tracking: Integration with their shipping system

Staff can update the knowledge base through a simple admin panel. New product? Add it. New FAQ? Add it. Changes reflect instantly.

Timeline

Results After 3 Months

Staff now spend their time on conversations that actually need a human—sales discussions, complaints, complex orders. They're less burned out and more effective.

What Didn't Work (At First)

Transparency: we had to iterate on several things.

• Initial responses were too robotic. We rewrote them to sound more natural and match the brand voice.
• Thai/English mixing caused issues. Many customers mix both in one message. Had to improve the language detection.
• Product search was too strict. "กางเกง" should find "pants" too. Added synonym mapping.
• Handoff threshold was too low. Initially escalated too many conversations. Adjusted confidence thresholds.

Costs

Should You Build One?

A LINE chatbot makes sense if:

• You get 50+ repetitive questions per day
• Staff are spending hours answering the same things
• You want 24/7 response capability
• You have structured data (products, FAQs) to pull from

It might not be worth it if:

• Your questions are mostly unique/complex
• You have low message volume (<20/day)
• Personal relationships drive your business (luxury, high-end services)