The hidden algorithm inside NSW workers’ compensation
How icare’s programming not Guidewire itself turned a digital overhaul into a long tail of remediation, and what that means for AI in public services.
The short version
Guidewire is a world-class insurance software platform used globally.
The problem in NSW workers’ compensation lies in how icare configured and governed it not in the technology itself.
The regulator-commissioned Dore Review (with EY support) found icare’s algorithmic triage missed “the subtleties of individual circumstances,” producing mis-triage and lack of dedicated case managers for some injured workers.
In 2024–25, icare issued new discovery and remediation contracts for its Guidewire environment, clear evidence that the system still requires stabilising.
Separate Guidewire builds (for other insurance schemes such as HBCF) show the technology stack expanding even as the core workers’ compensation platform continues to be remediated.
What actually went wrong — and why it isn’t Guidewire’s fault
Guidewire is not the villain here. The software itself is a proven global standard used by major insurers. What went wrong was the way icare programmed, trained, and governed the system.
Within its Guidewire-based Nominal Insurer Single Platform (NISP), icare developed an internal algorithmic triage model built on an “80/20” assumption that roughly 80 percent of claims were straightforward and could be processed automatically.
EY’s analysis (as part of the Dore Review) found that this model led to mis-triage, passive case management, and poorer return-to-work outcomes. The review cautioned that such automation “will miss the subtleties of individual circumstances.”
These were design and governance flaws, not defects inherent in Guidewire’s code.
Is it still a problem in 2025?
The public record says yes.
Feb 2025: icare issued a contract (CAN 4600003414) for “Guidewire v9 code and database remediation and analysis,” valued at $14.46 million over three years. Supplier: Tenzing Australia.
Oct 2024: icare paid approximately $764 000 for a “Discovery Phase for the icare Guidewire Upgrade” (CAN 4600003407), also to Tenzing.
icare’s 2024 public statement still speaks of a “comprehensive program of reform and remediation.”
When the official language remains “remediation,” the system isn’t fixed — it’s still being repaired.
(Sources: icare Procurement Improvement Program 2021; NSW Contracts Register CAN 4600003407 & 3414; iTnews Apr 2023.)
The cost to employers and workers
Injured workers: Regulator findings confirmed mis-triage and missing case-management support, leading to poorer outcomes.
Employers: Delays and classification errors distort risk pools and premium modelling, driving unpredictable costs.
Scale of impact: icare publicly confirmed in 2021 that 53 000 injured workers received back-payments for PIAWE underpayments under the Nominal Insurer. That figure alone shows tens of thousands have been negatively affected by administrative and systemic failures.
Why the Treasury Managed Fund (TMF) still hasn’t moved
The TMF covering public servants including teachers, nurses and police was meant to migrate to the same platform. It has not. No public contract shows a TMF Guidewire integration. The reasons are likely obvious: continued remediation, sensitive data-governance obligations, and Treasury’s reluctance to migrate billions in liabilities until the platform is stable.
What this teaches us about AI in government
icare’s experiment foreshadows the challenges every public agency will face as AI seeps into decision-making.
Algorithms embed human choices. icare’s triage logic reflected management assumptions, not machine objectivity.
Data lineage defines outcomes. Old claims data and actuarial biases became the system’s training set.
Transparency is non-negotiable. When technology governs access to compensation, decisions must be explainable and auditable.
“Remediation” is governance debt. If you’re still patching years later, the problem is your model not your machine.
What icare could publish tomorrow to rebuild trust
A plain-English model overview of triage logic and exceptions.
An independent validation pack on mis-triage rates (especially psychological injuries).
A public KPI series for claim duration, RTW and appeal rates.
A remediation register showing what’s being fixed and how success is measured.
A scheme-by-scheme Guidewire spend breakdown since 2016.
Sources / citations
SIRA Dore Review (2019) and EY support papers – Janet Dore Review of icare’s claims management and related matters (PDF) and SIRA Review of the Nominal Insurer resources page
icare Procurement Improvement Program (2021) and Improvement at icare (2024) – icare Procurement Improvement Program and Improvement at icare
NSW Contracts Register: CAN 4600003407 – Discovery Phase for icare Guidewire Upgrade and CAN 4600003414 – Guidewire v9 Code & Database Remediation
iTnews (Apr 2023): “icare signs Capgemini for $38 million claims system”
icare media release (Nov 2021): “icare to pay out $38 million to 53,000 injured workers”
ARNnet (Sept 2019): “NSW Government iCare claim platform cost $360M”
Sydney Morning Herald (Aug 27, 2020): “icare executives spruiked software company despite delays and blowouts” (may be paywalled)
Notes
This article relies entirely on publicly available information from the NSW Government Contracts Register, regulator-commissioned reviews (Dore Review and EY reports), icare’s own disclosures, and reputable media coverage (Sydney Morning Herald, iTnews, ARNnet, InnovationAus). All interpretations are drawn from those materials and presented in the public interest. References to “world-class technology” describe Guidewire’s global industry reputation, not any endorsement of icare’s configuration. Where scheme attribution is unclear because of redactions, contracts are labelled “likely” or “implied.” The phrase “tens of thousands of injured workers affected” derives from icare’s own confirmation that 53 000 workers received PIAWE remediation payments in 2021. This piece constitutes fair comment and truth-based reporting, produced with reasonable care and diligence using the best information available at the time of publication. If any factual inaccuracy is later identified, it will be corrected promptly in line with standards of responsible public-interest journalism concerning technology governance, workers’ compensation administration, and accountability in AI-enabled decision systems.
