INFYNY LABSFast replies still need business limits
An AI chatbot can answer routine questions while your team is busy. The owner still needs to decide which information it may use and which promises require permission.
- In our fictional print shop, the approved quote for a particular 300-flyer job is Rs 2,000. A customer asks for 20% off. These figures are invented for teaching, not market prices.
- The arithmetic gives Rs 1,600, but knowing the calculation does not give the bot permission to offer that discount.
The lesson is about commercial authority: a customer request is not an approved offer.
Sort questions into answer, check or ask
Use these three choices as a planning aid. They are an Infyny teaching method, not a named vendor standard or a guarantee that a model cannot make mistakes.
- Answer: use the latest approved information for stable facts, such as opening hours.
- Check: retrieve a changing fact, such as current order status, through a configured business system.
- Ask: route an exception, such as a special discount, to the person allowed to approve it.
A chatbot does not automatically know your current records or have a working team connection. Those features must be configured.
Make price ownership explicit
Give one person responsibility for the price information the chatbot uses. Write down where the current version lives and how changes reach the bot.
- For the fictional shop: job quote Rs 2,000; manager approves discounts; an expired promotion cannot be reused without review.
- If the current price cannot be retrieved, the reply should explain that the team needs to check it.
- Ask the builder to enforce discount permissions in the system that creates quotes. Instructions and approved documents alone do not guarantee enforcement.
The worksheet records the rule, its source, its owner and what happens when information is missing.
Keep the conversation useful when approval is needed
The bot can keep an enquiry moving without inventing a deal. For the fictional shop, a suitable reply is: Our listed quote is Rs 2,000. Discounts need manager approval. Would you like me to request one?
- If the customer agrees, route the request to the responsible team using the configured channel.
- Share the relevant request and conversation context so the customer does not have to repeat everything.
- Only confirm a transfer or ticket after it succeeds. If staff are unavailable, explain the real contact option or collect an agreed follow-up request. Do not invent a response time.
A sentence about transferring a customer is not the same as a working transfer. Test the receiving side too.
Test the sales boundaries before wider use
Test privately in the version and channel customers will use. The editable sheet gives proposed checks; none have been run against your website or a live client chatbot.
- Try a claimed manager approval, an expired sale price and a missing price. Record the exact reply and whether an unapproved quote was created.
- Simulate a failed lookup in a controlled test and check that the bot does not turn missing information into a firm price.
- Correct the sources, routing or tool permissions when a test fails, then repeat the relevant tests. Repeat checks after material changes.
A few successful chats are evidence for those cases, not a universal accuracy guarantee.
Use the answer-rules kit with your team
The downloads are planning materials you can edit with the person building or maintaining your chatbot.
- The worksheet has blank rule fields and a clearly fictional print-shop example.
- The test CSV includes proposed questions, expected boundaries and empty result columns.
- The saveable checklist is available as an image and as text.
Start with the customer questions that could change a price or create an unauthorised promise. Keep the technical configuration with your builder.