Our AI

Not a chatbot.
A specialist.

Powered by Anthropic's Claude — purpose-configured for Australian insurance document analysis with AFSL-compliant guardrails, clause-level citations, and zero hallucination tolerance.

Powered by Anthropic

Built on Claude. The world's most capable AI.

Claude leads on three dimensions that matter for insurance: reasoning depth over long documents, instruction-following precision, and safety-first design. It reads your entire 150-page PDS in a single pass.

  • 1 million token context window — 150+ pages in a single pass
  • Structured JSON output — guaranteed valid, zero parsing failures
  • Adaptive reasoning — allocates more thinking to complex clauses
  • Safety-first — designed to refuse rather than fabricate
Anthropic Claude
Sonnet 4 · Sydney region
1M context Structured output Adaptive thinking Citations API
150+
Pages per pass
60s
Avg. analysis
0
Hallucinations
How it works

From PDF to structured insight.

Here's exactly what happens when you upload a PDS document. Every step designed for accuracy, traceability, and speed.

1

Document ingestion

Your PDF is uploaded securely. Claude reads the entire document natively — no OCR needed for digital PDFs.

2

Structured extraction

7-step analysis — reconnaissance, definitions, coverage, exclusions, conditions, gap analysis, cross-reference audit. Output forced into validated JSON.

3

Confidence scoring

Every data point gets High, Medium, or Low confidence with rationale. Medium and Low findings flagged for human review.

4

Cited output

Results delivered with clause-level citations — section number, page, clause heading. Click any citation to jump to the source.

Terminology normalisation
QBE
"Words with Special Meaning"
Definitions
Allianz
"What you are not covered for"
Exclusions
Zurich
"Basis of Settlement"
Valuation
CGU
"Cover Comparison Table"
Tier Matrix
Vero
"Parts A through G"
Sections
Insurance-Specific

Trained on PDS structure. Not generic text.

A generic AI doesn't know that "Words with Special Meaning" is QBE's definitions section. Our system prompt encodes deep knowledge of Australian PDS formatting across every major insurer.

  • Insurer-specific terminology normalisation — QBE, Allianz, Zurich, CGU, Vero, Suncorp, IAG
  • Multi-level definition resolution — general, section-specific, schedule-level
  • Exclusion hierarchy mapping with carve-back detection
  • Composite document awareness — PDS + SPDS + Schedule + Endorsements
Safety & Guardrails

What the AI will never do. By design.

These aren't suggestions — they're hard constraints built into every API call.

Never provides personal advice

Constrained to general information only. Never recommends a specific product or crosses into personal financial product advice.

Never fabricates information

If a clause isn't in the document, the AI says "not addressed." Designed to refuse rather than guess.

Never answers from training data

Responses grounded exclusively in your uploaded documents. No internet, no other clients' data.

Always includes advice warning

General Advice Warning appended automatically to every output. Enforced at the API level — can't be removed.

Never accesses other tenants

Every firm's data completely isolated. Enforced at the database level with tenant-scoped queries on every request.

Always cites its sources

Every claim includes a source reference — section, page, clause. If it can't cite a source, it doesn't make the statement.

Accuracy & Citations

Every claim traced to the source.

Trust comes from verifiability. Every finding includes a precise citation — section number, page, and clause heading — so you can verify any claim in seconds.

  • Clause-level citations — "Section 4.3, p.28" not just "see page 28"
  • Click-to-source — citations link directly to the clause in the PDF
  • Per-field confidence scoring — not one score for the whole document
  • Ambiguous provisions explicitly flagged for expert interpretation
Confidence breakdown
High
Policy number, insurer, product name
Declarations p.1
High
Property limit: $5,000,000
Schedule p.3
Medium
BI cover: "up to 12 months"
S5.1 p.44
Low
Endorsement 3 may override S4
Not provided
Data flow
Your browser
TLS 1.3
Xano API — Sydney
AES-256
Anthropic Claude API
Zero retention
Results stored — Sydney
7-year retention
Anthropic does not train on your data
Data & Security

Your data stays in Australia.

Infrastructure hosted in Sydney on SOC 2 compliant servers. Documents encrypted in transit and at rest. Anthropic does not train on your data.

  • Australian-hosted infrastructure — Sydney region, always
  • Zero data retention by Anthropic — not used for training
  • End-to-end encryption — TLS in transit, AES-256 at rest
  • Multi-tenant isolation — every firm's data completely separated
Why Claude

Why we chose Anthropic over the alternatives.

We evaluated every major AI model. Three things made the decision clear.

Long-document reasoning

Claude's 1M-token context window reads an entire 150-page PDS in a single pass. Definitions on page 5 stay connected to exclusions on page 67.

Instruction following

Claude follows complex, multi-step analysis prompts with high fidelity. When told to extract exclusions verbatim and flag carve-backs, it does exactly that.

Safety-first design

Anthropic's constitutional AI approach means Claude is designed to refuse rather than fabricate. In insurance, a confident wrong answer is worse than no answer.

Contact

AI-powered insurance.
Coming soon.

We're here to answer your questions. We usually reply within one business day.

Prefer email? Write to hello@insuranceadvice.app