Automation You Can Leave Alone, Because It Tells You When It Stops

• 50+ scheduled tasks run the recurring work, from amortisation, depreciation and loan payments to recurring invoices, journals and bank-feed updates, without anyone having to remember.

Prepaid amortisation, depreciation, loan payments, recurring invoices and journals, commission accruals, bank-feed updates, year-end close reconciliation. The ones that touch money are the interesting case, and everything below is about them.

• Only one copy of a money-moving task can run at a time, and each posting carries a fixed reference, so a retry or an overlap cannot post the same entry twice.

• The system itself refuses an unbalanced entry before it can be saved. The rule is built into the foundation, not into a screen that can be bypassed.

Two things have to be true for a scheduled task to be safe when the system retries it. Only one copy of it may run at a time, so a slow run and its retry cannot overlap. And what it posts must be identified by the work itself rather than by the attempt, so that a second attempt is recognised as the same posting instead of becoming a duplicate.

Both are built into how every task is made, not habits each task has to remember. That distinction is the whole point: a convention holds until the twentieth task.

• Automation proposes and a person decides: scheduled work lands as a draft for approval unless you have deliberately switched on auto-posting, company by company.

By default the schedules post nothing on their own — work arrives as a draft for a person to approve, company by company. Auto-posting is available, and switching it on is a deliberate act, per company, by somebody with the authority to do it. Nothing becomes automatic because it was convenient at the time.

A standing authorisation is also re-asked rather than remembered: every time the task runs, it checks that the person who switched it on still exists, is not suspended, and still holds that authority. An approver who leaves in June does not keep approving in December.

• All 15+ of the automatic tasks that post to the books are watched by a check that notices silence: one that stops running raises an alert, because a task that quietly stops is the failure nobody sees.

A scheduled task that fails loudly is a good day: something is red and somebody looks. A scheduled task that simply stops running is the bad one — the software keeps working, every screen looks normal, and the numbers are quietly a month stale. The silence check exists for exactly that case: it asks whether each money-moving task has reported a run inside its own window, and alerts on the ones that have gone quiet.