Skip to content

AI for Insurtech

Simple claims settling themselves. Adjusters on the ones that matter.

We engineer AI for carriers, MGAs, and specialty insurers: FNOL triage, claims-packet document ingestion, fast-track decisioning, fraud screening before an adjuster opens the case, and underwriting copilots that read a submission like your best underwriter does.

Built on your claims system, your policies, and the audit trail your DOI will accept.

Discuss your claims workflowSee how the system works

The operational reality

FNOL hits five channels. Documents arrive all week. Simple claims and hard ones share one queue.

Adjusters drown in the easy ones. The hard ones wait.

Loss adjustment eats the margin. Customers remember the wait.

Felt by Head of Claims, SIU leads, Chief Underwriters, Chief Actuaries, and the Head of Direct Distribution watching combined ratio and NPS at the same time.

  1. 01

    FNOL intake

    First notice of loss arrives through the app, the call centre, the agent, email, or a broker portal. The claim is coded, the policy is matched, and reserves are set.

    Friction · The same loss shows up in three channels; the policy match is manual on specialty lines.

  2. 02

    Document ingestion

    Photos, estimates, police reports, medical bills, repair invoices, recorded statements, and prior-loss history land into the claim file over the following days.

    Friction · An adjuster re-reads the whole packet every time something new arrives.

  3. 03

    Triage and decisioning

    The adjuster decides whether the claim is a fast-track settlement, a routine assignment, or a case to investigate further. Reserves are updated as the picture firms up.

    Friction · Simple claims sit in a queue behind complex ones; backlogs stretch during cat events.

  4. 04

    Fraud and SIU triage

    Suspicious patterns, prior-claim networks, and inconsistencies are referred to the Special Investigations Unit. Everything else moves on.

    Friction · SIU capacity is finite; weak referrals crowd out the files worth working.

  5. 05

    Settlement, subrogation, and close

    Payment is issued, subrogation potential is flagged against the at-fault carrier, and the file is closed with a reason code the actuary can read.

    Friction · Subrogation opportunities leak when the adjuster is behind on the queue.

  6. 06

    Underwriting and renewal

    Submissions and renewals pull in loss runs, exposure data, and the broker narrative. The underwriter writes, declines, or non-renews with a documented rationale.

    Friction · The best underwriters spend their day on document assembly, not on judgement.

Swipe the workflow

Where the work breaks

The friction is in the places nobody budgets for.

  • 01

    Adjuster backlogs stretch for weeks during cat events.

    How we fix it · A system we build catches this before it costs you.

    Impact · Policyholder NPS falls, bad-faith exposure rises, and the simple claims that should have cleared in hours sit behind the complex ones.

  • 02

    Legacy policy and claims systems were not built to read a document.

    How we fix it · A system we build catches this before it costs you.

    Impact · Every new photo or bill means an adjuster re-reads the file; the system holds the record, not the understanding.

  • 03

    Fraud leakage hides in the files no one has time to work.

    How we fix it · A system we build catches this before it costs you.

    Impact · Weak SIU referrals get worked; strong signals in low-dollar claims pay out quietly and show up in loss runs a year later.

  • 04

    The underwriter is doing document assembly, not judgement.

    How we fix it · A system we build catches this before it costs you.

    Impact · The quote goes out late, the broker places the risk elsewhere, and the deals that come back are the ones no one else wanted.

  • 05

    Combined ratio pressure is squeezing both sides at once.

    How we fix it · A system we build catches this before it costs you.

    Impact · Rate action lags the loss trend, expense ratios creep upward, and leadership runs out of levers that do not touch headcount.

What we would build

Systems, not features. Each with its own job.

  • 01

    FNOL triage and fast-track settlement

    Low-severity, well-documented claims should settle without an adjuster. Today they wait in the same queue as everything else.

    Who benefits · Claims operations, fast-track teams, and digital-first direct lines.

    Inputs
    The policy, the FNOL record, submitted photos and documents, prior-loss history, and the carrier's own fast-track rules and reserve bands.
    What the system does
    A triage model reads the FNOL packet, matches it to policy coverage, scores severity and complexity, and either stages a fast-track settlement for review or routes to an adjuster with a prepared summary and reserve recommendation.
    Output
    A triaged file: a staged settlement with the audit note the regulator expects, or a routed assignment with the context the adjuster would have had to gather by hand.
    Benefit
    Simple claims close in hours instead of days. Adjusters spend their time on the files that need judgement.
    Human stays here
    A claims specialist reviews every fast-track settlement before it is issued. The system refuses to fast-track files with injury, liability dispute, or prior-loss ambiguity.
  • 02

    Claims document understanding

    Photos, estimates, police reports, medical bills, and repair invoices arrive in every layout and quality. An adjuster re-reads them each time.

