averoxi
Agentic claims · FNOL → settlement

The AI claims agent that takes the call, sees the damage and settles the claim.

Averoxi replaces the call centre and the adjuster visit with one autonomous, multimodal agent. It handles first notice of loss over voice and WhatsApp in the claimant's language, guides photo capture, estimates repair cost from imagery, scores fraud in real time and pays low-complexity claims straight through.

$3–20
per claim
<5 min
FNOL to decision
40+
languages & dialects
Averoxi Claims Agent
online · replies instantly
WhatsApp
Message… |

Drops into the stack carriers already run

  • WhatsApp Business
  • Voice / IVR
  • Guidewire ClaimCenter
  • Duck Creek
  • Sapiens
  • Majesco
  • Salesforce
  • Twilio
  • Audatex
  • Mitchell
  • M-Pesa
  • Flutterwave
  • Stripe
  • SAP
  • WhatsApp Business
  • Voice / IVR
  • Guidewire ClaimCenter
  • Duck Creek
  • Sapiens
  • Majesco
  • Salesforce
  • Twilio
  • Audatex
  • Mitchell
  • M-Pesa
  • Flutterwave
  • Stripe
  • SAP
The gap

The claims process was designed for phones with cords and cameras with film.

Every carrier feels it in two numbers: cycle time and leakage. Both come from the same place — a process that scales with headcount, not intelligence.

01

FNOL is a call centre.

Claimants wait on hold, repeat themselves, and get a case number. Every minute is churn risk — and every handoff drops context.

02

Assessment is a person in a car.

A physical adjuster visit costs $300–600 and takes days to schedule. For most claims the visit confirms what a photo already showed.

03

Cycle time bleeds customers.

Days-long claims are the #1 driver of non-renewal. Speed isn't a nice-to-have — it's retention.

04

Nobody has time to scrutinise.

Leakage — the money overpaid because no one could check — is a volume problem. Humans can't inspect every image. An agent can.

The agent

One multimodal agent. The entire claims journey.

Averoxi isn't a chatbot bolted to a form. It's an autonomous agent orchestrating speech, vision and decisioning models — with hard guardrails and a human in the loop wherever you want one.

Conversational FNOL, in their language

Voice or WhatsApp. The agent takes the story naturally, extracts the structured claim, and confirms coverage — in 40+ languages and dialects, code-switching included.

Yoruba → EN

Guided capture

Real-time on-device guidance tells the claimant exactly what to shoot — angle, distance, lighting — so every image is usable the first time.

Step back 2 m · Rear-left2 / 3

Vision-based estimates

Vision-language models map damage to parts and labour databases for a line-item estimate in seconds, with a confidence score.

Bumper cover, rearOEM · 71241-KE₦186,000
Tail lamp assy, LHOEM · 33551-TR₦94,500
Refinish + labour3.5 h × ₦37,800₦132,300
Estimate · 94% confidence₦412,800

Fraud signals, in-line

Image reuse, metadata inconsistency, staged-damage patterns and claimant velocity — scored before a human ever looks.

  • Image reuse0 matches across 2.1M claims
  • EXIF / GPSConsistent with FNOL location
  • Staged-damage patternLow similarity (2%)
  • Claimant velocity2 claims / 90 days

Straight-through settlement

Deterministic decision policies you author. Low-complexity claims are approved and paid instantly; everything else routes with a full dossier.

Decision policy · Motor / Low complexityv2.14
coverageconfirmed
estimate ≤ auto_limit412,800 ≤ 750,000
fraud_score0.08 < 0.35
actionSETTLE · PAY_INSTANT

Human-in-the-loop, by design

Adjusters supervise a queue, not a phone line. Every agent decision ships with its evidence, reasoning trace and confidence — override in one click, and the agent learns from it.

