Minerva Reasoning Engine [What It Changes]

TL;DR

• The Minerva Reasoning Engine is Carpe's August 20, 2026 underwriting engine that turns a carrier's small-commercial appetite into a source-backed quote, decline, or refer, per Carpe's launch.

• Carpe says it can evaluate more than 200 business characteristics against 50 million-plus U.S. business profiles, cut underwriting touches by up to 25%, and save 30–45 minutes per submission. Those are vendor figures, not a regulator trial.

• FinTech Global independently restated the same 25% touch cut and 30–45 minute save on August 24, 2026, quoting CEO Max Drucker on keeping judgment on the risks that deserve it.

• A two-truck HVAC shop, a 10-person agency, or a solo clinic should care because the same research tax shows up as a delayed certificate, a declined policy after the job is already booked, or an underwriter who never opened the file.

Key Takeaways

• Quote, decline, or refer is the product claim. Underwriters stay on judgment-intensive risks. Carpe's product page adds Confident, Refer, and Insufficient as the three score tiers.

• 25% fewer touches and 30–45 minutes saved are Carpe's claims. They are not loss-ratio results and they are not a NAIC examination finding.

• State insurance regulators still expect an AI system that supports underwriting to comply with unfair-trade and unfair-discrimination rules. The NAIC's December 4, 2023 model bulletin is the written reminder.

• The first honest test is one submission: identity resolved, appetite rule named, evidence chip clickable, and a human who can reject the refer.

The answer in plain English

The Minerva Reasoning Engine is Carpe's AI-native commercial underwriting engine that resolves a small business, enriches the record, applies the carrier's named appetite rules, and returns a quote, decline, or refer with a source-backed reasoning path. As of August 20, 2026, Carpe launched it from Santa Barbara, California, for carriers that write small commercial — the policies that cover the HVAC shop, the agency, and the clinic, not the Fortune 500 campus.

If you run two trucks, you already know the cheap version of this failure: the broker emails a submission, someone at the carrier rebuilds the file from public records and a "we do not write X" spreadsheet, and forty minutes later the certificate still has not moved. A 10-person agency waiting on professional-liability evidence and a solo clinic waiting on a certificate sit in the same queue. Minerva is Carpe's pitch that the carrier can run that research as a replayable pipeline. The useful question is whether the recommendation names the rule, names the source, and leaves the underwriter able to say no. Red flags: treating 50 million profiles as your insured; auto-declining without a named rule; skipping the evidence chip because the verdict sounded confident.

What Carpe launched on August 20

According to Carpe, the company launched the Minerva Reasoning Engine on August 20, 2026 as an AI-native commercial underwriting solution that uses live business intelligence to turn each carrier's small commercial appetite into evidence-backed quote, decline, or refer recommendations. August 20, 2026 is Carpe's published launch date. The same release says small-commercial underwriters face a constant challenge: the time spent researching a business must not outweigh the policy's potential profit.

Carpe's workflow, in the company's own order, is a single pass: resolve the business, enrich the record, apply named testable appetite rules to current Carpe intelligence, and return a source-backed reasoning path. Carriers can apply the same logic across submissions and channels while keeping underwriters on judgment-intensive risks. CEO Max Drucker is quoted: carriers need better decisions from the data already available to them; Carpe has spent more than a decade building commercial intelligence; Minerva is meant to expand capacity while keeping judgment on the risks that deserve it.

The launch lists four help claims: reduce underwriting touches by up to 25%; save 30–45 minutes per submission; evaluate more than 200 business characteristics; apply appetite across 50 million-plus U.S. business profiles with plain-language tuning. Carpe says it is available to commercial carriers. The media contact is Hallie Criste at INK Communications for Carpe. The homepage at carpe.io lists separate claims-side marketing counts (10 million-plus claims processed, $500 million-plus reclaimed since 2016). Those homepage counts are not Minerva underwriting trial results.

Sources: Carpe launch; FinTech Global.

