Claims and underwriting, run for you · P&C and life insurance
We run your claims and underwriting function. Claims decided in minutes.
Antino builds a claims brain — your products, rules, and decision logic as an executable graph — and operates underwriting and claims on top of it. Machines handle the clean path. Every exception lands on a named human owner.
Below: the whole pipeline — playable
01 · The problem
Your loss ratio is decided by a queue you can't see into.
Underwriting rules live in a binder and three veterans' heads. Claims pile up while adjusters chase photos and documents. Leakage and turnaround both creep, and you can't tell which decisions were consistent and which were a Friday-afternoon judgment call. Headcount scales linearly with volume — until it can't.
02 · How we'd run it
The claims brain: every claim in, two ways out.
Intelligent decisions. Zero busywork. Every exception, owned.
Claims in
Any channel. Any format.
Photos & videos
From the policyholder's phone
Policy & history
Endorsements, past claims
Vehicle & damage data
Make, model, telematics
Third-party & external
RTO, IDV, blacklists
The claims brain
Rules. Models. Decisions.
- Coverage check
- Policy validation
- Damage assessment
- Fraud signal
- Rules engine (DMN)
- Limits & deductibles
every decision logged · every outcome written back
Straight-through
~80%Settled. Payment initiated in minutes.
Exception
~20%Routed to a senior adjuster
Complex cases. Human judgment — with the case pre-packaged:
Executable rules
DMN decision tables & policy logic
Auditable by design
Every decision logged with traceability
Continuous learning
Every outcome sharpens the brain
Regulatory ready
Built-in checkpoints & controls
Scales by default
Volume without linear headcount
Shares are illustrative of a typical mature claims book.
03 · The pipeline, end to end
The science of shipping, stage by stage.
This is the anatomy of a claims function on the brain — nine stages from first notice to write-back. Drag the agentic-depth dial to watch each stage flip from how the work feels today to how it runs on the brain, and hover any stage for a worked example. The ceilings are deliberate: the calls that matter stay human.
Agentic depth — drag it
0%Intake & assess
Claims brain · assess0%stage 01 · dossier
FNOL intake
First notice of loss arrives from any channel and becomes a structured claim file with coverage confirmed.
- Parses photos, video and documents into a structured claim
- Reads the policy graph and confirms coverage and status on the spot
6:02 pm Friday, a rear-ender
The policyholder uploads photos and a 20-second video from the junction. By 6:03 the claim file exists — coverage clause attached, documents inventoried — without a single hold-music minute.
A call queue and a paper formFiled from the kerb in one upload
click the stage to collapse
stage 02 · dossier
Coverage & document check
Completeness is verified against the product line's requirements; missing items are requested immediately, not at day four.
- Checks endorsements, deductibles and document requirements per product
- Requests exactly what's missing, with the reason, in one message
One ask, not a correspondence thread
The claim needs an RC book and nothing else. The policyholder gets one request naming the single missing document — instead of three rounds of "please share all documents again."
Three follow-up emails for one RC bookGaps chased the moment they appear
stage 03 · dossier
Pre-inspection
Vision models assess damage and severity from submitted media against thousands of prior claims.
- Classes damage zones and severity from photos and video
- Attaches comparable settlements and the policy's tolerance band
- An adjuster spot-checks borderline severity classes.
Pre-inspection without an inspector
The pattern we built running digital-first insurance ops: a kerbside video is classed — rear bumper, quarter panel, no structural signals — before a field inspector could have been booked, let alone dispatched.
Three calls and a site visitPhotos assessed on arrival
stage 04 · dossier
Fraud & consistency screen
Fraud signals run on every claim: frequency, damage-versus-description mismatch, blacklist and history checks.
- Scores frequency, mismatch and network signals on 100% of claims
- Packages a flagged file with media, history and the exact rule that fired
- A flagged claim is never auto-denied. It goes to a person, case already built.
The twin that didn't sail through
Same model of car, a second claim inside a month, damage that doesn't match the story. The brain's response isn't a verdict — it's a dossier, routed to someone qualified to render one.
Gut feel on a busy deskEvery claim scored, not a sample
stage 05 · dossier
Reserve setting
The initial reserve is recommended from damage class and comparable settlement history.