Explainable
Every decision has a trace
Auditable
Immutable claim dossier
Tunable
Thresholds per LoB
0%
reduction in FNOL handling time
call centre → agent
0 min
median FNOL-to-decision
low-complexity motor
0%
settled straight-through
no adjuster touch
$0
saved per avoided site visit
midpoint of $300–600
How it works

FNOL to payout, one autonomous loop.

Six stages, orchestrated end-to-end. Each stage emits evidence into the claim dossier, so nothing the agent does is a black box.

01 · Intake

Notice of loss, conversationally

Speech-to-text, entity extraction and policy lookup run while the claimant is still talking. Structured claim, zero form.

ASR · LLM · Policy API
claimantAmara O.
policyMTR-2291-04 · Comprehensive
perilThird-party collision
locationLekki toll, Lagos · 14:32
injuriesNone reported
Fraud intelligence

Scrutinise every claim. Not just the ones you have time for.

Leakage is what happens when volume beats attention. Averoxi scores every image, every metadata field and every relationship — in-line, before payment, at zero marginal cost.

100%
of claims scored
<900 ms
per image
8–14%
leakage reduction*
*Design-partner range, motor, 12-month window.

Image reuse

Perceptual hashing and embedding search against every image the carrier has ever received — and the open web.

Metadata inconsistency

EXIF timestamps, GPS, device fingerprints and edit histories cross-checked against the FNOL narrative.

Staged-damage patterns

Models trained on confirmed-fraud corpora flag implausible damage distributions and repeat repair-shop signatures.

Network & velocity

Graph features across claimants, garages, phone numbers and bank accounts surface organised rings early.

Lines of business

Motor is done. Everything else isn't.

Every emerging insurance market is growing without any straight-through claims capability. Averoxi is built line-agnostic from day one — the same agent, new parts tables, new decision policies.

Live

Motor

Collision, glass, theft. Parts and labour databases for 30+ markets.

Live

Property contents

Room-by-room inventory from video walkthroughs, valued against retail and depreciation tables.

Early access

Agricultural loss

Crop and livestock loss from field imagery and satellite indices; parametric triggers supported.

Early access

Marine cargo

Container and cargo damage from port photos, matched to bills of lading and survey standards.

Design partners

Health

Outpatient claims from receipts and discharge notes, with tariff validation and upcoding detection.

Pricing

Priced like an adjuster you never have to schedule.

Two models. Both are a rounding error next to a site visit.

Per claim

Usage-based
$3–20
per claim processed

Pay only for what the agent handles. Price scales with line of business and how far the claim travels — FNOL only, through assessment, or full straight-through settlement.

  • Conversational FNOL, voice + WhatsApp
  • Guided capture & vision estimates
  • Fraud signals on every claim
  • Adjuster console & dossiers
  • Core-system write-back

Licence + outcome share

Aligned incentives
Platform fee
+ share of verified leakage reduction

A fixed platform licence with a success fee tied to independently measured leakage reduction. We only win when your loss ratio improves.

  • Everything in Per claim
  • Custom decision policies per LoB
  • Private model fine-tuning on your corpus
  • Dedicated deployment region
  • Quarterly leakage audit & attestation
Trust & security

Built for regulated carriers, not demo day.

Claims data is the most sensitive data an insurer holds. We built Averoxi around that from the first commit.

Encrypted end-to-end

TLS 1.3 in transit, AES-256 at rest, customer-managed keys available.

Data residency

Deploy in-region — EU, UK, Nigeria, Kenya, South Africa, India, Brazil, GCC — to meet local regulatory requirements.

Model governance

Versioned models and policies, bias monitoring per segment, full decision traces for regulators.

SOC 2 & ISO 27001

Controls designed to SOC 2 Type II and ISO 27001; audit reports available under NDA.

Your data stays yours

No cross-customer training without explicit opt-in. Deletion on request, with certificate.

Human override, always

Any decision can be paused, reversed or escalated by a licensed adjuster in one click.

Put an agent on every claim. Starting with your next one.

We onboard a handful of carriers per quarter. Tell us your lines, markets and volume — we'll come back with a pilot plan inside a week.