Agentic FabriqRestrict access by policyholder, claim, product, jurisdiction, and assigned team.
Scope resolves from the case in front of the agent — one policyholder, one claim — not from the carrier’s whole book.
SOLUTIONS / Insurance
Safely automate decisions that affect coverage and claims
Govern agents across underwriting, claims, policy servicing, and sensitive customer records.
01 / IDENTITY02 / POLICY03 / AUDIT
01 / THE CONTEXTMove the claim along.
Move the claim along.
Hand the decision to the adjuster.
Insurance agents work with personal, medical, financial, property, and behavioral data. They may summarize submissions, assess claims, recommend coverage, request documentation, or update policies.
CLAIM CL-77310 · EXAMPLE
claim documents
claim documents- 01First notice of lossthe new claim, summarisedclaims-agent
- 02Triage & routingthis policy, this claimclaims-agent
- 03Document analysisinside the case’s own filesclaims-agent
- 04Adjudicationcoverage decisionadjuster
- 05Settlementpayment, coverage changeadjuster
Between 03 and 04 the agent stops assembling and a person decides.
02 / IN PRACTICE
Autonomy, with accountability.
THE AGENT’S OUTBOX · CLAIM CL-77310Illustrative · claims-agent, acting for a claims handler
- sentTo the policyholder · this claimThe three documents still missing from your claim
Assembled from the claim’s own file and sent under the handler’s permissions. The message, its inputs and both identities are on the record.
Outlook - sentTo the field assessor · this claimInspection slot confirmed for the property
Scheduling is routine work inside the case, so it goes out without waiting for anyone.
Twilio - not sentTo the broker · this claimClaim summary, with the claimant’s medical report attached
Blocked. Field-by-field limits keep medical detail out of a workflow with no approved reason to use it — the attachment is never retrieved, so there is nothing to send.
Outlook - not sentTo another claimant · claim CL-77482Status of your claim
A different handler’s claim. Scope resolves from the case in front of the agent, so the file cannot be read and the message cannot be written.
Outlook - heldTo the policyholder · this claimSettlement offer for the damage assessed
Coverage and payment are the adjuster’s to decide, and neither is an action this agent holds. The draft stays where it is because there is no tool to send it.
NOT IN THE GRANTOutlook
Example messages. The cases, sensitive fields and grants are the ones you set.
03 / WHAT CHANGES
Control, in the details.
Separate data collection and recommendation from final decisions.
Agents assemble and suggest; decisions that affect coverage or payout remain with adjusters and underwriters.
Keep denials, pricing changes, settlements, and coverage modifications out of the grant.
The decisions regulators scrutinize hardest are not actions any agent holds, so they stay with the people who have the authority to make them.
Limit which sensitive details an agent may use in a workflow.
Field-by-field limits keep medical, financial, or behavioral data out of workflows that have no approved reason to see it.
Decide whether an agent may send externally at all, or only draft.
Draft and send are separate actions, so an agent can prepare a letter to a policyholder, broker, or third party and have no way to send it.
Record the inputs and the authority behind each outcome.
Every claim and policy outcome carries what informed it, which agent acted, and whose authority it ran on — market-conduct evidence by default.
THREE LAYERS. ONE DECISION.
The boundary travels with the work.
Explore the context attached to every request.
What is this agent allowed to do?
claims-agent has a defined purpose and a bounded set of tools.
Whose authority travels with it?
The request acts for claims handler, using that person’s permissions.
Where does the action land?
The policy applies to the requested action within Claims platform, before the tool executes.
04 / WAYS TO PUT IT TO WORK
Built for your real workflows.
01Claims intake
Intake agents summarize and route each new claim, limited to that claim’s policy and documents alone.
Identify→Authorize→Record
02Underwriting assistance
Assistants assemble submissions and flag risk factors; acceptance and pricing stay with the underwriter.
Identify→Authorize→Record
03Policy servicing
Routine changes are granted and apply automatically; coverage-affecting changes are not granted at all, so they stay with authorized staff.
Identify→Authorize→Record
04Fraud review
Review agents work assigned referrals read-only, preserving evidence and avoiding cross-case contamination.
Identify→Authorize→Record
05Broker support
Broker-facing agents answer from approved product and appointment data, without exposing other producers’ business.
Identify→Authorize→Record
06Document analysis
Analysis stays inside the case’s own documents, with sensitive details gated by purpose.
Identify→Authorize→Record
QUESTIONS
A closer look.
Does Fabriq replace our claims or policy systems?+
No. It drops in between your agents and the platforms you already run. Agents keep working in your claims and policy systems; Fabriq decides, per request, what they can read and which actions they can take there.
How do we keep sensitive details out of the wrong workflows?+
Field-by-field limits. A workflow with no approved reason to see medical or behavioral data simply cannot retrieve it — the block is enforced on every request, not requested politely in a prompt.
What happens when staff change roles?+
Grants resolve org → team → member, so authority follows the person — and revoking a session cuts their agents’ access immediately.
Can we choose what each adjuster’s or underwriter’s agent can do?+
Yes — per user. The same agent can assemble files under a junior adjuster’s tight limits and act more broadly for a senior underwriter, because reach resolves from the person the agent works for. Denials, settlements, and coverage changes stay with humans because no agent is granted them.
Can we stop an agent touching a payout?+
Yes — an adjuster’s agent can read and draft freely while payment and coverage-change actions stay out of its grant, so they never appear in its tool list. Every call it does make is part of the record.
How do agents connect to our systems?+
Through the hub — prebuilt connectors plus OpenAPI import for internal claims and policy APIs, with credentials held in the vault, not in bot configs.