The insurance model building & auditing company
We build the risk models insurance teams need — and independently audit, stress-test, and harden the ones already in production.
Flagship Model
Track the cascading impacts of an event
Most pricing tools answer the first hop — what was at the address. Our cascading-impact graphs walk outward, surfacing the second-, third-, and fourth-order exposures a single-asset lookup can't see.
Worked example
A minor fire in one production unit at an oil refinery. The classical catastrophe model logs a single small structural damage event and moves on. Our cascading-impact model walks the graph and traces what follows:
Downstream fuel depot
Daily despatch is scheduled against the refinery's output. With the line offline there is nothing to load and nothing to ship — the depot stands idle while contracted off-takers turn elsewhere. Business interruption.
Commercial enclave — four retail properties
The enclave's on-site diesel power system runs on a by-product stream piped from the refinery. With supply cut, the four tenants flip to grid backup at higher tariffs and reduced reliability. Utility-failure business interruption on the property line.
City arts & crafts fair
Cancelled by the local authority — precaution against the smoke plume and a pending environmental compliance review of the refinery's release inventories. Event-cancellation cover triggered; uninsured vendors lose the trading day.
One small structural fire — a single line in the cat-model output — becomes a portfolio impact map across business interruption, commercial property, retail tenant cover, and event cancellation.
Other Models
Forecasting, pricing & risk models — done properly
Generalised linear models for the frequency-severity work regulators and rating agencies expect; Bayesian and probabilistic models for forecasting with calibrated uncertainty; deep networks and transformers when the signal lives in sequence, text, or imagery. We pick the model class to fit the question — the data, the regulator, and the use case — rather than fitting every question to one toolkit.
- Classical actuarial. Frequency-severity, loss-cost, and rating-factor GLMs with full residual diagnostics — the form actuarial, pricing, and filing teams need.
- Probabilistic forecasting. Bayesian and probabilistic models for reserving, loss-ratio projection, submission-volume forecasting, and accumulation outlooks — with calibrated uncertainty bands attached to every estimate, not just a point number.
- Modern signal extraction. Deep networks and transformers for sequence, text, and imagery — claims notes, policy documents, satellite tiles, marine AIS streams — when the classical toolkit can't reach the data.
- Ships ready-to-run. Validation pack for regulator and rating-agency scrutiny, rate-file outputs, and integration hooks into Earnix, Akur8, or in-house pricing and broking platforms.
Auditing & validation
Independent review for the models you already run
Whether the model is internal, vendor-supplied, or AI-touching, three angles of independent review:
Stress testing
What banks do for credit, we do for insurance. Tail-event scenarios, climate-stress, accumulation shocks — applied to your pricing, reserving, and accumulation models with quantified breakage points and concentration findings.
Documentation review
Independent, third-party review of model documentation against best-practice and regulatory expectations (NAIC, IAIS, PRA, EIOPA). Identifies the gaps examiners or rating agencies will flag — before they do.
Robustness consulting
Engagement-style work to harden models against drift, data-shift, and adversarial attack — including AI-aware fraud detection and validation for LLM-touching pipelines like claims triage and underwriting copilots.
Use cases
What the work actually delivers
For claims & SIU
AI-aware fraud detection
Catch fabricated claims and AI-doctored evidence by cross-checking the claim against an independently curated ledger of confirmed peril events.
For underwriters & actuaries
Cascading-exposure pricing
Surface the second-order accumulation a curve can’t see — correlated assets in the same weather corridor, OEM cohort, or shared infrastructure dependency.
For lenders & diligence teams
Pre-close portfolio screening
Run a developer’s submitted addresses against the ledger to flag undisclosed prior incidents before the loan or acquisition closes.
For portfolio & risk teams
Live impact monitoring
Watch a portfolio of insured or owned assets; receive flag-and-trace context — including cascading second-order impact — whenever a confirmed event touches an exposure in your book.
For in-house modelling teams
Training & enablement
Hands-on training delivered alongside every model we ship and every audit we complete — so your actuarial, data-science, and risk teams can extend, monitor, and defend the work long after we hand it over.
About Arkline
What the name means
Ark — defence, stability, robustness. The ancient idea of a vessel built to carry what matters safely through what is coming.
Line — the Lloyd's-era word for a single underwriter's stake on a risk, the smallest unit of an industry whose written promise to make whole what gets broken has now run continuously for almost three and a half centuries.
We are two co-founders building tools that sit at the meeting point of those two ideas: models that price, stress-test, and audit the lines insurers write — so the promise can keep being kept against a world that is getting harder to underwrite.
Companies our team has built and audited models for
