Skip to content
Business-Gym Docs

The inbound triage cockpit

The inbound triage cockpit is the daily queue an operator clears — draft replies to approve or edit, categories to confirm or correct, notes to leave on the items that do not fit. It is deliberately not the CEO’s screen. It is the seat for whoever handles inbound work every day.

The queue is also where the system gets its training data. Every click you make in it is a labelled example.

Who sits in the cockpit?

Section titled “Who sits in the cockpit?”

The operator, not the CEO. The people who benefit from a summary dashboard and the people who clear a decisions queue are different people doing different jobs, and this queue is built for the second one.

That matters for how you work it. You are not reviewing a sample to check the system is behaving. You are expected to take the queue to the bottom.

What is in the queue when you open it?

Section titled “What is in the queue when you open it?”

Items pulled in from shared mailboxes over Microsoft Graph — eleven or more of them, each with an owner Head that the item is routed to.

By the time an item reaches you, four things have already happened, and all four are live:

  • it has been ingested from its mailbox
  • it has been classified — what kind of item it is, how urgent it looks
  • it has been routed to the Head that owns that mailbox
  • its sender has been resolved to a known person or account

So the queue is not a raw inbox. It is a sorted pile with a proposed next step attached to each item.

How do you get through a queue fast?

Section titled “How do you get through a queue fast?”

Work in one pass, in this order, per item:

  1. Check the category before the content. A misrouted item is a wrong draft no matter how good the wording is. Correct the category first, then judge what it proposes.
  2. Edit rather than rewrite. An edited draft is a much more useful example than a draft you deleted and replaced from scratch. Keep the parts that were right.
  3. Leave a note when you override. The note is what makes an unusual decision readable later — to a colleague, and to the system.
  4. Do not park items you could finish. A skipped item comes back and costs you the reading time twice.

Why does every click matter?

Section titled “Why does every click matter?”

Because every click is a labelled example. Approving a draft, editing it, correcting a category, adding a note — each of those is a recorded judgment about what the right answer was for that item.

This is the part of the job that is easy to do carelessly and expensive to undo. A category you waved through because it was close enough is a wrong label, and a wrong label is worth less than no label.

What if an item needs more than one step?

Section titled “What if an item needs more than one step?”

It comes back to you. Today a proposed action is a single step — draft this reply, set this category — not a chain the system works through on your behalf.

Matching an item to a playbook, the reusable “when this arrives, do these things” recipe that would turn single steps into sequences, is largely not built yet. So plan your session around single decisions. When a case needs three things done in order, you will see it three times.

What happens when a connector is not connected?

Section titled “What happens when a connector is not connected?”

The step says so plainly. You get a “not connected” state, and the item queues until the connector is configured. It is never reported as done.

That is a deliberate choice: a fabricated success is worse than no automation, because you find out weeks later that nothing was sent. If an item looks stuck, check its connector before you assume the queue is waiting on you.

What does a good session look like at the end?

Section titled “What does a good session look like at the end?”

There is no analytics dashboard to tell you. Operations analytics — time saved, automation rate — is not built, so do not go looking for a number that grades your session.

Judge it by the queue instead. A finished session looks like this:

  • The queue is empty, or every item still in it has a named reason for being there — waiting on a connector, waiting on a person, or genuinely undecidable today.
  • Every category you touched is one you would defend, not one you waved through.
  • Every override carries a note.
  • Nothing was marked done that did not actually happen.

If all four are true, the session was good — whatever the volume was.

Where do I go next?

Section titled “Where do I go next?”

The queue itself is documented separately, and the two pages are meant to be read together: this one is the working rhythm, that one is what each card means.