The first week of a diagnosis is not for drawing the final architecture. It is for seeing how the company actually operates: where work enters, where context is lost and who closes exceptions. Without that picture, any AI layer is designed on a polite version of the process.

What the first days are for

This is not about interviewing everyone or inventorying every tool with equal weight. It is about following a few real cases from the moment they appear until someone calls them closed.

Onezix Systems uses that week to locate repeatable friction: messages without an owner, documents nobody can find, updates done twice and decisions that only live in a thread.

Channels and queues

WhatsApp, email, forms and any shared mailbox should be seen as queues, not as apps. Who watches them, on what schedule, what counts as urgent and what is left for “when we can”.

If two channels handle the same case without a common identifier, the diagnosis already has a finding: context splits before any AI is involved.

Tools and copies of the truth

The tool list matters less than the map of copies. Where the record is created, where it is truly updated and where it is only reflected. CRM, sheets, tasks and chats often contradict each other without anyone deciding that.

The first week is enough to mark, per process, a provisional source of truth. If the team cannot agree, that disagreement is the data. It is not resolved by picking the newest tool.

Documents and living knowledge

Open Drive, Notion or whichever repository the company uses and ask: is this used, or archived by habit? A procedure with no review date and a procedure the team quotes from memory do not weigh the same.

It also matters which knowledge is not written down: exceptional pricing, sensitive clients, shortcuts only one person knows. That bounds what a knowledge layer will be able to answer at the start.

Owners, exceptions and what stays untouched

Every process has a real owner, even when the org chart says otherwise. It is the person who receives the exception and who suffers if the case is left half-done. The diagnosis names that person or role.

Just as important is the list of what is not automated in the first deployment: payments, especially sensitive data, open conflicts, any step without a rule. That list protects the perimeter.

  • Three or four real cases followed end to end.
  • Intake channels and urgency criteria.
  • A source of truth per process, even if provisional.
  • Current documents versus abandoned documents.
  • The owner of the exception and steps left outside the first scope.

What a Onezix Systems diagnosis delivers

At the end of that first reading, the company should see a sober map: where noise lives, where context is split and which Onezix OS module makes sense first — Inbox, Knowledge, Workflows or a small combination.

It is not a catalogue of promises. It is a perimeter proposal and a next step, ready for infrastructure to be designed on top of the tools that already exist.