The same spine, applied to underwriting and claims.
Appetite, completeness and risk scoring for commercial property submissions; coverage, reserve and leakage for auto claims. An unexamined fraud flag is treated as unknown, never as clean.
Run it
Run an underwriting submission or a claim, including a poisoned extraction that tries to hide a repeat-VIN claim.
Unedited file: the engine runs its committed facts and no model is called. Synthetic data only. Identifiers are redacted before any model sees the text.
How the agents hand off
One file through the fleet: verification, the decision, and a reviewer who samples what was auto-rendered.
Where it fits in the process
The lifecycle of a file, and the first workflow worth automating.
01Submission
Day 0A business (the customer) asks for coverage. Its broker fills out standardized forms describing the business and sends them to one or more insurance companies. This bundle of forms is called the 'submission.'
Who: Insured business + their broker/agent
Submission triage & intake for commercial underwriting
Intake sits atop the funnel where slow quotes and off-appetite junk destroy capacity.
- Extract ACORD, loss runs & SOVs to record
- Match against carrier appetite rules
- Rank winnable risks, route to underwriter
- Underwriter approves borderline appetite calls
- Human keeps the pricing & bind decision
Lines of business
What each line covers, how it is underwritten and how its claims flow.
Commercial property
Commercial linePays to repair or rebuild a business's buildings, equipment and inventory - and replaces lost income - after a fire, storm, theft or similar event.
Who buys it: Any business that owns or leases physical premises.
Submission → appetite check → classify the occupancy → rate the building value (TIV) and hazards (construction, protection, exposures) → quote → bind.
First notice of loss → confirm coverage → inspect / estimate the damage → set a reserve → settle the repair or rebuild.
Who else serves this
7 companies already selling into this workflow, described from their public material, and the gap each leaves for a carrier that needs a replayable decision. None is affiliated with Arbiter.
Dominant P&C insurance core suite (InsuranceSuite: PolicyCenter, ClaimCenter, BillingCenter, plus UnderwritingCenter) on Guidewire Cloud.
Gap for a carrier: Core-platform conservatism and human-in-the-loop governance cap autonomy; AI is largely an orchestration/assist layer atop workflows designed for adjusters/underwriters.
Source: guidewire.comEnd-to-end P&C core SaaS (Duck Creek OnDemand: policy, billing, claims) and the main challenger to Guidewire.
Gap for a carrier: Smaller installed base than Guidewire; agentic apps are newly launched and unproven at scale; like all core suites it keeps humans on final underwriting/claims decisions, capping true autonomy.
Source: duckcreek.comAI decision-optimization specialist for insurers: a fraud-detection engine that scores each claim against hundreds of evolving fraud scenarios using structured + unstructured data (docs, images, external/weather/location), plus…
Gap for a carrier: Specialist layer dependent on core platforms (Guidewire/Duck Creek) for distribution and could be encroached by their native AI.
Source: shift-technology.comComputer-vision AI that assesses vehicle (and property) damage from smartphone photos, generating repair estimates and triage (total loss / repairable / cash settlement) in seconds, trained on hundreds of millions of damage images.
Gap for a carrier: Value concentrated in one step (visual assessment); accuracy bounded by photo quality and blind to hidden/structural damage, so human review remains for edge cases.
Source: tractable.aiActuarial AI platform that automates insurance pricing and reserving using proprietary Transparent Machine Learning (TML) on GLM/GAM structures - building predictive technical models/pure premiums up to 10x faster while keeping full…
Gap for a carrier: Automates model-building, not the bind/decision - actuaries retain control and regulators approve rates, so it accelerates rather than removes human judgment; narrow to pricing/reserving.
Source: akur8.comGenerative-AI underwriting solution that ingests an insurer's own guidelines/risk appetite, then assesses submissions (SOVs, applications, loss runs + web-pulled data) and produces a 0-5 risk score with a cited rationale.
Gap for a carrier: Early-stage and venture-funded vs. entrenched cores; produces a risk assessment/score that supports the underwriter rather than autonomously binding risk.
Source: sixfold.aiSells "The Agentic Decision Platform": a decision engine, an AI agent manager, a case manager and a context layer for financial services.
Gap for a carrier: It is a platform the customer builds decisions on; the credit policy, its validation and its test suite remain the customer's work.
Source: taktile.comField notes
Reference material behind the design: who is involved, the documents in the file, the cost structure, the metrics a pilot would move, and the agent specification.
StakeholdersWho touches the file and what each owns
The business that buys the insurance and pays the premium. It wants protection if something goes wrong - a warehouse fire, a customer slipping and suing, a delivery truck crash.
An intermediary who works for the business, gathers its information, and shops it to insurers to get the best coverage and price. Brokers represent the insured, not the insurance company, and earn a commission.
A specialist firm an insurance company hands its 'pen' to - meaning the MGA is allowed to underwrite, price, and sometimes handle claims on the insurer's behalf for a niche it knows well (e.g., trucking or restaurants). It takes the risk decisions but usually doesn't hold the risk itself.
The company that actually promises to pay claims and holds the money to do so. It collects premiums, decides which risks to take, and bears the financial loss when claims happen.
A company that insures the insurance company. Carriers buy reinsurance so that if a hurricane causes thousands of claims at once, the reinsurer absorbs much of the hit and the carrier doesn't go bankrupt.