What FinTech Global independently reported

According to FinTech Global, Carpe introduced the Minerva Reasoning Engine in August 24, 2026 coverage as a tool to help small commercial insurers underwrite more profitably. 30–45 minutes per submission is the time-save range FinTech Global restated from Carpe. The outlet also restated the up-to-25% touch cut, the 200-plus characteristics, and the 50 million-plus profiles, and reprinted Drucker's quote. It is a news restatement, not a timed study. Do not treat the second publisher as a second measurement.

Inside the workbench Carpe ships

Carpe's Minerva product page describes the engine as deterministic, rule-based, and replayable, with every field sourced and every rule named. Four views: decision detail with evidence chips, appetite-authoring studio, automation-first review queue, and leadership dashboard. Uncertain identity is supposed to Refer, not write blind. Missing data is not supposed to silently fire a rule. Score tiers — Confident, Refer, Insufficient — are not automatically the same labels as quote, decline, or refer. The six-stage pipeline is Parse, Resolve Identity, Enrich, Run Appetite, Score, and Explain. A vendor comparison table on that page is Carpe's own market analysis as of 2026, not a lab result.

Agencies already comparing Applied Epic versus Salesforce Financial Services Cloud are choosing a system of record for producers. Minerva sits on the carrier side of the same submission. It does not replace the agency management system. Retention tools in Agency Revolution versus Better are a different step: keeping the account after the policy is bound. Insurance automation workflows are the broader map; this hub is the August 20 reasoning engine only.

What the labor market around underwriters actually shows

According to the U.S. Bureau of Labor Statistics Occupational Outlook Handbook, insurance underwriters had a May 2025 median pay of $81,370 per year, or $39.12 per hour, with 125,600 jobs in 2025 and a projected −4% employment change from 2025 to 2035 (−4,800 jobs). $81,370 is BLS's published 2025 median for insurance underwriters. BLS projects about 6,800 openings each year on average over the decade from replacement, not from growth. Typical entry-level education is a bachelor's degree. The Handbook page was last modified August 27, 2026.

The BLS Industries at a Glance page for NAICS 524 lists a preliminary 2,935.6 thousand jobs in the insurance-carriers subsector in July 2026 and 97,600 insurance underwriters in the subsector in 2025 (OES), with the same $39.12 / $81,370 medians. Handbook and OES employment counts differ because the frames differ. Data on that glance page were extracted September 1, 2026.

None of those BLS counts measure Minerva. They measure the occupation Carpe is selling into. A 25% touch cut would not, by itself, change the −4% occupational outlook.

Sources: BLS Insurance Underwriters; BLS NAICS 524.

Why a shop, an agency, or a clinic is in this file

The U.S. Small Business Administration still lists "Get business insurance" as a launch step. Location-specific costs it names include business insurance rates. That is why a two-truck operator, a 10-person agency, or a solo clinic is the named insured on the small-commercial submission Minerva is built to score. The Census Bureau's Statistics of U.S. Businesses counts employer firms by industry and enterprise size; latest featured data on that program page is 2022. SUSB is not Carpe's resolver. Do not treat a Census establishment count as the 50 million-plus profile file.

How small-business automation actually gets used and form-to-CRM tools are the insured's side of the same paperwork. Executive-assistant task automation is the broker-office version of the research tax.

USTA analysis: minutes saved at the BLS median wage

Working only from Carpe's 30–45 minutes per-submission save on the launch page and BLS's $39.12 median hourly wage on the underwriter Handbook:

• Low end: 30 ÷ 60 × $39.12 = $19.56 of median-wage time per submission.

• High end: 45 ÷ 60 × $39.12 = $29.34 of median-wage time per submission.

• Midpoint of Carpe's range: (30 + 45) ÷ 2 = 37.5 minutes; 37.5 ÷ 60 × $39.12 = $24.45.

• Touch-cut remainder: Carpe's "up to 25%" leaves at least 75% of underwriting touches in the book (1 − 0.25 = 0.75).

Inputs: 30 and 45 from Carpe; $39.12 from BLS. The dollar figures and 75% are arithmetic on those inputs, not a loss-ratio. Carpe's 25% is "up to." The wage is a national median, not your loaded cost.

