Walk me through a claim from first notification to closure, and tell me what the reserve is doing at each stage.

Notification and the first estimate

The lifecycle begins with first notification of loss, universally abbreviated to FNOL. Someone reports that something has happened: a date of loss, a description, a location, possibly injuries and third parties. At this point you know very little, and the system's job is to capture the loss date accurately above almost everything else, because the loss date determines which policy term and which terms of cover govern the claim. A claim notified today against a loss eight months ago is priced, covered and reserved under the contract as it stood eight months ago.

Setting up the claim creates one or more features, sometimes called claim lines or coverage lines: separate buckets for vehicle damage, third-party property, bodily injury, and so on. Each feature carries its own reserve, because they behave completely differently over time. Vehicle damage resolves in weeks against a repair estimate. Bodily injury can stay open for years while a prognosis develops.

The initial reserve is often set from a table of average costs by claim type, because nobody has looked at the claim yet. That is a legitimate placeholder, but it means the earliest reserve figures carry almost no information about the individual claim.

What the reserve is, and why it changes

The case reserve is the handler's current estimate of what remains to be paid on that feature. The identity that anchors everything is:

incurred = paid to date + outstanding reserve

Paid moves up as payments go out. Outstanding moves down as those payments consume the estimate, and moves up or down independently as the estimate itself is revised. Incurred is what the business reports as the cost of the claim, and it is the number that flows into loss ratios and into the actuarial triangles.

Reserves change for reasons worth being able to list, because "the reserve went up" is not an explanation anyone accepts:

• Information arrival. An engineer's report, a medical report, an invoice, a liability admission. The commonest reason and the most benign.
• Prognosis change. Chiefly in injury claims, where recovery is slower than expected or a further procedure becomes necessary.
• Litigation. A claim entering proceedings acquires legal cost and a wider range of outcomes.
• Severity drift. Repair, care and legal costs rise between loss and settlement, and long-tail claims are exposed to that for years.
• Coverage or liability reassessment. A partial liability split, a policy limit becoming binding, an exclusion coming into play.

Each of those is a reason code on a reserve transaction, and the transaction is what you store. Overwriting the outstanding figure in place destroys the history of estimation, which is exactly what reserving and management reporting need.

Adjudication

Adjudication is the decision-making core: is this covered, for how much, and to whom. Coverage asks whether the policy in force at the date of loss responds to this event, checking perils, exclusions, conditions, whether the claimant is a covered party and whether the loss falls inside the term. Quantum asks what the loss is worth: the indemnity basis, applicable limits and sub-limits, and the deductible or excess. The excess is the insured's retained portion and can be applied either by deducting it from the settlement or by collecting it from the insured, which changes the cash flows even though the net indemnity is identical. A declined claim is not a deletion; it remains with a decision, a reason and a reserve released to zero, because declinatures get challenged and complaints get upheld.

Handling costs stay separate from indemnity. Loss adjusters, engineers, solicitors and medical experts are claim expense, and the industry further distinguishes expense attributable to a specific claim from expense that is not. Blending them into indemnity ruins both loss ratio analysis and reserving.

Payment

A payment is issued against a specific feature, to a payee who is often not the insured: a garage, a hospital, a third party's solicitor. It reduces the outstanding reserve on that feature by the amount paid, unless the reserve is revised at the same time. Partial payments are the norm on anything long-running, and interim payments on injury claims are made long before the final settlement is known. Once a payment has cleared, correcting a mistake is a reversal transaction and never an edit.

Recovery, and why subrogation is the piece candidates skip

Subrogation is the insurer stepping into the insured's rights to pursue whoever caused the loss, having indemnified the insured. If your policyholder was hit by another driver, you pay your policyholder and then pursue that driver's insurer for what you paid. Salvage is the related but distinct case of realising value from the damaged property itself, typically a written-off vehicle.

Recoveries are not income, and modelling them as income is a mistake with consequences. They reduce the net cost of the claim, and reporting must show gross incurred, recoveries and net incurred separately, because the actuary needs the gross figure to see the underlying loss experience while finance needs the net. Recoveries also arrive late, often after the claim would otherwise have closed, which is why recovery is usually a separate workflow with its own queue, its own anticipated-recovery reserve and its own ageing.

Closure is a state, not an ending

A claim closes when there is nothing outstanding: reserve at zero, payments made, recoveries concluded or written off. It can reopen, and long-tail claims do. A settled injury claim reopens when the claimant's condition deteriorates; a property claim reopens when a further defect emerges from the same event. The reopening is a new dated movement, not a revision of the closure, and the original history survives it untouched.

The edge worth tracing is the one leaving closure and re-entering adjudication, because it is the transition that forbids a destructive update anywhere in the model.

stateDiagram-v2
    state "Coverage review" as Review
    state "Recovery open" as Recovery
    [*] --> Notified
    Notified --> Review
    Review --> Declined
    Review --> Adjudicated
    Adjudicated --> Paying
    Paying --> Recovery
    Paying --> Closed
    Recovery --> Closed
    Declined --> Closed
    Closed --> Reopened
    Reopened --> Adjudicated
    Closed --> [*]

That constraint is where a claims data model earns its keep, because the question the business asks is never "what is this claim worth now". It is "what did we think this claim was worth at the end of last quarter", asked of a claim that has since closed, reopened and closed again.