A firm hired to process and pay claims on behalf of an insurer or MGA. It does the day-to-day adjusting work - taking the loss report, verifying coverage, investigating, and cutting the check - without being the company that holds the risk.
The person (at a carrier or MGA) who reviews each submission, judges how risky it is, sets the price, and decides whether to accept, decline, or modify the coverage.
The person who handles a loss after it happens - confirms it's covered, figures out how much is owed, and approves payment. Their job is to pay what's fair and owed, no more and no less.
Documents in the fileWhat each document is and why it matters to the decision
ACORD application (e.g., ACORD 125)
An industry-standard fill-in-the-blanks form that describes the business applying for insurance - its name, address, operations, and ownership. ACORD is the nonprofit that publishes these standardized forms so every insurer reads the same format.
It is the front door of underwriting - the underwriter can't price a risk without knowing what the business is and does. Standardization lets insurers process thousands of submissions consistently.
ACORD 125 for 'Sunrise Bakery LLC, 412 Main St, Ohio, retail bakery with on-site oven, 14 employees, owned by two members.'
Cost structureWho pays for what, with illustrative figures
The price of the coverage; the insurer's core revenue. A small business might pay a few thousand dollars a year; a large manufacturer, millions.
How brokers get paid - a slice of the premium, so it costs the business nothing extra directly but is baked into the price.
The actual money paid out when a covered loss happens. This is what the loss ratio measures against premium.
What it costs the insurer to investigate and settle claims - adjuster salaries, lawyers, experts. Separate from the payout itself.
What the carrier pays to offload catastrophic risk. Rises sharply after big disaster years (a 'hard market').
Overhead to run the business - staff, systems, commissions, taxes. The US P&C industry expense ratio was about 25% in 2024, near historic lows.
The portion of a loss the business pays itself before insurance kicks in. Higher deductible = lower premium.
Metrics a pilot would moveIndustry reference figures from the sources listed; not results of this demo
How long it takes from receiving a request for insurance to sending back a price. Faster is better - slow quotes lose business to competitors.
Of all the prices an insurer quotes, what percentage of customers actually buy. Higher means the insurer is pricing well and targeting the right risks.
Of every dollar collected in premium, how many cents are paid out in claims. Lower is more profitable, but too low can mean overcharging.
The single most important profitability number - claims plus expenses as a percentage of premium. Below 100% means the insurer made an underwriting profit; above 100% means it lost money on insurance (before investment income).
The share of claim dollars paid out that shouldn't have been - overpayments, missed recoveries, paying for things not covered. Money quietly lost through sloppy handling.
The percentage of simple claims that get paid automatically with no human touching them. Higher means lower cost and faster service.
What it costs to handle claims, as a share of premium or losses - adjusters, lawyers, experts. Lower means a leaner claims operation.
Sources: practoinsura.com · sutherlandglobal.com · insurancejournal.com · federato.ai · selectsys.com · vcasoftware.com · vcasoftware.com · fivesigmalabs.com
Agent specificationThe build specification for Arbiter Risk + Arbiter Adjuster
GlossaryEvery term used on this page, searchable
The price a business pays an insurer for coverage, usually yearly.
e.g. Sunrise Bakery pays a $9,800 annual premium for its property and liability policy.
Claims paid divided by premiums collected, shown as a percentage - how much of the price goes back out as claims.
e.g. An insurer collects $100M in premium and pays $55M in claims: a 55% loss ratio.
Loss ratio plus expense ratio - the key test of whether insuring is profitable; under 100% is a profit.
e.g. 55% claims + 30% expenses = 85% combined ratio, meaning a 15-cent underwriting profit per premium dollar.
Money an insurer sets aside to pay claims it knows about but hasn't fully paid yet.
e.g. When a $50K injury claim is reported, the adjuster books a $50K reserve so the money is earmarked even before it's paid.
The nonprofit that creates the standardized insurance forms everyone uses, so submissions look the same across the industry.
e.g. A broker submits an ACORD 125 application and ACORD 139 statement of values for a new restaurant account.
The very first report that a loss or accident has happened, which starts the claims process.
e.g. After a kitchen fire, the bakery owner calls the hotline - that call is the FNOL.
An insurer's right to recover money from whoever actually caused a loss after it has paid its own customer.
e.g. After paying for fire damage caused by a faulty oven, the insurer sues the oven manufacturer to get its money back.
Money paid out on claims that shouldn't have been - overpayments, missed recoveries, or paying for non-covered items.
e.g. An adjuster approves a $42K roof estimate without noticing $6K was for upgrades the policy doesn't cover - that $6K is leakage.
The types of risks an insurer wants (or refuses) to cover.
e.g. A carrier's appetite includes offices and retail but excludes nightclubs and fireworks factories.
The moment coverage becomes legally active, before the formal policy is even printed.
e.g. The underwriter emails 'we're bound effective today' and the business is protected from that minute.
Insurance that insurers buy to protect themselves against huge or many-at-once losses.
e.g. A coastal insurer buys reinsurance so a single hurricane doesn't wipe it out.
The cost of investigating and settling claims, separate from the claim payout itself.
e.g. Paying a $20K claim also costs $2K in adjuster time and an expert's fee - that $2K is LAE.