Sources for inputs: Carpe; BLS. Dollar figures and 75% are USTA analysis.

Teams already routing submissions, exception lists, and broker follow-ups through US Tech Automations can treat Minerva as the model behind one step — parse, resolve, recommend, wait — without rebuilding the human refer queue. The engine does not bind the policy.

Data-extraction agents are the USTA shape for "read a packet, propose a field, wait." They are not Carpe.

The regulatory overlay the engine does not replace

According to the NAIC's Artificial Intelligence topic page (last updated 4/3/2026), 88% of 193 responding auto insurers reported they use, plan to use, or plan to explore AI/ML models. 88% of 193 auto-insurer respondents is the NAIC survey figure. Home: 70% of 194. Life: 58% of 161. Health: 92% of 93. The NAIC formed the Innovation Cybersecurity and Technology (H) Committee in 2021 and the Big Data and Artificial Intelligence (H) Working Group in 2019. As of March 2026, an AI Systems Evaluation Tool was being piloted by 12 participating states. The NAIC homepage at content.naic.org states the organization has supported state insurance regulators since 1871.

The NAIC Model Bulletin on the Use of Artificial Intelligence Systems by Insurers was adopted by Executive (EX) Committee and Plenary on December 4, 2023. It reminds licensed insurers that AI-supported decisions impacting consumers must comply with unfair-trade and unfair-discrimination law, and it expects a written AIS Program. Third-party AI providers sit in the bulletin's "Third Party" definition. Minerva, sold to carriers, sits in that box from the insurer's side of the exam. The bulletin does not name Carpe.

The U.S. Department of the Treasury Federal Insurance Office was established under Title V of the Dodd-Frank Act. FIO monitors the insurance sector, underserved access to non-health products, and represents the United States at the International Association of Insurance Supervisors. FIO does not license Minerva. State departments still do.

Source: NAIC Artificial Intelligence topic.

Trust, security, and what a breach still requires

According to NIST, the AI Risk Management Framework (AI RMF) was released on January 26, 2023 for voluntary use. January 26, 2023 is NIST's AI RMF 1.0 release date. The same page points to the January 26, 2023 news note, the July 29, 2021 Federal Register RFI, and the download at doi.org/10.6028/NIST.AI.100-1. The NIST AI Resource Center launched March 30, 2023. The generative-AI profile landed July 26, 2024. The framework is voluntary. It is not a Carpe certification.

ISO/IEC 27001:2022 is the published information-security management-system standard (Edition 3, 2022, 19 pages). ISO reports over 70,000 certificates in 150 countries in the ISO Survey 2022. Carpe's Minerva pages fetched for this hub do not publish an ISO/IEC 27001 certificate number. Do not invent one.

The Federal Trade Commission data-breach guide still requires a business that exposed personal information to secure operations, notify law enforcement, and notify affected people; all states plus D.C., Puerto Rico, and the Virgin Islands have breach-notification laws. CISA's cybersecurity best-practices page still starts with passwords, updates, and multi-factor authentication, and it publishes a 24/7 incident number: 1-844-729-2472. Advisory AA26-237A on that page is dated August 25, 2026. Neither FTC nor CISA scores Minerva.

Why now: the research tax outran the premium

Carpe's diagnosis is that small-commercial research is too slow and too inconsistent relative to the premium. The three constraints the launch names are manual research, disconnected data, and appetite applied differently by desk. Named rules plus a 50 million-plus profile file are the proposed close of that loop. Named rules are also how a bad proxy becomes a systematic decline. Keep the refer path. US Tech Automations is the routing layer for "submission in, exception out, human on the refer," not the appetite compiler.

It does not establish that 50 million profiles include your named insured, that the 25% touch cut landed on a published book, or that 30–45 minutes is a timed study. It does not replace the underwriter's license.

A buyer's evaluation sequence

US Tech Automations can keep the reject reason when a recommendation fails click-through. It should not issue the quote.

