Agentic InsuranceAI employees that read, check and route every claim
Software for insurers. AI employees read claim documents, check them against the policy, score fraud risk and send each claim down the right path. Your adjusters approve what matters.
The problem
Insurance still runs on paperwork and hand-overs.
From the first lead to the final claim, a policy passes through many teams and systems. Every hand-over adds re-typing, waiting and room for error. Customers feel each delay, most of all when they make a claim.
One policy, six hand-overs
- 01 Lead
- Delay
- 02 Sale
- Re-typed
- 03 Policy issued
- Bottleneck
- 04 Premium paid
- Hand-over
- 05 Servicing
- Delay
- 06 Claim
- 01
Manual, repetitive work
Staff type the same data from forms, invoices and medical summaries into several systems.
- 02
Teams and systems that do not connect
Sales, policy administration, finance and claims each hold part of the picture, so work waits at every hand-over.
- 03
Operating costs that keep rising
More claims are met with more people, not with better processes. Medical costs keep rising too.
- 04
The claim decides loyalty
A fast, clear claim is the moment a customer decides to stay. A slow or confusing one is when they leave.
01 Manual, repetitive work
Staff type the same data from forms, invoices and medical summaries into several systems.
02 Teams and systems that do not connect
Sales, policy administration, finance and claims each hold part of the picture, so work waits at every hand-over.
03 Operating costs that keep rising
More claims are met with more people, not with better processes. Medical costs keep rising too.
04 The claim decides loyalty
A fast, clear claim is the moment a customer decides to stay. A slow or confusing one is when they leave.
AI employees
One AI employee for each job.
Each AI employee has one job description and its own access rights, and works only on that job. Together they cover the whole policy lifecycle. Every one of them reports to a person on your team.
Acquisition
Find and qualify the right prospects.
-
Prospect researcher
Analyses prospect data. Personalises digital campaigns. Scores leads.
-
Market analyst
Spots market trends. Segments customers.
Sales
Help prospects choose and apply.
-
Product advisor
Suggests products that fit the customer. Finds cross-sell and up-sell opportunities. Suggests product bundles.
-
Sales assistant
Answers prospects in chat straight away. Qualifies prospects. Helps them finish the application.
Issuance & payment
Underwrite, issue the policy and collect premiums.
-
Policy issuer
Prepares underwriting checks. Creates the digital policy document. Validates applicant data.
-
Payment manager
Processes premium payments securely. Flags suspicious payments. Sends bills on schedule.
Servicing
Answer policyholders and keep records current.
-
Service agent
Answers questions at any hour. Resolves first-level requests. Escalates to the right person.
-
Account keeper
Updates policyholder data. Records preferences. Sends notices before they are needed.
Claims
Assess claims and investigate the unusual ones.
-
Claim assessor
Reads claim documents and damage photos. Estimates the claim amount. Verifies the documents.
-
Claim investigator
Detects unusual patterns. Gathers digital evidence. Recommends a decision to the adjuster.
Retention
Keep policies in force and renew them.
-
Retention analyst
Spots policies at risk of lapsing. Reads customer sentiment. Recommends retention actions.
-
Loyalty manager
Personalises rewards. Sends renewal offers on time. Analyses customer feedback.
Oversight
-
Supervisor
An AI supervisor checks the quality of every AI employee’s work and reports it to the people responsible on your team.
01 Acquisition
Find and qualify the right prospects. Prospect researcher: Analyses prospect data. Personalises digital campaigns. Scores leads. Market analyst: Spots market trends. Segments customers.
02 Sales
Help prospects choose and apply. Product advisor: Suggests products that fit the customer. Finds cross-sell and up-sell opportunities. Suggests product bundles. Sales assistant: Answers prospects in chat straight away. Qualifies prospects. Helps them finish the application.
03 Issuance & payment
Underwrite, issue the policy and collect premiums. Policy issuer: Prepares underwriting checks. Creates the digital policy document. Validates applicant data. Payment manager: Processes premium payments securely. Flags suspicious payments. Sends bills on schedule.
04 Servicing
Answer policyholders and keep records current. Service agent: Answers questions at any hour. Resolves first-level requests. Escalates to the right person. Account keeper: Updates policyholder data. Records preferences. Sends notices before they are needed.
