APRA
CPS 234
Information security
Claims documents, medical records and policy data flowing to a model make that model an information asset. It needs tested controls and a named owner.
What we build in
Asset registerControl testingVendor review
Insurance · Sydney
We help Australian general, life and health insurers ship AI that moves the loss ratio and clears APRA review. Claims automation, underwriting support and conduct surveillance, designed with your risk team in the room.
Why insurers call us
"Claims volume tripled after the storms. Adjuster headcount did not. Touch time per claim is crushing our NPS."
"Conduct review is one person sampling 2% of calls. We found a breach in month fourteen we should have caught in month three."
"Our underwriting model works in pilot. Legal will not sign off until someone owns the bias testing."
Regulation map
Information security
Claims documents, medical records and policy data flowing to a model make that model an information asset. It needs tested controls and a named owner.
Operational risk management
If claims triage depends on a model provider, that provider may be material. You need fallbacks and tolerance levels for when it is slow or down.
General Insurance and Life Insurance Codes of Practice
Claims decisions carry timeframes and fairness duties. AI can approve simple claims and speed up the rest, but declines go to a person.
Design and distribution obligations
Products must reach the target market they were designed for. Underwriting and pricing models need monitoring for drift away from it.
Conduct and sales rules
Unsolicited selling is restricted and sales calls are scrutinised. Surveillance AI must flag risk without becoming a black box.
Australian Privacy Principles
Health and claims data is sensitive. New transparency rules for automated decisions start in December 2026.
This is a design guide, not legal advice. Your compliance and legal teams make the final call. We give them the evidence to make it quickly.
Use cases
Ranges are typical of what we see once a workload is in production. Your baseline decides your number, so we measure it first.
Agents read claim forms, invoices and photos, match them to cover, check for fraud signals and settle simple claims. Complex ones reach an adjuster with a summary.
Explainable models that help underwriters price and accept risk faster, with drift monitoring and bias testing your actuaries can review.
Every sales and service call reviewed, with hawking language flagged and evidence your compliance team can act on.
Graph and machine learning models alongside your rules, so investigators chase real fraud instead of false alarms.
Actuarial, claims and distribution dashboards on open-source Superset and ClickHouse, hosted in Australia.
How we keep it compliant
Every build follows the same path. Data stays onshore, a person signs off on decisions that touch a customer, and every step leaves a record.
Ways to engage
Two weeks: one claims or underwriting workload taken to an evaluated prototype, with the evidence your risk team will ask for.
from A$10K
Get a proposalSix to ten weeks: the workload live in your cloud, Sydney region, with evals, audit trail and human review wired in.
from A$50K
Get a proposalA squad that works alongside your claims, underwriting and risk teams, two to four days a week, with quarterly outcomes.
from A$12K/mo
Get a proposalQuestions
We design for AI to recommend and a person to decide on every decline. Simple approvals can be automated if your claims philosophy allows it.
Yes. Models and data are hosted in the Sydney region, and every data flow is documented for your CPS 234 register.
We test outcomes across customer segments before launch, monitor drift in production, and keep the reasons behind each decision so they can be explained.
Claude, GPT and open-weight models, chosen per workload on accuracy, cost and where they can be hosted. We do not resell any vendor.
Usually claims. It has the clearest baseline, the fastest payback and a regulator story that is easy to tell.
Start a conversation
A few lines is enough. A senior person reads every message and replies with a next step, not a sales sequence.
We reply within one business day.