CLAUSTRUM

A pricing engine for insurance risk that returns a distribution, not a number.

Claustrum models each account as a predicted loss curve. Calibrated, explainable, and exposed through a single API. Underwriters see the shape of the risk, not just its midpoint.

CLAUSTRUM / SUBMISSION_SCORE
EXPOSURE
$84M
TENURE
12Y
LOSS HIST.
5Y
GEOGRAPHY
MULTI-STATE
P10P50P90μ
P(loss > 2× expected) = 0.094
P10
$31,400
P50
$48,200
P90
$94,600

BUILT ON BAYESIAN DEEP NETWORKS + SDE MODELING

SCOPE
INSURANCE RISK
OUTPUT
FULL LOSS DISTRIBUTION
METHOD
BAYESIAN DEEP NETS + SDE
INTEGRATION
API / BATCH
01PROBLEM

A single number is the wrong shape for the answer.

Most pricing systems hand the underwriter an expected loss and stop there. The variance, the worst credible outcome, and the reasons behind both are left to instinct. Two accounts with the same midpoint look identical — until one of them produces the claim that moves the book.

LEGACY_GLMPOINT ESTIMATE
LEGACY GLMOUTPUT
expected_loss$42,000
industry_factor1.18
revenue_band_factor0.92
records_factor1.05
one number. no shape, no drivers, no uncertainty band.
CLAUSTRUMOUTPUT
loss_p50$38,400
loss_p90$184,300
loss_p99$1,420,000
cv0.71
top_driverseverity_trend
reserve_p75$96,400
a calibrated curve, ranked drivers, and a reserve point you can defend.
SIGNALS LEGACY MODELS IGNORE
× severity_trend
× exposure_concentration
× loss_development_pattern
× claim_frequency_shift
× controls_maturity
× peer_benchmark_drift
02WHY NOW

The conditions for distributional pricing are finally in place.

01

Inference is cheap enough

Posterior inference that was a research workload a decade ago now runs inline with submission flow. Probabilistic models can finally sit in the underwriting path, not next to it — at the latency a real desk needs.

02

The market is ready for it

After a decade of volatile loss ratios, severity shocks, and systemic-event scares, underwriters know point estimates do not describe the risk. Buyers are actively choosing distributional, evidence-backed pricing.

03API

One call returns the predicted loss curve, the drivers, and a decision.

POST /v1/underwriteREQUEST
{
  "segment": "mid_market",
  "geography": "US_MULTISTATE",
  "exposure": { "basis": "revenue", "value": 84000000 },
  "policy": { "limit": 5000000, "retention": 50000, "term_months": 12 },
  "loss_history": [
    { "year": 2024, "incurred": 180000, "status": "closed" },
    { "year": 2022, "incurred": 940000, "status": "open" }
  ],
  "signals": {
    "exposure_concentration": 0.42,
    "controls_maturity": 0.78,
    "third_party_dependency": 0.34,
    "claim_frequency_trend": 0.21
  }
}
200 OKRESPONSE
recommended_premium
$48,200
confidence_interval
$31,400–$94,600
loss_distribution
{p10, p50, mean, p90, p99}
tail_probability
P(loss>2×mean)=0.094
reserve_guidance
$36,500
expected_loss_ratio
0.58
decision
refer
model_version
claustrum_core_0.3
RANKED RISK DRIVERS
Severity trend above peer benchmark+18% premium
Exposure concentration in correlated segments+11% premium
Open claims aging beyond expected developmentrefer
Controls maturity above 90th percentile-7% premium
P10P50P90μ
04WHERE IT FITS

Things you can do with a curve that you cannot do with a point.

01

Price the tail

Anchor premium to a chosen quantile of the predicted loss curve. Margin reflects the shape of the risk, not the average of it.

02

Reserve with shape

Case reserves and IBNR carry the same uncertainty the pricing model saw. Reserve to a band, with the drivers attached, defendable to actuaries and regulators.

03

Select on variance

Two accounts with the same expected loss are not the same risk. Surface the wider distribution — usually the concentrated or systemic exposure — before it ends up in the book.

05USERS

Built for the desks that own the price.

PROGRAM UNDERWRITING

MGAs & Programs

Triage submissions, set technical price, and route referrals using the shape of the predicted loss — on a single underwriting surface.

PORTFOLIO PRICING

Carriers

Score accounts against appetite and target loss ratio with explicit uncertainty. See which segments of the book carry the systemic and tail exposure.

TREATY & TAIL

Reinsurers

Re-underwrite ceded portfolios from the ground up. Price treaty layers and aggregation covers against a modelled loss curve, not aggregated cedant summaries.

06EARLY ACCESS

Get on the list.

We work with a small number of underwriting, pricing, and program teams at a time. Tell us where you sit and we will be in touch.

JOIN_WAITLISTSECURE

We respond to qualified inquiries within 48 hours. Your information is used only to evaluate fit and will not be shared.