    Who benefits · Adjusters, examiners, and vendor-management teams.

    Inputs
    The documents in the claim file, the policy, the vendor-estimate schema, state-specific fee schedules where they apply, and the carrier's own prior files.
    What the system does
    Each document is classified, OCR'd, and extracted into structured fields with a confidence per field. Values are validated against the policy, the fee schedule, and prior extractions. Low-confidence fields are flagged, not guessed.
    Output
    A structured claim file: coverage decisions grounded in the policy, estimates reconciled against the fee schedule, and every extracted value linked to the page it came from.
    Benefit
    The second read goes away. New documents update the file in minutes, with citations the adjuster can open.
    Human stays here
    Fields below confidence threshold are queued for a human; nothing writes to the claims system until an adjuster confirms coverage decisions.
  • 03

    Fraud screening and SIU triage

    Fraud leakage hides across low-dollar files that never reach SIU, while weak referrals burn investigator time.

    Who benefits · SIU investigators, fraud operations, and claims leadership.

    Inputs
    The claim file, the policy and prior-loss history, the claimant and provider network across prior claims, device and session data from digital FNOL, and public records permitted under FCRA.
    What the system does
    A screening model scores each claim against known fraud patterns and the carrier's own historical SIU outcomes. It surfaces the top contributing signals so an investigator can decide, not a black box.
    Output
    A ranked SIU referral queue: each file carries its score, its top three contributing signals, and the network view that connects it to prior claims.
    Benefit
    SIU works the files that pay back. Leakage on the long tail of low-dollar claims drops because screening is universal, not sampled.
    Human stays here
    An investigator decides. Adverse-action procedures follow FCRA. The model's version and inputs are logged for every scored file.
  • 04

    Subrogation and recovery

    Subrogation potential leaks when the adjuster is behind. The at-fault carrier is already paying out before anyone has filed.

    Who benefits · Subrogation teams and recovery vendors.

    Inputs
    The claim file, the police report, prior-loss narratives, and the carrier's own historical recovery outcomes by state and loss type.
    What the system does
    The system reads the loss description and the supporting documents, scores subrogation potential, drafts the demand package, and routes it to a recovery specialist before the clock on the at-fault carrier runs down.
    Output
    A drafted subrogation demand, with the liability theory, the supporting evidence, and the carrier's own recovery band for this loss type.
    Benefit
    Recoveries that would have aged out get filed. Loss runs reflect net losses, not gross.
    Human stays here
    A subrogation specialist reviews every demand before it is sent. Recovery decisions on contested files remain human.
  • 05

    Underwriting submission copilot

    The best underwriters spend their day assembling the submission packet, not pricing the risk.

    Who benefits · Underwriters, line-of-business leaders, and portfolio actuaries.

    Inputs
    The broker submission, loss runs, exposure schedules, the carrier's own appetite rules, and prior-year underwriting notes on the same account.
    What the system does
    The submission is parsed, the exposure schedule is normalised, the loss runs are read against the carrier's appetite, and a draft underwriting summary is staged alongside the raw file.
    Output
    A draft underwriting worksheet: exposure summary, loss analysis, appetite match, and the broker narrative with the fields the underwriter will need to decide.
    Benefit
    Underwriters quote faster on the risks that fit, and decline cleanly on the ones that do not.
    Human stays here
    Pricing and binding stay with the underwriter. The system drafts and surfaces; it does not quote.

Show the system

What it could look like in your hands.

A design concept, not a shipped product. Interfaces like this get drawn out before any code lands, so the system reads back the way the people using it work.

claims ops / triage queue

Open queue

  • CLM-82041 · R.H.

    $3,420

    Auto, collision

    Aging 1d

    low

  • CLM-82038 · M.A.

    $18,900

    Homeowners, water

    Aging 4d

    medium

  • CLM-82029 · D.K.

    $42,700

    Auto, bodily injury

    Aging 9d

    high

  • CLM-82024 · L.P.

    $6,850

    Homeowners, theft

    Aging 6d

    medium

  • CLM-82019 · T.S.

    $480

    Auto, glass only

    Aging 1d

    low

  • CLM-82012 · J.N.

    $74,200

    Homeowners, fire

    Aging 12d

    high

Claim CLM-82041

Claimant R.H.

PA-04-11-882 · Auto, collision, $500 deductible, in force 3y 4m

Loss description

Single-vehicle collision with fixed object in a parking garage, no injuries claimed, photos and police report submitted at FNOL.

+6 images processed

Triage recommendation

Fast-track settle

confidence 86%
  • Loss description, photos, and police report are internally consistent.
  • No injury claimed; no third-party liability; no prior-loss conflicts on policy.
  • Repair estimate is within the fast-track band for this coverage and ZIP.
Synthetic data. This is a design concept of how an adjuster could see the open queue, the AI triage recommendation for the focused claim, and the actions a regulator would expect audited.