05 Claims
Assess claims and investigate the unusual ones. Claim assessor: Reads claim documents and damage photos. Estimates the claim amount. Verifies the documents. Claim investigator: Detects unusual patterns. Gathers digital evidence. Recommends a decision to the adjuster.
06 Retention
Keep policies in force and renew them. Retention analyst: Spots policies at risk of lapsing. Reads customer sentiment. Recommends retention actions. Loyalty manager: Personalises rewards. Sends renewal offers on time. Analyses customer feedback.
+ Oversight
Supervisor: An AI supervisor checks the quality of every AI employee’s work and reports it to the people responsible on your team.
Role names are examples. You choose which AI employees to start with.
How it connects
It works inside your systems, under your rules.
Agentic Insurance is software you run. Matajari delivers it and sets it up with your IT team, connected to the systems you already have.
- 01
Channels
Where requests arrive from customers, agents and staff.
Chat · Voice · Web · Email
- 02
Orchestration
Gives each task to the right AI employee and applies your rules and approval steps.
- 03
Language model layer
Reads, writes and reasons. Models can be swapped without rebuilding the rest, so you are not tied to one AI vendor.
- 04
Integration layer
Connects through secure interfaces and your workflow engine.
- 05
Your core systems
Stay the system of record.
Policy administration · Claims · Payments · Customer records
-
Sensitive data stays with you
Health records and personal data stay in your own data centre or private cloud.
-
Heavy work can scale out
Workloads that need more computing power, such as reading large batches of documents, can run in a cloud environment you approve.
How a claim moves
From claim to payment, with people at the gate.
Every claim follows the same path. Simple claims go straight through. Anything complex or unusual goes to an adjuster, with the groundwork already done.
- 01
Claim comes in
The customer sends photos of the invoice, medical summary and ID card by web, chat or email.
- 02
Documents are read
Text recognition (OCR) reads each page and picks out the fields that matter.
- 03
Claim is checked
Coverage, benefits and limits are checked against the policy. The claim gets a complexity score and a fraud-risk score.
- 04
Decision gate
Your rules choose the route: confidence, claim value and risk score against thresholds your team sets.
- Simple
Straight through
High confidence, low value and low risk: approved automatically. Your team still reviews a sample of these.
- Medium
Fast track
Moves to the front of an adjuster’s queue, with the documents already summarised.
- Complex or suspicious
Manual review
An adjuster or investigator decides, with the AI’s findings and evidence attached.
Decision and payment
The customer hears the outcome and approved claims are paid. Every step is on record.
- 01
Claim comes in
The customer sends photos of the invoice, medical summary and ID card by web, chat or email.
- 02
Documents are read
Text recognition (OCR) reads each page and picks out the fields that matter.
- 03
Claim is checked
Coverage, benefits and limits are checked against the policy. The claim gets a complexity score and a fraud-risk score.
- 04
Decision gate
Your rules choose the route: confidence, claim value and risk score against thresholds your team sets.
Three routes
SimpleHigh confidence, low value and low risk: approved automatically. Your team still reviews a sample of these.
MediumMoves to the front of an adjuster’s queue, with the documents already summarised.
Complex or suspiciousAn adjuster or investigator decides, with the AI’s findings and evidence attached.
- 05
Decision and payment
The customer hears the outcome and approved claims are paid. Every step is on record.
The thresholds and the size of the review sample are your own settings.
What it checks
Three checks on every claim.
Before a claim reaches the decision gate, the AI reads the documents, looks for signs of fraud and checks the claim against the policy.
Reads the paperwork
Photos and scans become structured data, so nobody has to type them in.
- 01
Collect
Invoices, medical summaries, diagnosis letters and ID cards arrive as photos or scans.
- 02
Recognise
Text recognition (OCR) turns each page into text.
- 03
Understand
The layout is read to find fields such as patient name, service date, diagnosis and amount billed.
- 04
Validate
Rules and models check each field and give it a confidence score. Low scores go to a person.
01 Collect
Invoices, medical summaries, diagnosis letters and ID cards arrive as photos or scans.
02 Recognise
Text recognition (OCR) turns each page into text.
03 Understand
The layout is read to find fields such as patient name, service date, diagnosis and amount billed.
04 Validate
Rules and models check each field and give it a confidence score. Low scores go to a person.
Looks for fraud four ways
Each claim is scored from several angles at once, and every score comes with the reasons behind it.
- 01
Learns from past cases
A model trained on your own confirmed fraud cases.
