Why canned replies are not enough
The questions customers ask most are about their own data: where is my order, how much is this invoice, is this item in stock. A bot that only knows prepared text can help with little of that.
A useful bot is connected to the systems that hold that data, and answers within the asker’s permissions.
Start with the 10 most-asked questions
Export one to three months of chat history and group the questions. Usually a handful of groups make up more than half. Get the bot answering those well first, then widen it.
Connect the data the answers need
Every lookup of personal data needs the customer verified first, and the bot should say where its answer came from.
- Order and delivery status from ERP
- Outstanding balances and invoices
- Stock and prices
- Manuals, policies and warranty terms
Decide when to hand over
Be explicit about which cases need a person: returns, complaints, or any time the customer asks for one. The bot should hand over with a summary and the chat history so the customer never has to repeat themselves.
What to measure
- Share of conversations the bot resolves on its own
- Average reply time
- Share handed to people, and why
- Satisfaction score after each conversation
PDPA and chatbots
Tell customers they are talking to a bot and what their data will be used for. Keep only what you need, set how long chat history is kept, and give customers a way to ask for deletion.