Accounting automation

Automation in accounting ranges from features already in the ERP (recurring journals, auto-reversing accruals, bank-feed matching rules, three-way matching) through workflow and reconciliation tools, scripts and queries, and software robots that copy what a person does on screen, to AI-assisted tools that read invoices or suggest coding. The question isn't which tool, but which tasks — and what controls keep the automated version reliable.

Good candidates are high-volume, rules-based, stable and built on structured digital data: matching bank lines to receipts, recurring and allocation journals, prepayment and depreciation schedules, intercompany recharges, invoice capture and matching, reconciliation matching, and producing standard reports and variance flags.

Poor candidates are tasks that depend on judgment (provisions, impairment, complex revenue contracts), are low volume, change often, or run on poor-quality inputs. A messy process automated is a messy process running faster: fix the process and data first.

What automation needs to be trustworthy - Design and documentation — rules, data sources, outputs and exception handling written down and approved by the process owner. - Testing before go-live — with real historical data, including the awkward cases (partial payments, duplicates, credit notes), with results compared against the manual answer; often a period of running both in parallel. - Change control — rule and code changes are requested, tested, approved and versioned; the people who can change the automation aren't the people who rely on its output. - Access — the automation runs under its own service account with only the rights it needs, never under a person's login, and every action is logged. - Input completeness — record counts and control totals: did the bot process every line in the bank file? - Exception queues — items the rules can't handle go to a named owner with a deadline; an exception queue nobody works is an unrecorded error. - Output review and monitoring — reasonableness checks and thresholds on what's posted, alerts when a run fails, and periodic re-validation that the rules still fit the business.

The business case. Estimate hours saved (volume × time per item × automation rate, less the time to work exceptions and review output), value them, and compare with build and running costs to get a payback period. Count quality too: fewer errors, faster close, better audit trail.