End-to-end workflow

From the trigger to the outcome.

A representative claim path, from FNOL to close. Automated stages run the moment the data lands; human stages are explicit and never skipped.

  1. 01

    FNOL arrives

    The notice is coded, deduplicated against open claims, and the policy is matched. Initial reserves are set from the carrier's own rules.

    Automated
  2. 02

    Policy and coverage pulled

    Policy terms, endorsements, prior-loss history, and reserve bands are retrieved from the policy admin and claims systems through supported APIs.

    Integration
  3. 03

    Documents ingested

    Photos, estimates, police reports, bills, and recorded statements are classified, OCR'd, and extracted into structured fields with a confidence per field.

    Automated
  4. 04

    Triage recommendation drafted

    The model scores severity, complexity, and fraud risk and proposes fast-track, routine, or SIU with the top contributing signals.

    Automated
  5. 05

    Low-confidence fields flagged

    Where extraction or triage confidence is below threshold, the system holds the recommendation for an adjuster rather than acting on it.

    Guardrail
  6. 06

    Adjuster reviews and decides

    The adjuster confirms coverage, accepts or overrides the triage, and signs off on the fast-track settlement or the next assignment.

    Human in loop
  7. 07

    Downstream actions fire

    Payment is issued through the claims system, subrogation demands are drafted where applicable, and the file is updated with the adjuster's reason code.

    Integration
  8. 08

    Audit trail logged

    Model version, inputs, confidence, and adjuster edits are logged against the claim ID. A regulator can reconstruct how any decision was reached.

    Guardrail

How exceptions are handled

  • If the policy cannot be matched with confidence, the file routes to a specialist rather than guessing coverage.
  • If fraud signals exceed threshold, no fast-track path is offered and the file routes to SIU.
  • If a required document is missing, the system drafts a documents-requested letter rather than inferring the value.
  • If a cat event is declared, triage switches to the carrier's cat protocol and fast-track thresholds tighten.

How it is engineered

Built on your stack, not ours.

Design choices that make this run against a real claims system, not a demo stack. We build to Guidewire, Duck Creek, Majesco, Sapiens, or your own policy admin, not around them.

01

Claims and policy integration

Supported APIs and messaging interfaces for Guidewire ClaimCenter, Duck Creek Claims, Majesco, and Sapiens. Writes land through the integration layer the carrier already governs.

02

Document pipeline

Classification, OCR, and extraction tuned to the claim document types that actually arrive: photos, estimates, police reports, medical bills, recorded-statement transcripts. Every extracted field carries a confidence and a source page.

03

Decisioning and triage

A decisioning layer that composes carrier-authored rules with model scores. Rules win where the carrier wants deterministic behaviour; models score where judgement is appropriate.

04

Identity and access

Single sign-on through the carrier's identity provider. Adjuster actions are attributable; the system never acts under a shared service account.

05

Model risk governance

Every model carries a documented risk tier, a validation pack, a monitoring plan, and the controls expected under NAIC's model AI bulletin and state DOI AI bulletins where applicable.

06

Observability and evaluation

Every triage recommendation, extraction, and SIU referral is logged with model version, inputs, confidence, and the adjuster's decision. An evaluation harness watches drift and flags regressions on business metrics.

What the business will measure

Numbers the operator already watches.

We agree on metrics and baselines before building. These are the numbers a claims and underwriting leader usually watches when deciding whether a system is working.

  • 01

    Cycle time to close on fast-track files

    Baseline today

    Measured today: median hours from FNOL to settlement on files eligible for fast-track, by line of business.

    How it moves

    Simple files are triaged and staged for same-day settlement; adjusters sign off instead of assembling.

  • 02

    Loss adjustment expense per claim

    Baseline today

    Allocated and unallocated LAE per claim, by line and severity band.

    How it moves

    Document re-reads collapse into one structured read; adjuster time shifts from assembly to decisioning.

  • 03

    Fast-track settlement rate

    Baseline today

    Share of eligible claims settled under the carrier's fast-track rules, by line.

    How it moves

    The triage model routes every eligible file to the fast-track path with the documented rationale.

  • 04

    SIU hit rate

    Baseline today

    Share of SIU referrals that result in a confirmed fraud finding or recovery.

    How it moves

    Universal screening replaces sampled referral; investigators work the files most likely to pay back.

  • 05

    Subrogation recovery rate

    Baseline today

    Dollars recovered as a share of recoverable loss, by state and loss type.

    How it moves

    Subrogation potential is flagged at FNOL; demand packets are drafted before the at-fault carrier closes its file.

Human in the loop

Oversight, compliance, and the audit trail

Where the humans stay in control.

