Human Oversight
Part of RPA bot operations
Planning manual processing when a bot stops
Define when to switch to people, classify unfinished bot work, preserve controls and reconcile records before automation resumes.
A manual fallback needs a switch rule, authorised workers, an item register and a reconciliation step. Define them before a bot stops so staff can identify work already done, work still waiting and work whose outcome is uncertain.
Decide when to switch
Set the switch decision against the business deadline and the bot's realistic recovery window. Name the process owner and deputy who can authorise manual processing. List tasks people may complete and tasks that must wait for a specialist or the application to recover.
Australian Government continuity guidance recommends identifying important services and backup arrangements. For a bot-dependent process, the organisation must decide the manual route, approvals and records that fit its work.
Stop and classify affected work
When continued processing would create uncertainty, prevent new starts through the authorised platform procedure. Check whether a job is still running before stopping it. A stop instruction does not by itself establish that an in-progress business action was rolled back.
Capture the stop time, last known good item, running jobs, queued items and unclear results.
Item state / Handling
- Confirmed incomplete
- Use the authorised manual route
- Confirmed complete
- Retain evidence and do not repeat
- Uncertain or partly completed
- Hold for a target-system check and owner decision
- Waiting for approval
- Preserve the ordinary approval requirement
Use a unique business reference for every item. An error may occur after a record was saved, so do not treat a retry as harmless.
Run the manual route
Give workers the required input, approval, target system, steps, completion evidence and exception contact. Have them use their own authorised identities.
Record who handled each item, when, what changed and where the result can be checked. Escalate any item whose required control cannot be preserved manually.
Keep the register available during the expected outage type. Limit access and avoid copying more personal or financial information than the task requires.
Assign one person to control claims on the work queue so two workers do not take the same item.
Manual Fallback Process Steps
- Authorise and AssignIdentify authorised workers and assign tasks using unique business references
- Provide Inputs and ControlsSupply required input, target system access, steps, and completion evidence
- Record ActionsLog who handled each item, when, what changed, and result location
- Escalate ExceptionsRaise issues where controls cannot be preserved manually
- Secure AccessLimit access to the register; avoid unnecessary copying of personal or financial data
Reconcile before resuming
Compare the manual register, bot queue and target-system records by business reference. Classify each item as complete, incomplete or still uncertain.
Hold or remove manually completed items from automated input using the approved platform method. Run a small controlled set of remaining items and verify their business outcomes before restoring normal volume.
Record unresolved exceptions and who authorised the restart. Rehearse the plan with safe records covering a failure before a write, a partial write and a manually completed item still queued for the bot.
Reconciliation Before Resuming Automation
- Compare records by business referenceManual register vs bot queue vs target-system records
- Classify each itemComplete, incomplete, or still uncertain
- Handle completed itemsRemove manually completed items from automated input using approved method
- Test controlled setRun small number of remaining items and verify outcomes
- Record exceptionsNote unresolved issues and who authorised restart
