Accounting software automation covers a wide range, from features that genuinely run themselves to ones that are a person's judgement with a rule written on top. The distinction matters because the second kind fails silently. A bank rule that categorises a recurring payment is deterministic and safe: if it is wrong you will see the wrong category. A rule that decides which invoices are approved automatically is a judgement, and if it is wrong the consequence is money leaving with nobody having looked. This page sorts the common automations into the two categories and gives the question to ask before switching any of them on.
Automations that genuinely run themselves
Bank feed matching against known payees, recurring journals for known amounts, invoice capture reading a header, and reminders. Each is mechanical, checkable, and produces a visible result if it goes wrong.
Automations that are judgement in disguise
Auto-approving invoices below a threshold, auto-clearing matching differences within a tolerance, and any rule that writes off a small balance. These are decisions, and delegating them to a rule is a legitimate choice that should be made deliberately and reviewed, not switched on because it was offered.
The question to ask before enabling anything
If this runs incorrectly, will anyone notice? Automation with a visible result is safe to try. Automation that quietly changes or clears things is not, because by the time somebody notices the evidence may be gone. On the worked example on this site, 400 invoices at three touches of six minutes with a 12% exception rate is 134.4 hours a month, $4,300.80 at a $32 loaded rate and $10.75 an invoice.
Questions people ask about accounting software automation
Is automation risky?
The mechanical kind is not. The judgement kind carries exactly the risk the judgement was there to manage, which is fine if you decided that knowingly and review it.
Should we automate the close?
Parts of it: recurring journals, standard reconciliation matching, and the reminders. The review and sign-off are the point of the close and automating them removes the control rather than the work.
How do we review an automation?
Periodically sample what it did and check you agree. A rule that has not been sampled since it was written is a decision nobody has revisited.