From 15 Hours to 2 Hours: Automating NDIS Payroll Compliance

How we built custom automation to handle complex allowances, break compliance, and clock-in validation that standard integrations couldn't solve.

The Challenge

An NDIS provider was using ShiftCare + Xero, but their real payroll rules were too complex for the standard integration.

Complex allowances (especially active night)
Break compliance across grouped shifts for 55+ staff
Inconsistent mileage handling
Clock-in issues requiring manual investigation

The admin team spent about 15 hours/week checking shifts, fixing break records, and chasing anomalies.

The Approach

Phase 1: Streamlit checker

First, we built a Python + Streamlit checker to flag risky items and safely auto-fix deterministic break-time issues.

It flagged allowances, break compliance, and clock-in anomalies.
It auto-fixed break timings only where rules were deterministic.
Everything uncertain stayed in a human review queue.

This phase proved detection quality without risky mass updates.

Phase 2: Safe auto-approver

Once phase 1 was stable, we introduced an auto-approver with strict safety rules: auto-fix when clear, flag when uncertain.

For each selected pay period, the system:

Applies the proven break-time fixes
Applies active night allowances where eligible
Uses shift logs + Gemini 2.5 Flash Lite for ambiguous attendance cases
Auto-approves when evidence is sufficient
Flags anything uncertain for manual review

Mileage automation was deprioritised because the client's process changed and mileage became too irregular for reliable automatic handling.

New Workflow

Handle exceptions for the period.
Run the auto-approver.
Review only the flagged shifts.

This keeps humans involved where judgment is needed, not on routine checks.

The Results

Manual effort dropped from ~15 to ~2 hours per week.
Payroll quality improved and recurring errors were removed.
Compliance checks became consistent across 500+ shifts/week.
Admin effort shifted from data entry to review/oversight.

Technical Highlights

ShiftCare API integration with audited writes
Python + Streamlit checker app
Auto-approver workflow with Gemini-assisted decisioning
Shift-grouping logic for break compliance
Clock-in validation using notes/tasks/event logs

Safety & Testing

Staging tests before production rollout
Audit logs for every change and approval
Conservative thresholds (ambiguous cases are flagged, not auto-written)
Gradual rollout from low-risk to higher-impact automations

TL;DR

We started with a conservative checker, proved reliability, then expanded into a safe auto-approver. The team now spends about 2 hours/week on payroll oversight instead of 15.