TT Lab
Blog Get started
Learn Learning paths Courses

Insurance Domain Deep Dive

Claims and Payouts — The Loss Ratio Is the Report Card

Continue in TT Lab

In one line

Paying a claim is a procedure of notice, adjustment, and payment, and its result is tallied into report cards called the loss ratio and the combined ratio, while reinsurance shares large losses and fraud detection stops money from leaking out.

Why this knowledge is needed

Half of an insurance company's IT requirements come from the claims side. "Assign to adjustment within 3 business days of notice," "calculate late-payment interest automatically," "loss ratio dashboard by product, coverage, and channel," "add an SIU referral rule." If you do not know where these words come from, you can build the screens but cannot tell why those fields are needed or which numbers being wrong is an accident.

Notice, adjustment, and payment

Notice      The loss is reported, with claim form, medical certificate, receipts. The report date starts every deadline.
Adjusting   Is it covered by the policy, is an exclusion triggered, how much? Field investigation or medical review if needed.
Payment     The decided amount goes to the payee account. Paying past the policy's deadline adds late-payment interest.
Denial      Exclusion, breach of disclosure duty, or outside coverage. The reason must be notified; the customer may appeal.

For a case under adjustment, you set a case reserve. It is the amount the adjuster estimates will go out. It has not been paid yet, but from the company's point of view it is a loss that has already occurred, and it goes into the numerator of the loss ratio calculation.

The loss ratio and the combined ratio

loss ratio      = incurred losses ÷ earned premium
expense ratio   = expenses ÷ earned premium   (some use written premium as the base)
combined ratio  = loss ratio + expense ratio

Earned premium is the premium "earned" in this period. Even if a year's worth is received in advance, what was earned this quarter is only three months' worth. Incurred loss is the paid amount plus the estimated amount not yet paid, adjusted by the amount by which already-set estimates have changed. If the combined ratio exceeds 100%, it means the insurance operation by itself is running a loss, and it has to be covered by investment income. Rate increases, tighter underwriting criteria, and suspending sales are decided on this number.

There are many axes along which this number is aggregated. Product, coverage, channel, region, and accident year. It is the heaviest aggregation in the information layer, and if the period definitions of denominator and numerator are misaligned, the loss ratio moves by 1 percentage point at a time. You must confirm in the requirements whether it is "accident-basis" or "reporting-basis," and "earned" or "written."

Reinsurance

Large losses that a single company has difficulty bearing (a major fire, a typhoon, an aircraft accident) are shared through reinsurance. The primary insurer pays part of the premium to the reinsurer, and when a loss occurs, the reinsurer pays that proportion or the part above a certain amount. There are proportional methods (premiums and losses are shared in the same proportion) and non-proportional methods (the reinsurer bears only the excess over a certain amount). For IT, this comes down as "calculate the ceded ratio and expected recovery amount for each claim and reconcile with the reinsurer." The loss ratio is also looked at separately before ceding (gross) and after ceding (net).

Fraud detection

Insurance fraud quietly pushes up the loss ratio. Typical patterns are a claim right after enrollment, high-value claims concentrated at a particular hospital or repair shop, claims from different contracts that use the same payee account, contact number, or address, and a claim after many enrollments in a short period. Rule-based detection scores these patterns and refers them to the investigation unit (SIU). A rule is a basis, not a conclusion, and a denial decision is made only from the investigation result. Telling a customer "fraud is suspected" without a basis is itself a source of complaints and legal problems.

The shape in which this work comes down as IT requirements

The concepts above actually arrive as sentences like these. What to ask back and what to pin down differ for each.

"Please calculate late-payment interest automatically." There are three things to ask back. Whether the accrual start date is the notice date or the day all documents are complete, whether the interest rate is fixed by the terms or tied to market rates, and whether the period delayed through the customer's fault is subtracted. If you do not ask the last item, an objection that the amount does not match will come in later.

"Loss ratio dashboard by product, coverage, and channel." Pinning down the period definition comes first, and there is one more thing. When you split by coverage, how do you split the denominator? Premiums are received per contract but losses occur per coverage, so the units of numerator and denominator differ. Each company has a rule for allocating contract premium to coverages, and if you build it without knowing that rule, the numbers are entirely wrong.

"Please add an SIU referral rule." More important than the rule itself is a structure that leaves the judgment history. What score each rule gave and what the total was when the referral was made must remain so that when the rules are changed later you can see what differs. Cases that were investigated as false positives and ended as normal must also be kept to measure the rules' accuracy, and if you do not keep them, the rules can never get out of "we tried it and it seems to work."

"Please set the estimated amount for open claims automatically." This must be accepted carefully. The estimated amount goes directly into the loss ratio and is also reflected in accounting. It is safer to limit automatic calculation to suggesting an initial value and have the adjuster confirm it, and if you track the difference between the automatic value and the confirmed value as a metric, you can tell whether the suggestion is usable.

One more common point. Most of this work's numbers change later. Estimates are updated, denials are reversed on objection, and reinsurance recoveries come in late. So when you build an aggregation table, you must also store which point in time the number was viewed as of, and if you do not, you cannot explain why pulling last month's reported loss ratio again today gives a different value.

What really matters in practice

First, the notice date is the basis of every deadline. If the notice date is stamped wrong, the payment deadline, late-payment interest, and processing-time statistics are all wrong.

Second, an open claim is not 0 won. If there is no estimated amount, the loss ratio looks good, and that illusion bursts all at once in the next quarter.

Third, pin down the aggregation period definition in the requirements. Accident-basis versus reporting-basis, and earned versus written, are different numbers.

In the next lab, you will calculate loss ratios and combined ratios by product and quarter from contract and claim data, find three fraud patterns, and build an investigation referral list.