Renata Cherniavska
← All work AI chatbots · Psychologist

A bot that knows when to call a human

An AI system for a psychologist. It answers questions, sells products and books sessions, and it passes sensitive questions to a person.

Client
Practising psychologist
My role
System design, build in ManyChat and Make, integrations
Stack
ManyChat · OpenAI · Make · Stripe · Google Sheets

the problem

Too much time went to direct messages

The expert spent a lot of time on manual chats in direct messages and on routine tasks.

The goal was a “digital office”: warm up the audience, answer difficult questions, sell digital products and manage the schedule, without the expert being there all the time.

what I built

One system around the chat

I mapped the whole system in Miro first. Then I built it in ManyChat and Make and connected the other tools.

  1. 1Recognise the client. A keyword funnel recognises the client, qualifies them and makes a personal offer.
  2. 2Answer and diagnose. OpenAI answers from the expert's knowledge base. An AI diagnostic reads the client's answers and gives personal recommendations.
  3. 3Call a human. Complex, unusual and sensitive questions go to a real person.
  4. 4Sell and deliver. After a Stripe payment, the PDF is sent right away. For physical products, the courier gets a notification.
  5. 5Book a session. The bot checks free slots in real time, holds the slot and sends a reminder 24 hours before, to reduce no-shows.
  6. 6Save the data. Leads, orders and diagnostic data go to Google Sheets for reporting.
1Keyword funnelRecognises and qualifies the client
2AI consultant and diagnosticAnswers from the expert's knowledge base
3Complex or sensitive?Unusual questions too
YES →A real person answersThe chat goes to a person
4Sell and deliverStripe, then PDF or courier
5Book a sessionFree slots, reminder 24 h before
6 · Google SheetsLeads, orders and diagnostic data
The system, simplified.

In Make and ManyChat · click to enlarge

Three of the Make scenarios behind the system.
The ManyChat automations: keywords, FAQ, reminders, booking and delivery.

what I took away

AI works best when it supports a clear process. As a separate chatbot, it does much less.

the result

What changed

~40% less manual work

On processing requests.

~10 minutes

First response time.

+40% engagement

After the personal AI diagnostic was added.

Open when the expert is offline

The bot still sells and books.

Let's talk Write to me