to Agentic AI
what happens when you DO use it.
Who we are: Young and ambitious company in the actuarial field, providing actuarial support to various financial institutions.
How we work: Collaboration, alone you go faster, together you go further. People-focused, engaged and professional.
Core values: Together – different – better – more fun.
Mission: Dedicated pet insurer in the Netherlands. Fair, transparent and affordable insurance solutions.
Approach: Strong digital-first approach, enabling online policy management and fast claim processing.
In-house
Sales · Claims · Finance · Compliance · Reporting
Smart Data Layer & Tooling
AI agents · Automated pipelines · APIs · Own intelligence layer on top of outsourced engine
Outsourced
Actuarial function · Risk & compliance (2nd line) · Core insurance engine
DORA Compliance
Digital Operational Resilience Act. ICT-risk management, incident reporting, third-party risk. Not nice-to-have, a requirement.
Core Engine: outsourced
The insurance engine is outsourced. Policy admin, premium collection, claim handling. That’s what you buy.
Own intelligence layer
Own databases, APIs, MCP integration. The differentiation is in the data layer, not in the outsourced engine.
Set premiums
Policy conditions
Risk assessment
Accept / decline
Draft contract
Customer receives
Assess & process
Payout / denial
Risk analysis
Performance
claims, invoices, data
control, decision
Excel, dashboard
action, escalation
Harry Claims Assistant
Automated claim processing + weekly review.
PetSecur AI Agent
Local model for simple queries, own memory and database. Complex queries routed to frontier models with privacy protection.
Workflow & Process
Process documentation, workflow automation, process mining and optimization. Initial pricing automation framework in place.
Pricing Automation
Claims data in, pricing advice out. GLM Tweedie calculates loss ratios per cell (~340 cells for pets). LLM generates management summary per run. Full audit trail: which data, which model, which parameters.
2026: GLMs run, LLM explains why
Invoice Classifier
Receive veterinary invoices as PDFs, ML model automatically classifies and extracts amounts. 99% accuracy. From weeks of manual work to seconds.
2026: Same principle, now in production on our invoices
the actuary or the developer?
Claims processing automation
RPA processes claims automatically, retrieving data, matching, reporting. Humans handle the assessment and four-eyes check through a weekly review.
Internal AI Agent
Local model answers simple questions. Own memory and database, gets smarter over time. Complex questions routed to frontier models, with privacy protection for personal data.
P&C: motor, property, liability
Claims: Garage invoices, expert reports, photos. Automatically classify and extract. Same pipeline as our vet invoices.
Pricing: GLM per risk profile, rate differentiation across thousands of cells. LLM explains why the rate changes.
Fraud: Patterns in claims, photo analysis, cross-referencing. AI flags, humans decide.
Underwriting: AI-assisted, scoring risk profiles, matching terms, checking exclusions.
Life, pensions & income protection
Decision support: Benefit calculations, transitional law, scenario analysis. AI prepares, actuary decides.
Compliance automation: Solvency II, IORP II, pension agreement. Generate reports, flag deviations, maintain audit trails.
Second-line evidence: Merkle trees, verifiable AI. Cryptographic proof the model did what it should.
Automated testing: Validate calculation rules, run regression tests, run scenarios, in minutes, not weeks.
Acceptance
Score risk profiles, summarize medical statements, match policy terms. AI provides recommendations, underwriter decides.
Claims handling
Triage, classification, fraud scoring, document processing. From days to hours, with human oversight on every decision.
Actuary
Run pricing models, validate reserves, calculate ORSA scenarios. The actuary becomes faster, not obsolete.
Compliance & Risk
Regulatory reports generated, deviations flagged, audit trails built. Evidence for 2nd and 3rd line, automatically.
Operations
Docs, data entry, reconciliation, invoices. If it’s repetitive, an AI assistant can do it.
Service desk
AI agents that understand policy terms, search internal knowledge bases, and draft answer templates. Privacy-first.
AI processing
Decision + audit
Read invoice
Check policy
AI proposes
€340 payout
Human
approves
2017: “Black box problem” as a challenge
2026: Don’t open the box. Make the chain verifiable
Every step is hashed as a leaf. Pairs combine into branches, branches into a single root. If any step changes, the root hash breaks.
Selective disclosure: share one branch for audit, keep the rest private. Ideal for Solvency II, GDPR, and compliance reporting.
to “seeing what’s possible when you take action”
The question
ML was 0% adopted. Fraud, telematics, pricing. The chances were there, but untouched.
Tools are changing, fast
GenAI, LLMs, Claude, GPT. The barrier to building is vanishing. Exponentially, not gradually.
What can we do today?
Pricing, claims, agents, compliance: one person can build the intelligence layer in months, not years. We went through the entire Reactive → Proactive curve in just two months.
No barriers remain
No more vendor lock-in. Open models, open standards. SaaS costs drops. Compute is a commodity.
AI is commoditizing because models use the same public internet data. The true competitive edge isn’t the model itself anymore, but access to exclusive, proprietary datasets.
The data and insights created by the actuary, loss ratios, risk cells, claims patterns, reserving triangles, are the only moat left. If you capture them in your own data layer. Knowledge in spreadsheets and heads walks out the door. Knowledge in structured, versioned pipelines compounds.
The answer: yes, if you take action.”