The system is designed around what claims regulators, SIU investigators, model-risk committees, and policyholders already expect. Nothing is autonomous that should not be.

  • 01

    State fair-claims-practices

    Timelines, acknowledgement, and documentation requirements under each state's fair-claims-practices act are encoded as deterministic rules the system respects, not model behaviour.

  • 02

    NAIC and DOI AI governance

    Every model carries the documentation expected under NAIC's model AI bulletin and the state DOI bulletins that have adopted it: risk tier, validation evidence, monitoring plan, and governance owner.

  • 03

    FCRA for adverse action

    Where a decision affects a consumer under FCRA, the required notices and dispute paths are supported, and the model inputs that drove the outcome are retrievable.

  • 04

    Adjuster sign-off

    No settlement is issued, no SIU referral is logged, and no non-renewal is sent without the responsible adjuster or underwriter signing it in the system of record.

  • 05

    Attribution and audit

    Every automated recommendation logs the model version, the inputs, the confidence, and the adjuster's decision. Audits can reconstruct how any claim was handled.

  • 06

    Low-confidence refusal

    When extraction or triage confidence is below threshold, the system flags the case for a human rather than acting on a guess.

Proof

Work we have shipped, and the shape of what we build.

Real engagements where we can speak to them, and representative use cases where we cannot.

How the build runs

Scoped tightly, shipped in phases.

Fixed price per phase, approved before the phase starts. You own the code, the models, the data, and the deployment.

  1. 01

    Discovery

    Two weeks. We shadow claim handlers and underwriters, map the real workflow against your policy admin and claims system, and agree on the metrics we will move and how we will baseline them.

  2. 02

    Prototype on real files

    Four to six weeks. One line of business, one claim type, one proof of mechanism. Adjusters use it in parallel to real work; we measure against the baseline.

  3. 03

    Production rollout

    Eight to twelve weeks per line. Policy and claims integrations, SSO, model-risk documentation, DOI-ready audit trails, and the oversight workflow your model-risk committee signed off on.

  4. 04

    Handover

    You get the full source, the model-risk register, the validation evidence, the evaluation suite, and a two-week walkthrough with your engineering and claims operations teams.

  5. 05

    Optional ongoing support

    A fixed-term support agreement for model monitoring, policy and fee-schedule updates, and claims-system interface changes. Never a retainer dependency.

Common questions

Answers before the first call.

01Which claims and policy systems do you integrate with?
We build against Guidewire ClaimCenter, Duck Creek Claims, Majesco, and Sapiens through their supported APIs and messaging interfaces, and against custom and legacy policy admin via the integration layer you already govern. Writes land where your change control already lives.
02How does this meet NAIC and state DOI expectations on AI?
Every model we ship carries a documented risk tier, a validation pack, a monitoring plan, and a named governance owner, aligned to NAIC's model AI bulletin and the state DOI bulletins that have adopted it. The model-risk register is yours; your committee signs off before production.
03Will adjusters trust fast-track recommendations?
Trust is a workflow problem, not a model problem. Every recommendation carries its top contributing signals, its confidence, and a documented reason an adjuster can accept or override. We measure override rate as a product metric and tune to it.
04How do you handle fraud screening without over-referring to SIU?
The screening model is tuned to your SIU's historical outcomes, not a vendor benchmark. We measure precision at the top of the queue and work to the referral capacity you actually have, not to a theoretical maximum.
05What about catastrophe events, when adjuster capacity is swamped?
Cat protocols are a first-class part of the design. When a cat is declared, triage thresholds tighten, document requests pre-stage, and the fast-track path is governed by the cat rules your claims leadership has approved for that event.
See the other details on the FAQ page

Next step

A thirty-minute conversation. Bring one line of business where cycle time or LAE is where you want it to be, and we will walk through what the first phase would look like, where the measurements come from, and what has to be true for your model-risk committee to approve it.

Discuss your claims workflow

Helpful to have on the call

Helpful to have on the call: your claims and policy systems, one line of business with a documented LAE or cycle-time pain point, and the person who owns the metric you would like to move.

Explore

A different problem in every sector, one engineering standard underneath.

Featured · AI for healthcare

Healthcare

Why is your clinician still charting at midnight?

Your clinicians are charting until midnight and your front desk is drowning in prior auths. We build AI that writes clinical notes automatically, triages patients to the right care faster, and finds any record with a simple question, so your people spend their hours on patients instead of paperwork.

See AI for healthcare
stallwart / healthcare

Live run

Healthcare

Running
as of nowhealthy
See every industry we build for

An AI-First Engineering Company

Get a price in one call.

One 30-minute call. You leave with an architecture, a fixed price, and a start date. No charge, and no slide deck.

Booking now for December 2026

Book a callSee how it works