Skip to content
An executive’s desk at night, with a phone, an open planner and the city lit up in the window.

Case 03 of 07 · Fourfecta, a large education group

No promise slips through.

Decisions and commitments were scattered across emails, meetings and messages. A searchable memory brings these sources together and prepares context for the next meeting.

AI-generated photo.

Scene · 09:48

What happens at 09:48

Ten minutes before the meeting, the executive asks on WhatsApp what was agreed, and the briefing is already there, with the open commitments.

Copilotonline
  1. Customer: I have the pilot meeting at 10. What did we agree last time?09:48
  2. Company: Pilot in one location, report every Monday. Two commitments still open; the full briefing is in the app.09:48
  3. Customer: Who was supposed to name the owner?09:49
  4. Company: You, by Thursday. The revised timeline went out on Tuesday.09:49
Meeting briefingThursday, 10 AM
  • Email
  • WhatsApp
  • Calendar
  • Previous minutes

Meeting on the customer service pilot

From the last meeting

The pilot starts with one office. Weekly report on Mondays.

Open commitments

Name an owner for the pilotdue Thursday

Send the revised timelinedone Tuesday

Sources: 1 email · 3 messages · 1 set of minutes

Illustrative scene, with synthetic data.

In production

What we built

  • Email, WhatsApp, calendar and meetings in one searchable memory, with priorities, promises and deadlines detected automatically

  • A single agent, in the app and on WhatsApp, with the same memory, voice and transcription

  • AI proposes, policy decides and a person reviews in a queue

    Tasks create themselves.

  • A briefing before every meeting, with the decisions from the last one

  • The memory opened to Claude over MCP, with secure login

  • Data isolated per client and encrypted keys

Architectureat a glance
  1. Email, WhatsApp, calendar and meetings
  2. Searchable memory with priorities and promises
  3. A policy that decides what the AI may execute
  4. Review queue, tasks and a briefing before every meeting

Governance

Where a person decides

A person in control

  • The AI proposes, company policy decides what can be executed and a person reviews it in a queue.

Risks handled

  • Data isolated per client and encrypted keys.
  • Claude reaches the memory over MCP, with secure login.

Technology · Hybrid search with reranking, MCP server with OAuth 2.0, voice transcription

40-minute call, free

Want something like this?

On the 40-minute call, free of charge, we show you how a system like this would work in your operation.