- Recommends a reserve from comparable settled claims in the same damage class
- Adjusters own reserves on high-severity and injury claims.
The reserve that doesn't drift
Rear-impact, drivable, no injury: the recommendation comes from the settled distribution of that exact class — so month-end reserve reviews stop being archaeology.
A round number from experienceA reserve grounded in comparables
stage 01 · dossier
FNOL intake
agentFirst notice of loss arrives from any channel and becomes a structured claim file with coverage confirmed.
- Parses photos, video and documents into a structured claim
- Reads the policy graph and confirms coverage and status on the spot
6:02 pm Friday, a rear-ender
The policyholder uploads photos and a 20-second video from the junction. By 6:03 the claim file exists — coverage clause attached, documents inventoried — without a single hold-music minute.
A call queue and a paper formFiled from the kerb in one upload
click the stage to collapse
stage 02 · dossier
Coverage & document check
agentCompleteness is verified against the product line's requirements; missing items are requested immediately, not at day four.
- Checks endorsements, deductibles and document requirements per product
- Requests exactly what's missing, with the reason, in one message
One ask, not a correspondence thread
The claim needs an RC book and nothing else. The policyholder gets one request naming the single missing document — instead of three rounds of "please share all documents again."
Three follow-up emails for one RC bookGaps chased the moment they appear
stage 03 · dossier
Pre-inspection
agenthumanVision models assess damage and severity from submitted media against thousands of prior claims.
- Classes damage zones and severity from photos and video
- Attaches comparable settlements and the policy's tolerance band
- An adjuster spot-checks borderline severity classes.
Pre-inspection without an inspector
The pattern we built running digital-first insurance ops: a kerbside video is classed — rear bumper, quarter panel, no structural signals — before a field inspector could have been booked, let alone dispatched.
Three calls and a site visitPhotos assessed on arrival
stage 04 · dossier
Fraud & consistency screen
agenthumanFraud signals run on every claim: frequency, damage-versus-description mismatch, blacklist and history checks.
- Scores frequency, mismatch and network signals on 100% of claims
- Packages a flagged file with media, history and the exact rule that fired
- A flagged claim is never auto-denied. It goes to a person, case already built.
The twin that didn't sail through
Same model of car, a second claim inside a month, damage that doesn't match the story. The brain's response isn't a verdict — it's a dossier, routed to someone qualified to render one.
Gut feel on a busy deskEvery claim scored, not a sample
stage 05 · dossier
Reserve setting
agenthumanThe initial reserve is recommended from damage class and comparable settlement history.
- Recommends a reserve from comparable settled claims in the same damage class
- Adjusters own reserves on high-severity and injury claims.
The reserve that doesn't drift
Rear-impact, drivable, no injury: the recommendation comes from the settled distribution of that exact class — so month-end reserve reviews stop being archaeology.
A round number from experienceA reserve grounded in comparables
Decide & settle
Claims brain · decide & learn0%stage 06 · dossier
Decision tables fire
DMN rules execute: appetite, completeness, fraud flags, severity tolerance — encoded judgment, not improvised judgment.
- Executes the decision tables with every input logged
- Writes the rule trace a regulator can replay end to end
The decision table that doesn't have moods
The Camunda/DMN pattern from our insurance work: in appetite, documents complete, severity inside tolerance — the decision is identical to what it would be on a different day, on a different desk.
Judgment improvised under backlogRules fire the same way at 2 am
stage 07 · dossier
High-severity approval
Total-loss, injury and fraud-flagged claims go to a senior adjuster — with the file assembled, never auto-decided.
- Assembles the full dossier: media, history, comparables, and the rule that escalated it
- The senior adjuster makes the call. This gate is permanent.
Ninety seconds to a decision, not a day to a file
The flagged claim arrives with the evidence and the triggering rule laid out. The adjuster spends her time on the judgment — fraud or false alarm — not on assembling the file to judge with.
Big claims wait in the same queueA built dossier on a senior desk
stage 08 · dossier
Settlement & payment
Payment is initiated, the settlement notice goes out, and the full audit trail is written at decision time.
- Initiates payment and notifies the policyholder with the maths
- Writes every rule fired and datum consulted into the claims file
The file the auditor can replay
Behind the payout sits the complete decision record — every rule fired, every datum consulted, every model score — assembled as it happened, not reconstructed months later under deadline.