A week-one test that still opens the source

Write the test before auto-decide. Pick one write, one decline, and one refer you already know. Run all three. For each verdict, click the evidence. Score only: named rule and correct, named rule and wrong, unnamed rule, and wrong identity. If identity is uncertain and the file Writes, stop. If a proxy is unlabeled, stop. Keep Carpe's homepage claims-recovery figures out of this test; they are a different product family on carpe.io. If Carpe later publishes methods behind the 25% and the 30–45 minutes, read the denominator. Until then, the board-deck numbers are 200-plus characteristics, 50 million-plus profiles, August 20 and August 24, BLS's $81,370 / $39.12, the NAIC survey shares, and your three-file tally.

Signal vs Speculation

Demonstrated signal: as of August 20, 2026, Carpe launched the Minerva Reasoning Engine to turn small-commercial appetite into quote, decline, or refer with a source-backed path, claiming up to 25% fewer touches, 30–45 minutes saved per submission, 200-plus characteristics, and 50 million-plus U.S. business profiles. FinTech Global restated those claims on August 24, 2026. The product page adds a six-stage pipeline and Confident / Refer / Insufficient scoring. BLS publishes $81,370 median pay, 125,600 underwriter jobs, and a −4% outlook. The NAIC AI topic page (updated 4/3/2026) reports 88% / 70% / 58% / 92% survey shares and a 12-state evaluation-tool pilot. The model bulletin was adopted December 4, 2023. NIST AI RMF 1.0 is dated January 26, 2023. ISO/IEC 27001:2022 is the published ISMS standard. FIO exists under Dodd-Frank Title V. SBA still lists business insurance as a launch step. SUSB's latest featured year is 2022. None of those federal or NAIC pages name Minerva.

Our read: over 12–36 months, small-commercial carriers and MGAs will trial Minerva on a slice of straightforward appetites, with click-through mandatory and auto-decide off until identity and proxy labeling survive a three-file test. Some will confuse quote/decline/refer with Confident/Refer/Insufficient in an RFP. Agencies on Epic will not migrate for a carrier engine they cannot hang on their record. Unreviewed Writes still do not count. If explainability is the claim, the test is a known appetite exception, not a happy demo.

Frequently asked questions

What is the Minerva Reasoning Engine?

It is Carpe's August 20, 2026 small-commercial underwriting engine that resolves a business, applies named appetite rules, and returns a source-backed quote, decline, or refer for a human to use or reject.

Does it replace the underwriter?

No. Carpe's own launch says judgment stays on the risks that deserve it, and the product page routes borderline and insufficient files to a person.

Are the 25% and 30–45 minute figures independent?

No. They are Carpe's claims. FinTech Global restated them. They are not a NAIC exam finding or a BLS series.

What do 200 characteristics and 50 million profiles measure?

Carpe's stated feature count and U.S. business-profile file. They are not your submission count.

Does the NAIC bulletin allow this?

The December 4, 2023 bulletin does not ban underwriting AI. It requires the insurer's use of AI systems that affect consumers to comply with existing law and to be governed. Minerva does not hold the certificate of authority.

How should a team plug this into work they already run?

As a model swap on a bounded research-and-recommend step they already review, with click-through and a stop button. Binding stays human. The launch is sold to commercial carriers; the shop is the named insured in the file, not the buyer.

Glossary

• Appetite: the carrier's written rules for which small-commercial risks it wants to write.

• Quote / decline / refer: the three press-release outcomes Minerva is built to recommend.

• Confident / Refer / Insufficient: the three score tiers on Carpe's product page; not automatically the same labels as quote / decline / refer.

• Resolver: Carpe's identity-matching step that confirms which business the submission is about.

• AIS Program: the written AI-system governance program the NAIC model bulletin expects insurers to maintain.

• SUSB: Census Statistics of U.S. Businesses, the public employer-firm series by size and industry.

• Proxy (Carpe usage): a stand-in signal the engine says it will label when a direct measurement is missing.

If you already route submissions, evidence packs, and refer queues through automated steps, map an agentic workflow that keeps the named rule next to every proposed verdict. US Tech Automations is the homepage for that routing layer, not an underwriting system.