The AI Drafts the Reply Before You Type a Letter
The moment a customer message lands, the AI drafts three separate replies. It reads the intent, summarises the conversation and pulls from the customer card and your knowledge base. Your team picks one, approves it, and sends. The feedback you give teaches the tool your way of writing over time.
When a customer message arrives, the AI drafts a reply in the inbox from the conversation summary, the intent and your knowledge base. Your team picks and approves one. The AI that writes the suggestion never messages the customer on its own. Your feedback brings the drafts closer to your tone.
This page is about the reply draft in the inbox. The suggested action in the Intent AI drawer is a separate job: there the AI tells you what to do and shows its reasoning, and you copy the text yourself.
Zeynep Y.
Hi, how much is your implant price and what do I need for an appointment?
14:23
AI suggestion · Intent: price and appointment question
The reply is ready the moment the message arrives
Your team picks with one tap and the answer goes out. Look at this example:
Conversation summary & intent
Zeynep asked for the implant price and what an appointment needs. She has the price range, wants to pay in instalments and is undecided on the exam day. Tone stays positive.
Intent
APPOINTMENT
92%
Objection
PRICE
Medium
Urgency
HIGH
Deciding today
Satisfaction
8.4/10
Positive
⚡ Suggested next action
Share the instalment options + offer available exam days → appointment
Three ready replies · conversation summary · intent reading
The AI weighs every incoming message against your knowledge base, sector patterns and the customer's history. It drafts three tones at once (formal, short, friendly). Your team taps to send or edits first. This page covers the suggestion side for agents; in the flows you build, the bot does send without your approval. Conversation summary, intent score and the suggested next step sit in the same panel.
- Three drafts at once: formal · short · friendly
- Conversation summary: catch the context of a long thread in seconds
- Intent score: purchase, question, complaint, return or undecided
Automatic tags & an instant customer card
The AI reads the conversation and applies the tag itself (price question, campaign, undecided, follow-up needed). The conversation is matched to the right customer record. That person's card and earlier threads stay on the right side of the panel. Your team does not have to look for the history somewhere else.
- Smart tags: topic, intent, customer profile and urgency together
- Customer card and earlier threads stay open on the same screen
- The agent who takes over reads the summary instead of the whole thread
Customer card (auto-matched)
AUTOMATIC TAGS
CONVERSATION HISTORY
3rd conversation · Previous topic: check-up appointment · Preferred channel: WhatsApp
Feedback learning, this month
Sample panel142 approved · 8 edited · 0 rejected. Agent tone learned.
94 approved · 12 edited. Instalment table template now added automatically.
38 approved · 22 edited. AI apology tone too formal; learning a warmer version.
📚 Learned this month
- • "Hocam" form of address (Turkish friendly tone)
- • 5 new KVKK (Turkish data protection law) phrases
- • Delivery times within Istanbul
Learns from feedback and sounds more like you over time
When your team approves, edits or rejects a suggestion, the AI learns from it. Over time it picks up your sector's terms, the way you talk to customers and your team's tone. Which suggestion was approved and where it was corrected counts as feedback. The drafts follow the way your team writes.
- Keeps learning from approve, edit and reject signals
- Sector-specific vocabulary grows on its own
- Accuracy is tracked; the acceptance rate shows in the report
Knowledge base integration (RAG)
Load your PDF, Word or web page content into the system. When a customer asks something, the AI drafts from the documents you uploaded. Your team sees which document the draft rests on. The source feeds the suggestion, it does not limit it. The agent is the one who catches a wrong suggestion before it goes out.
- Quotes directly from the documents you uploaded
- The source document is listed next to the draft; approval stays with the agent
- Cross-language search: a question asked in Turkish is found in an English document
Customer question
"How many days do I have to return? Do I pay the shipping?"
Draft from the KB, the agent approves
Our return period is 14 days. For unused items with tags still attached, we cover the return shipping. For damaged or used items the shipping cost is paid by the customer.
Give your team a higher gear
Instead of typing into a blank box, your agent edits a ready draft and sends it. A new hire finds the clinic's saved replies and knowledge base in front of them from day one.
How to do it
Step-by-step guides from the help center for this feature:
