An AI trained on your own sales team.
Your best salespeople have a voice customers trust — the greeting, the mix of English and Chinese, the emoji, the way they close. WhatsTeam’s AI twin learns that voice from their real WhatsApp conversations and drafts the next reply the way they would write it. A person always reviews and sends.
- Learns from your own chats
- A human reviews every draft
- Never sends by itself
From your chat history to a working twin
No scripts to write and no bot flows to design. The twin is trained on what your team has already typed — thousands of real replies that won real orders.
- Step 1
Connect your numbers
Each connected WhatsApp number gets its own profile, so Amy’s twin writes like Amy and Ken’s twin writes like Ken — not a company average.
- Step 2
Analyze real conversations
One click reads that number’s past chats and studies how the salesperson actually writes — with the conversations that closed deals weighted most heavily, so the twin learns the moves that win orders.
- Step 3
Or train a shared Team profile
Upload an export of closed-deal conversations from a previous platform and train one company-wide Team profile. Any number without its own profile falls back to it automatically — useful from day one.
- Step 4
Every run becomes a version
Training never overwrites anything. Admins see the full history of training runs, open and compare past versions, and pin the one the team should use.
- Step 5
One tap drafts the reply
In the inbox, one tap drafts the next reply in that salesperson’s voice. If a frequently sent file clearly answers the customer — a price list, a catalog — the draft can suggest attaching it.
- Step 6
A person reviews and sends
The draft appears in the reply box, not in the customer’s chat. Your teammate edits it, or sends it as is, or ignores it. Nothing leaves without a human decision.
What a draft gets right
A generic chatbot answers like a manual. A twin answers like the person your customer has been talking to all along.
The language mix
If your rep switches between English and Chinese mid-sentence, the draft does too — in the same conversation’s language, not a translation of it.
The signature phrases
Training extracts the greetings, expressions and closing lines each person actually uses, taken verbatim from their own chats.
The small habits
Message length, punctuation, emoji habits, how they follow up when a customer goes quiet — the details that make a reply feel like a person.
The closing instinct
From deal-closed conversations the twin learns how each rep handles price pushback and asks for the order, and applies it when the moment fits.
Drafts, not decisions.
The AI twin is an assistant for your team, not a replacement for it. These guardrails are built into how it works.
It never sends by itself
There is no auto-reply. A draft only becomes a message when a member of your team reviews it and presses send. You are always talking to your customer, with better first drafts.
It never invents prices or promises
The drafting rules forbid making up prices, discounts, stock levels or delivery dates that aren’t already in the conversation. When it can’t know, it drafts the way your rep would: “let me check and get back to you.”
Customer identities stay out of training
Transcripts reach the model labelled only as the two sides of a chat, with no customer name or phone-number fields attached. Uploaded closed-deal exports go further: customer names and phone numbers are scrubbed out of the message text before any model call.
Admins hold the keys
Only admins train profiles, review version history and pin the active version. If a training run isn’t an improvement, pin the previous version and carry on.
Honest by design: drafts are generated by third-party AI models through our AI routing provider. AI output can be wrong or incomplete, so a human must review every draft before sending — and your team stays responsible for every message actually sent. We don’t claim guaranteed accuracy, and neither should anyone.
Common questions
- Will my customers be talking to a bot?
- No. The twin only writes drafts inside your team’s inbox. Nothing reaches a customer until a person reviews it and sends it, so every message your customers receive was approved by a human.
- What does it need in order to learn?
- The past WhatsApp conversations of each connected number — the ones that closed deals matter most. If you’re new and have little history here, upload a closed-deal export from your previous platform and train the shared Team profile instead.
- What about a brand-new number with no history?
- It automatically uses the company-wide Team profile until it has a profile of its own, so drafting works from the first day.
- What if a training run makes it worse?
- Every run is kept as a separate version. Admins compare versions side by side and pin the one that should be active — a bad run never overwrites a good one.
- Can it quote the wrong price?
- It is explicitly instructed never to invent prices, discounts, stock or delivery commitments, and a human reviews every draft before it is sent. AI can still make mistakes — that is exactly why the review step is not optional.
Give every rep their best first draft.
Connect your numbers, run the first analysis, and watch drafts arrive in your team’s own voice — reviewed by your people, sent by your people.