iHRAustralia × prepared by Haloweave

On 1 January 2025, Australia made wage theft a crime.

The compliance engine that turns 25 years of expert judgement into code.

Yellow Canary tells an employer they have a problem. iHR fixes it. This is a working demonstration of the hybrid AI + rules engine that lets you do both — audit every pay run against Australian Modern Awards, flag the underpayments, and prove the maths.

HYBRID RULES + AI/RAG NURSES AWARD 2020 POC TARGET · AUG 2026
Begin the walkthrough
01

The market moved. The tooling didn't.

Penalties now reach 10 years' imprisonment and $7.825M fines. Demand is acute — adoption is in its infancy. Early movers with credible brands win.

Source: iHR market research, June 2026 · Fair Work Ombudsman 2024–25.

02

Detection is a feature. The fix is the business.

The market leader stops at "you have a problem." iHR's moat is the entire lifecycle — and that's what no competitor can replicate.

Yellow Canary sells

A technology platform.

  • Detection & quantification of variances
  • Dashboards and reports
  • EA interpretation by the client's own lawyers
  • Australian Awards only

Then it stops. Full stop.

03

Two engines. One verdict.

A deterministic rules engine for the top Awards — 100% auditable, zero hallucination — running in parallel with an AI/RAG engine for Enterprise Agreements, edge cases and new countries. A validation layer reconciles them and escalates disagreement to a human.

DETERMINISTIC

Rules engine

Award rules hand-coded as tested logic. The source of truth for AU, today.

  • Top 10–20 Modern Awards
  • Legally defensible, millisecond checks
  • Primary for the POC
VALIDATION LAYER

Cross-checks both engines. Agreement → trusted. Disagreement → human review. Confidence compounds over time until AI can lead.

PROBABILISTIC

AI / RAG engine

Retrieves the exact clause, reasons over it, computes the rate. Scales to anything.

  • Enterprise Agreements & rare Awards
  • New-country expansion from day one
  • Co-pilot now, primary later
  1. 01Data ingestionCSV / API / DB → canonical schema
  2. 02Legal knowledge base~122 Awards chunked, embedded, date-aware
  3. 03Compliance engineRules + AI + validation
  4. 04Analysis & reportingVariance, root cause, board packs
  5. 05Privacy & securityAU residency · ISO 27001 · SOC 2
04

Watch the engine audit a real pay run.

A faithful preview running on synthetic data — one hospital, twelve nurses, the Nurses Award 2020. Production needs 122 Awards benchmarked to 95%+ accuracy on live payroll. That gap is the build.

LIVE AUDIT SIMULATION src: stvincents_nurses_Q1-2026.csv · SYNTHETIC
05

The dashboard is 10%. This is the other 90%.

The pretty report is the easy part. Correctness lives in retrieval, date-versioning and the legal knowledge base. Two of the hard parts, made tangible.

RETRIEVAL · SIMULATED

Ask the legal knowledge base

Naïve keyword search misses clauses that don't share words. Hybrid retrieval finds the right one, date-aware, with a citation and a confidence score.

Type a question, or pick one above. The engine retrieves over the Nurses Award knowledge base.

DATE-AWARE

The same clause, two years apart

Every retrieval is date-stamped. Use last year's rate and the audit is wrong — and indefensible.

UNDER THE WATERLINE
Variance dashboard
10% visible
  • 122-Award rules library
  • Legal knowledge base + parsing
  • Date-versioned retrieval
  • Validation & human-review loop
  • AU data residency · ISO 27001 · SOC 2
06

Where iHR wins on capability.

Same audit power as the leader — plus the things only a 25-year consultancy can do.

07

Size the exposure.

Underpayments typically run 1–3% of headcount cost. Move the slider to model a client's annual exposure — and the recoverable value the engine surfaces.

Estimated annual underpayment exposure $3.51M
Criminal penalty ceiling$10.5M
Engine price @ $3.50 / emp / mo$105K
Exposure : cost33×

Illustrative. Penalty ceiling is the greater of fixed maxima or 3× the underpayment.

08

An 8-week proof of concept, with gates.

Milestone-based, with go/no-go checkpoints. You see working software at Week 4, validation results at Week 6, and an end-to-end audit on real data at Week 8.

09

Investment, by phase.

Exactly the breakdown Jacqui asked for — discovery, rules engine, RAG, AI engine, dashboard MVP, testing. Milestone-linked payments. Figures finalised on the kickoff call.

AUD, ex-GST. Milestone-based schedule. A fixed-price POC de-risks the production decision — you commit to the build only after Week 8 proves the accuracy.

10

You asked 13 questions. Here's where we land.

Before the kickoff call. The full set is in the brief — these are the ones that decide feasibility.

NEXT STEP

Let's walk through it live.

Book the kickoff call. We'll align scope, confirm the phased figures, and start the data-feasibility sprint — on track for a working POC by August.

iHR × Haloweave