Carriers, MGAs, and brokers
AI worker teams for claims and policy operations
Claims economics are largely decided in the first ten minutes. Every field missing at first notice becomes a callback, a re-route, and a cycle-time day — so the highest-leverage intervention is completeness, not persuasion.
Our view
Where we think the real opportunity is
The instinct is to point AI at claim decisions, because that is where the expertise is. We think that is the wrong end. Coverage and liability are regulated judgements with real consequences for being wrong, and they are also not where most of the cost sits. The cost sits in intake that arrives incomplete, gets re-routed twice, and forces a distressed claimant to retell a bad day three times. That failure is structural, not a skill problem, and it is exactly the kind of thing a consistent conversational layer fixes. So the design is deliberately unambitious about judgement and aggressive about structure: capture the event once, completely, route it with recorded reasoning, and hand your adjuster a file rather than a voicemail.
Operating reality
How demand actually behaves here
Volume has a shape, and the shape is what breaks teams. This is what we assume about your operation before we design anything.
Volume is event-driven and violently spiky
A storm or flood generates a month of claims in two days. Staffing for that peak is uneconomic and staffing below it means the queue becomes the customer experience precisely when the brand is most exposed.
Claimants contact you on their worst day
First notice is often given in shock or distress. Consistency and patience matter more than efficiency, and the cost of making someone repeat the narrative is reputational as well as operational.
Claims are multi-party by nature
Claimant, third party, repairer, medical provider, and broker all touch one file. Context that lives in one person's memory of one call is context that will be lost.
The process is documentation-bound
Progress depends on evidence arriving, and evidence arrives late because the request was vague, mistimed, or sent to the wrong party.
Where the hours go
The cost we would try to move first
Almost every operation spends its most expensive time on its least valuable work. Naming that precisely is what makes a pilot measurable rather than impressive.
Intake repeated at every handoff
Unstructured first notice means the second and third person on the file re-gather what the first already heard, which is pure duplicated cost plus claimant friction.
Evidence chasing
A significant share of claims-handling time is spent asking again for documents, in sequences that could be determined from the claim type at the outset.
Mis-routing to the wrong queue
A claim that lands in the wrong specialist queue loses days and often gathers a second incorrect assessment before it is corrected.
Leakage from slow triage
Delay increases severity — mitigation not authorised in time, hire periods extended, minor injury claims that harden. The cost of slow triage rarely appears in the contact-centre budget where it is caused.
The trust envelope
What constrains the design, and what never gets automated
We design to the boundary first. A control added after launch is a control nobody believes — and in this sector the boundary is not negotiable.
Sector constraints
No coverage representations
Any statement that implies a loss is or is not covered creates exposure. This has to be excluded from the workers' scope structurally, not discouraged by prompt wording.
Licensure for adjusting activity
Assessing and settling claims is a licensed activity in most jurisdictions. Intake, verification, and evidence collection are not — and that line is the design boundary.
Sensitive and medical data
Injury claims carry health information into a contact-centre workflow, which raises the access-control bar and narrows what any worker should ever see.
Fraud handling protocol
Indicators must be captured for specialist review without being raised with the claimant, and without becoming an accusation recorded in the wrong place.
Stays human, permanently
These are structural boundaries, not trust levels waiting to be relaxed.
Coverage and liability
The determination that a policy responds is the regulated core of the business. It stays with a licensed handler, working from a complete file.
Reserving and settlement
Setting reserves and agreeing settlement are financial judgements with reporting consequences, and they require accountability a model cannot carry.
Any fraud conclusion
Signals are inputs to a specialist investigation. Concluding that a claim is fraudulent is a human decision with legal weight.
Injury and emergency conversations
Where someone is hurt or at immediate risk, intake stops and a person takes over on that turn.
Where to start
Sequencing matters more than scope
The first blueprint should be the one whose success criteria your team already agrees on. Each step below links to the blueprint that implements it.
Start
One claim type, human-reviewed
Take your highest-volume, lowest-complexity claim type on one channel. Workers capture and structure; a handler reviews every file before it moves. What you are proving here is completeness, not autonomy.
Read the blueprintThen
Add evidence chase and recorded triage
Turn on claim-type-specific evidence collection and triage recommendations. Handlers still confirm routing, which is what builds the rationale baseline you will later rely on.
Read the blueprintNext
Premium arrears and new business
Arrears recovery and new-business intake reuse the identity gating and approval thresholds already reviewed for claims, so the second and third blueprints inherit the governance work.
Read the blueprint
Systems we would expect to read and write
- Policy administration system
- Claims management platform
- Document and media store
- Broker and partner portals
- CRM and complaints case management
Blueprints that apply
3 operating blueprints for Insurance
Each one carries the full team, process, and control detail — the mechanics this page deliberately does not repeat.
Money & regulated intake
Turn first notice of loss into a complete, correctly routed claim
First notice of loss
Read the blueprintMoney & regulated intake
Recover overdue balances without losing the customer
Collections & payment recovery
Read the blueprintDemand & growth
Qualify and onboard new matters without spending billable hours on admin
Client intake & onboarding
Read the blueprint
Push back on this
Objections we actually hear in Insurance
Answered as we would answer them in the room, including the cases where the right answer is that this is not for you.
- An AI cannot handle a claimant who has just had a serious accident.
- It should not try. Injury and emergency language triggers human takeover on the turn it appears, with the transcript and policy context transferring so the person picking up does not restart the conversation. The blueprint is aimed at the large volume of first notices that are stressful but not emergencies — a stolen phone, a leaking pipe, a bumper in a car park — where consistency and not having to repeat yourself is the actual kindness.
- Our claim types are too varied for this to work.
- Variety is why we scope by claim type rather than by department. One type, one channel, human review on every file — then widen. If your portfolio genuinely has no high-volume repeatable type, this is the wrong first investment for you and we would rather say so.
- Regulators require a human to adjudicate claims.
- Agreed, and the blueprint is built on that assumption. Coverage, liability, reserving, and settlement are reserved for licensed handlers with no automation path around them. What is automated is intake, policy verification, evidence collection, and a routing recommendation that records the signals behind it — none of which is adjudication.
Other sectors
How the argument changes elsewhere
Providers, clinics, and health systems
AI worker teams for patient access operations
Read the analysisRetailers, marketplaces, and D2C brands
AI worker teams for order and delivery support
Read the analysisNetwork operators and retail energy
AI worker teams for outage and service operations
Read the analysis
Next step
Argue with this in a working session.
Bring your own numbers. We will map your version of the operating reality, agree what must stay human, and scope the first blueprint against a measurable win — or tell you it is not worth doing yet.