Zayit

Software that shows clinics where their sales conversations break.

In Brazilian health clinics, most sales start as a WhatsApp conversation with the front desk. A patient asks about a procedure, and what the receptionist says next decides whether that patient books, shows up and buys.

Zayit builds software for that part of the clinic. Our conversation analysis system is in production. Our own CRM for clinics is in development, at prototype stage.

In production since September 2026

Conversation analysis

Reads the WhatsApp conversations between a clinic's front desk and its patients, and ties each one to what happened next.

Reviewing these conversations by hand every week is how we used to find the mistakes that cost clinics bookings. The system does the collecting, transcribing and organizing, so the analysis can be done by AI agents working over the full history instead of a sample.

  1. A conversation arrives from the clinic's WhatsApp, triggered by an event in their sales funnel.
  2. Voice messages are transcribed, so the whole conversation exists as text.
  3. The conversation is stored and linked to the contact's outcome: still talking, booked, booked and did not show up, showed up and did not buy, or bought.
  4. Agents read the conversations of a given outcome to find what went wrong and what worked.
  5. The best conversations are saved as a knowledge base that agents retrieve from when analyzing others and when writing scripts for the team.

Models. The system is model-agnostic by design: each agent is registered with its own model and its own access to the knowledge base. Claude is the model we use most.

In production

  • Event-driven intake of WhatsApp conversations
  • Voice message transcription
  • Storage linked to contact and funnel outcome
  • Model registry
  • Agent registry, with knowledge base access set per agent
  • Knowledge base of exemplary conversations
  • Investigation agent: answers questions about selected conversations, cites the messages behind each claim and says what the material does not support

Running on conversations from the clinics served by Zayit.

In progress

  • Agent for conversations that did not turn into a booking
  • Filling the knowledge base as conversations come in

Planned

  • Agents for the booking and attendance stages
  • An agent that audits front desk service quality
The investigation agent answering a question about a clinic conversation
The investigation agent answering a question about one conversation, with each claim tied to the messages it came from. The interface is in Portuguese. Patient and clinic details are covered.
In development, prototype stage

Zayit CRM

A CRM built for how clinics sell: a shared WhatsApp inbox, a sales pipeline, and follow-ups that do not depend on someone remembering.

It is multi-tenant from the first migration, with each clinic's data isolated by row-level security in the database. Most of the backend for a first usable version is built and tested, and the main screens exist. It is not yet running with clinics.

The AI work comes after the core is in use. The first piece planned is a briefing: when the front desk hands a patient over to the person who will see them, the CRM summarizes the conversation, the signals in it and the risks. The conversation analysis system above is where we are learning what those briefings need to say.

Stack. Go, PostgreSQL 16 with row-level security, Redis, River for background jobs, WebSocket for real time, S3-compatible storage. Vue 3, Tailwind and TanStack Query on the front end. WhatsApp through Evolution Go.

Built and tested

  • Multi-tenancy with row-level security
  • Authentication with rotating refresh tokens
  • Custom roles, permissions and audit log
  • Teams with automatic lead distribution and business hours
  • Several WhatsApp numbers per clinic
  • Contacts, tags, custom fields, CSV import and export
  • Pipelines, opportunities, products and per-stage follow-up sequences
  • Activities, scheduled messages and attachments
  • Real-time inbox and chat with media
  • WhatsApp reliability: durable webhook intake, idempotent messages, outbound reconciliation, connection alerts

Screens built: inbox and chat, WhatsApp channel settings, kanban, contact page.

Next

  • Activity center
  • Settings for users, teams and hours
  • Message scripts and triggers

Planned

  • Appointments and patient handoff
  • AI briefing at handoff
  • Visual automation editor
  • Dashboards and analytics
  • Self-signup and billing
  • AI assistant over won and lost deals
Zayit CRM sales pipeline screen
Sales pipeline in the prototype, with demo data. The Copilot panel and the menu items for automations, reports and campaigns are interface designs, not yet connected to the backend.
Zayit CRM inbox and chat screen
Inbox and chat in the prototype, during a test conversation. Items tagged MOCK are placeholder data.

Company

Zayit started as a marketing and sales advisory for health, aesthetics and dental clinics, and still runs that service today. The software on this page comes out of that work: the clinics we serve are where the conversation analysis system runs, and their sales process is what the CRM is designed around.

The advisory side is at zayitsolutions.com.br.

Legal name
Zayit Tecnologia Marketing e Produções Ltda
CNPJ
61.032.179/0001-53
Registered
May 2025
Location
Anápolis, Goiás, Brazil
Founder
Samuel Brito de Oliveira
Contact
samuel@zayit.cloud