A cheque run and a status callPaid before the policyholder is home
stage 09 · dossier
Write-back
Settlements and fraud resolutions write back; the straight-through tolerance and the fraud signal both get sharper.
- Feeds outcomes into tolerance and severity calibration
- Sharpens fraud signals with each confirmed or cleared flag
- Quarterly: humans review what the brain changed and why.
Monday's claims meet a smarter function
Friday's clean settlement tightens the tolerance data; the flagged twin's resolution sharpens the signal that caught it. The veterans' judgment stops leaking out the door and starts accruing.
Closed files go quietEvery outcome tightens the tolerances
stage 06 · dossier
Decision tables fire
agentDMN rules execute: appetite, completeness, fraud flags, severity tolerance — encoded judgment, not improvised judgment.
- Executes the decision tables with every input logged
- Writes the rule trace a regulator can replay end to end
The decision table that doesn't have moods
The Camunda/DMN pattern from our insurance work: in appetite, documents complete, severity inside tolerance — the decision is identical to what it would be on a different day, on a different desk.
Judgment improvised under backlogRules fire the same way at 2 am
stage 07 · dossier
High-severity approval
agenthuman ownsTotal-loss, injury and fraud-flagged claims go to a senior adjuster — with the file assembled, never auto-decided.
- Assembles the full dossier: media, history, comparables, and the rule that escalated it
- The senior adjuster makes the call. This gate is permanent.
Ninety seconds to a decision, not a day to a file
The flagged claim arrives with the evidence and the triggering rule laid out. The adjuster spends her time on the judgment — fraud or false alarm — not on assembling the file to judge with.
Big claims wait in the same queueA built dossier on a senior desk
stage 08 · dossier
Settlement & payment
agentPayment is initiated, the settlement notice goes out, and the full audit trail is written at decision time.
- Initiates payment and notifies the policyholder with the maths
- Writes every rule fired and datum consulted into the claims file
The file the auditor can replay
Behind the payout sits the complete decision record — every rule fired, every datum consulted, every model score — assembled as it happened, not reconstructed months later under deadline.
A cheque run and a status callPaid before the policyholder is home
stage 09 · dossier
Write-back
agenthumanSettlements and fraud resolutions write back; the straight-through tolerance and the fraud signal both get sharper.
- Feeds outcomes into tolerance and severity calibration
- Sharpens fraud signals with each confirmed or cleared flag
- Quarterly: humans review what the brain changed and why.
Monday's claims meet a smarter function
Friday's clean settlement tightens the tolerance data; the flagged twin's resolution sharpens the signal that caught it. The veterans' judgment stops leaking out the door and starts accruing.
Closed files go quietEvery outcome tightens the tolerances
Depth bars are illustrative of a mature claims book. The human gates — high-severity approval, fraud rulings — are permanent, not transitional.
04 · One brain, two engines
How the function runs on top of the brain.
Revenue OS
Underwrite and rateWe run underwriting on a DMN and Camunda decision graph. Clean risks bind straight through with a consistent, rule-rated premium; anything outside appetite is referred — with the rule that fired and the data it needs.
- Camunda/DMN underwriting: eligibility and rating as executable decision tables
- Straight-through binding for in-appetite risk, with a full audit trail
- KYC and endorsements processed inside the same governed workflow
- Out-of-appetite risk referred to a human owner with the reason attached
Delivery OS
Settle claimsWe operate claims end to end on the claims brain. AI pre-inspection assesses damage from submitted media, the routine claim settles, and the disputed or fraud-flagged claim goes to an adjuster pre-packaged.
- AI pre-inspection: damage and severity assessed from photos and video at FNOL
- Routine, in-tolerance claims settled straight through with documented logic
- Fraud and complexity signals surfaced before a human ever opens the file
- Exceptions escalated to a named adjuster, not a shared inbox
05 · What actually changes
The same function, before and after the brain.
06 · The outcome
Figures are illustrative unless tied to a named proof engagement.
07 · Live reference · Go Digit
Digital-first insurance operations built and run with Antino.
More use cases
The same brain, a different function.
09 · Run a function
Stop renting hours. Start running functions.
Pick the function you want off your plate. We'll map the brain and name the outcome we'd commit to — before you do.