Communicating during an interruption

Two audiences, two messages#

An interruption produces two communications with different obligations. The internal one has to be specific enough to act on: which line, which mailbox, what the health verdict says, what has already been retried. The external one has to be true, brief and free of anything a customer cannot use.

Detection to all-clear, stage by stage#

1. Trigger — something stops working: a mailbox loses authorisation, a line stops accepting calls, a provider degrades, an allowance runs out.
2. Internal event — the health verdict changes. Connected is not health: a mailbox that authenticates and returns nothing is a quiet mailbox, which is a signal rather than an error.
3. Authentication and workspace resolution — the fault belongs to one workspace's line or mailbox, and is raised inside it rather than as a global banner.
4. Ingest — retries and their outcomes are recorded. A request that cannot be lost is retried through a restart or a proxy error, while a 4xx is an answer and is never retried.
5. Canonical record — work that cannot be done is held rather than dropped, and read cursors deliberately do not advance past what was not processed.
6. Classification — degraded or stopped, and for which channel. The difference decides whether customers need telling at all.
7. Knowledge and rules — what the business has authorised to be said. A closed line's spoken message is deterministic and carries no hours in it, so it cannot become wrong when hours change.
8. Autonomy and approval — a message to customers about an interruption is a business communication and follows the workspace's own rules for one.
9. Action — the internal queue entry appears immediately; any external message is deliberate rather than automatic.
10. Result — customers hear something plain, or nothing, and the business decides which. Callers on a closed line always hear a spoken refusal rather than silence.
11. Relationship and timeline — anything said to a customer lands on their record like any other message.
12. Audit, usage and Needs You — the queue entry drains by itself as the cause clears, which is what makes the queue trustworthy rather than a list somebody has to tidy.

What is held back, and why#

1. Provider error text. It carries identifiers, endpoint names and sometimes account references. It also tends to be wrong in an interesting way that invites the reader to diagnose, which is not their job.
2. Internal paths, traces and identifiers. They tell an outsider about the structure of the system and tell a customer nothing about their own problem.
3. A cause that is still a theory. A cause published early and corrected later is remembered as the cause. Say what is affected now; say why once it is known.
4. A restoration time that is really a hope. Missing an announced time turns one interruption into two conversations. Where a time can genuinely be committed, commit to it and schedule the follow-up that proves it.
5. Blame directed at a named third party. It is rarely as clear-cut as it looks during the incident, and it is not information the customer can act on.

Who is told first#

• The workspace, immediately and automatically: line health and mailbox health both raise items into the queue, and each drains by itself as its cause clears.
• The person who owns the channel, through the same queue rather than a separate alerting product, so that the fix and the evidence are in one place.
• Customers, only where the interruption affects something they are waiting for — and only after somebody has decided what to say.
• Anybody with an open commitment: a scheduled follow-up that cannot run is a promise at risk, and is more urgent than a general announcement.

Held work is the reason a short interruption often needs no customer message at all. Work refused while an allowance was spent or a mailbox was disconnected is waiting, not lost — the cursor did not advance past it, and it resumes when the cause clears.

Afterwards#

1. Confirm the queue entries have drained on their own.

   Result An entry that persists after the cause is fixed means the cause was not what you thought.

2. Check what was held rather than processed, and let it run.

   Result Held work resuming is the difference between a delay and a gap in the record.

3. Answer anybody who was told something during the interruption.

   Result An all-clear that never arrives is remembered longer than the interruption.

4. Write down what actually happened where the next person will find it.

   Result The pattern across several small interruptions is usually more informative than any single one.

Questions#

Related#

• Troubleshooting
• Mailbox health
• Phone line health
• Reviewing mailbox health as a routine
• Escalating a complaint, end to end
• The product status manifest