AI Team Lead: a training simulator for chatters inside Telegram

The simulator has four customer types: a newcomer, a big spender, a fan who has gone cold and a difficult one. Each needs a different approach, so the chatter learns to adjust tactics to the person instead of repeating one script.

The personalities are based on real chats, so the customer's reactions are close to what the chatter will run into in live conversations.

The chatter can't see how much the customer is willing to spend or what they're interested in. They have to find out in conversation, the same as with a real person. If the conversation doesn't go well, the customer may just leave without a word.

If the chatter gets it right, the customer buys, haggles or leaves a tip. That way the chatter sees a direct link between what they write and the revenue.

After each training session, the bot sends a review: a score on six criteria, response speed and revenue for the conversation. Quotes from the chat sit next to the scores, so the chatter can see which reply worked and which one killed the sale.

The review ends with advice on what to change next time. Scoring is the same for everyone and doesn't depend on the reviewer's mood.

The team lead sees the team ranking and each chatter's weak spot. They can give an assignment with a deadline and a minimum score: for example, complete a session with a difficult customer and reach the required result.

For a fair comparison, the whole team can get the same customer. Then the difference in scores shows a difference in skill, not luck with the customer. Over the course, chatter skills improved by 60%.

The chatter opens the bot and starts a conversation with a customer of the chosen type. From there it works like a normal shift: they write, the customer replies, haggles or loses interest. At the end comes a review with a score and advice, and the result goes into the team ranking. This case page has a short video showing how the bot plays the customer and reviews the conversation.