- 02
Spots outliers
Flags claims that do not fit the normal pattern.
- 03
Maps relationships
Finds unusual links between members, providers and agents.
- 04
Reads the text
Notices odd wording or document formats.
01 Learns from past cases
A model trained on your own confirmed fraud cases.
02 Spots outliers
Flags claims that do not fit the normal pattern.
03 Maps relationships
Finds unusual links between members, providers and agents.
04 Reads the text
Notices odd wording or document formats.
Checks the claim against the policy
The same policy rules are applied the same way to every claim.
- 01
Coverage
Is the treatment covered by the policy and its riders?
- 02
Benefit
Which benefit does the claim fall under?
- 03
Limits
How much of the annual and per-event limit is left?
- 04
Waiting periods and exclusions
Does any waiting period or exclusion apply?
01 Coverage
Is the treatment covered by the policy and its riders?
02 Benefit
Which benefit does the claim fall under?
03 Limits
How much of the annual and per-event limit is left?
04 Waiting periods and exclusions
Does any waiting period or exclusion apply?
Controls
People approve what matters.
AI employees prepare, check and recommend. Your people make the decisions that carry risk, and every step is on record.
AI employees can
- Read and sort incoming claims
- Check documents, coverage and limits
- Score risk and recommend a decision
- Approve simple, low-value claims within the limits you set
Your people decide
- Claims above your thresholds
- Every decline
- Every suspected fraud case
- Any change to rules or thresholds
Risks designed out from the start
01 Inconsistent explanations
Answers come only from a knowledge base you approve, and are checked against compliance rules before they go out.
02 Wrong answers
The AI looks things up in curated sources and declines questions outside its remit instead of guessing.
03 Wrong actions in your systems
Approval steps, a separate test environment and a full log of every action.
04 Personal data exposure
Personal data is masked, encrypted and visible only to the roles that need it.
05 Dependence on one AI vendor
A model layer that lets you change AI models without rebuilding.
Built-in controls
- An audit trail of every decision and every human override
- Role-based access and data-loss prevention
- Monitoring of model accuracy and drift, with rollback
- Bias checks and explanations for each recommendation
- Consent records and data-retention rules
Built to support your obligations under
- Law No. 27 of 2022 on Personal Data Protection (UU PDP)
- POJK No. 4/2021 on IT risk management for non-bank financial institutions
- POJK No. 22/2023 on consumer protection in the financial services sector
Compliance stays your responsibility as the licensed insurer. The software gives you the controls and records to show it.
Implementation
Start small. Grow claim by claim.
A phased rollout lowers risk and lets your team learn at every step. Each phase delivers something your claims team can use before the next one starts.
- 01 Phase 1
Foundation
Business case, data inventory, target architecture and an AI governance policy.
- 02 Phase 2
Document reading
Digital claim intake and automatic data extraction for reimbursement claims.
- 03 Phase 3
Triage and fraud scoring
Claims routed by complexity and risk, with model monitoring from day one.
- 04 Phase 4
Automatic approval
A rules engine approves low-risk claims, and your team reviews a sample.
- 05 Phase 5
Assistants and the wider lifecycle
An internal knowledge assistant, medical summaries, and AI employees for sales, servicing and retention.
The data we work with
- 01
Policy and benefit data
Coverage, riders, limits, waiting periods and exclusions.
- 02
Claim records
Dates, claim types, amounts, status and timestamps.
- 03
Claim documents
Invoices, medical summaries, diagnosis letters and ID cards.
- 04
Provider and medical data
Diagnosis and procedure codes, the provider network and tariffs.
- 05
Past outcomes
Which claims were approved or declined, and which were confirmed as fraud.
01 Policy and benefit data
Coverage, riders, limits, waiting periods and exclusions.
02 Claim records
Dates, claim types, amounts, status and timestamps.
03 Claim documents
Invoices, medical summaries, diagnosis letters and ID cards.
04 Provider and medical data
Diagnosis and procedure codes, the provider network and tariffs.
05 Past outcomes
Which claims were approved or declined, and which were confirmed as fraud.
Your programme team
- Business sponsor
- Product owner
- Architecture lead
- Integration lead
- Insurance specialist
- Compliance lead
- Quality supervisor
Where we suggest starting
Small-value health claims, with document reading first and people approving every decision until the results earn trust.
Frequently Asked Questions
Everything you need to know about Agentic Insurance.