Solutions / Minerva Reasoning Engine
The Reasoning Engine for Commercial Underwriting
Turn appetite into auditable, automation-ready decisions over live business data.
SMB AI So Good, We'll Give You Our Data for Free
Great decisions are made using great data, and we've got the best there is. But great data on its own isn't enough; small commercial carriers need an engine that turns that data into a confident, auditable verdict with more speed and accuracy than their peers.
The Minerva Reasoning Engine is that competitive edge: deterministic, rule-based, and replayable end to end, with every field sourced, every rule named, and without the confidently incorrect noise so common in other AIs.
Inside the Minerva Workbench
Four views of the same workbench: a single legible decision, the appetite authoring studio, the automation-first review queue, and the leadership dashboard.
Decision Detail
Verdict, reasoning trace, evidence
A legible verdict with the reasoning trace, evidence chips, and the data lineage behind every field that mattered.
Transparent, rule-based decisioning over live, resolved business data, fully replayable end to end. Every decision stores each pipeline step, each data pull (source plus fields), and each rule evaluation (the value read, the threshold, whether it fired, and the outcome).
Where a direct signal does not exist, the engine uses an explicit, clearly-labeled proxy and records it as a proxy on the trace, so a reviewer is never misled into reading it as a direct measurement. A multi-source fusion view groups evidence by field and shows where sources agree and where they conflict.
No False Positives-- Ever
Most AI is built to sound confident, even when it isn't. We took that out of the equation with three clean categories:
Confident
Confident
Clear enough to auto-decide (Write or Decline). This is the number the automation-rate metric tracks, and the work an underwriter never has to touch.
Refer
Refer
Borderline. Routed to a human, with the specific uncertainty named and how to resolve it surfaced on the file.
Insufficient
Insufficient
Not enough evidence to decide. The missing factors are named explicitly, so the underwriter knows what to chase: no confidently incorrect guesses, ever.
How a Submission Becomes a Decision
Every submission flows through a transparent, six-stage pipeline. Each stage is recorded, so the whole decision is replayable end to end.
Stage 01
Parse
Reads the submission as it arrives (broker email, loss run, or ACORD-style field block) and extracts the underwriting facts, keeping the raw snippet each value came from.
Stage 02
Resolve Identity
Matches the business against the Minerva resolver to confirm who this actually is and pull firmographics. If identity is uncertain, the decision is downgraded to Refer rather than written blind.
Stage 03
Enrich
Layers in the resolved firmographic profile and optional public hazard signals, with the source recorded for each field.
Stage 04
Run Appetite
Evaluates the carrier's compiled appetite rules against the facts. Missing data never silently fires a rule.
Stage 05
Score
Lands the decision in one of three tiers: Confident, Refer, or Insufficient, each calibrated to abstain rather than guess.
Stage 06
Explain
Produces the human-readable reasoning trace, the evidence picture, and the field-level lineage that lets a reviewer reconstruct exactly how the verdict was reached.
Five Capabilities at the Heart of the Engine
How MRE Stacks Up
Seven capabilities a modern underwriting platform actually has to deliver, scored across Minerva Reasoning Engine and the platforms carriers most often compare it to.
| Capability | Carpe MRE AI-native decisioning | Palantir Foundry / Ontology | Duck Creek Guidewire Core | BriteCore Origami / Others | Generic Data Vendor Feeds | Homegrown Excel / Scripts |
|---|---|---|---|---|---|---|
| Appetite-as-code, UW-editable | ||||||
| Evidence with provenance | ||||||
| Online signals (CAT, news, web) | ||||||
| Decision audit trail | ||||||
| Time-to-deploy | ||||||
| Insurance-domain fit | ||||||
| Cost / value ratio |
Comparison reflects Carpe's market analysis of vendor capabilities as of 2026.
Design-Partner Stage
Taking on a Small Number of Carriers
Minerva Reasoning Engine is in design-partner stage, and Carpe is taking on a small number of carriers to prove the platform against real appetite, real submissions, and real underwriting teams. If your underwriting leadership is ready to write its appetite in plain English and see every decision the engine makes against your own book, we want to talk.