Anyone who serves customers every day already has the best answers: how to explain a deadline, how to handle a “let me think about it”, how to answer a frequent question. ChatSense’s team learning turns that into material for the AI agent, without anyone rewriting everything from scratch and without anything reaching a customer without human approval.
What ChatSense learns, and from where
ChatSense learns from the replies your team gives, including those sent through the WhatsApp Business app and imported history. It works whether the AI agent is active, paused or off.
It starts with the owner’s LGPD consent
The feature only turns on with the consent of the owner or an administrator. When opening the Learning area (web, iPhone or Android), the person reads a short data-protection text, the “Assisted learning” annex of the DPA, and accepts. ChatSense records who accepted, when, from which IP and which version of the text. Without consent, nothing is read.
- Before reading, phone numbers, CPF/CNPJ, e-mail addresses, postal codes and order numbers are stripped from the text.
- Reading is done by AI providers (some outside Brazil) that do not use the text to train models.
- Nothing your team writes is used for another company.
- The owner can turn it off whenever they want: reading stops immediately and anything not approved is deleted.
The full annex is at chatsense.app/legal/dpa/aprendizado-assistido, with a notice template the owner can share with the team.
Human review: nothing reaches a customer without approval
Everything that comes from the team goes into a single review queue. There the system groups similar questions, shows how many times they appeared and how confident it is in the answer, and flags when the answer differs from what is in the knowledge base.
When approving, the owner chooses how it is used: ready answer, knowledge-base entry (rewritten, without duplicates) or tone example. Everything can be undone. Answers containing price, deadline or schedule go back to review every 90 days, and customer data such as phone, ID number, order and address is never learned.
How it helps day to day
- In the inbox, when a customer asks something the team has already answered and the owner approved, the answer appears for the team to use. It only fills in the field; nothing is sent on its own.
- Every week the owner gets a summary on WhatsApp, in the app and in the bell: what was learned, what awaits review and what differs from the knowledge base.
- When the AI agent is activated, it already knows what the team approved, and it never replies over a person who is handling the conversation.
Frequently asked questions
Does the AI answer on its own what my team answered?
No. What comes from the team becomes a suggestion in a review queue, and only what a person approves is used. In the inbox, the suggestion just fills in the field for the team.
Do I have to accept anything to turn it on?
Yes. The owner or an administrator reads and accepts the “Assisted learning” annex of the DPA. Without consent, nothing is read.
What if I change my mind?
The owner can turn it off whenever they want. Reading stops immediately and anything not approved is deleted. Unapproved suggestions are also deleted after 90 days.
Is my customers’ data used?
Phone numbers, CPF/CNPJ, e-mail addresses, postal codes and order numbers are stripped before reading, and customer data is never learned. If a customer asks for their data to be deleted, what was learned from their conversations is removed too.
Go further: see how ChatSense works for clinics, e-commerce; compare it with other tools on the comparison page and check the plans and pricing.
Want to see this working in your business? Book a 30-minute demonstration with the ChatSense team.