Why Underwriting KRIs Need a Design, Not Just a Dashboard
A key risk indicator (KRI) is a measure selected to provide an early or continuing signal about a risk exposure. For an underwriting portfolio, a useful KRI does more than report a number: it identifies what is being monitored, how the number is produced, what level warrants attention, and who must respond. A dashboard full of metrics without these decisions can show activity without supporting risk management.
This workshop builds a hypothetical KRI set for an insurer’s commercial property portfolio. The figures and thresholds are illustrative, not industry benchmarks. In practice, thresholds should be supported by the organization’s risk appetite, historical experience, business plan and data limitations, and approved through appropriate governance. The approach is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.
A KRI is not automatically a risk limit. A limit states a boundary on an exposure or activity; a KRI is a signal used to assess whether risk conditions may be changing or require attention. A KRI can support monitoring against a limit, but a threshold breach does not by itself establish that a limit has been breached or that a loss will occur.
Build Each Indicator Around a Decision
Start with a material underwriting concern, then select a measure that is relevant, reproducible and timely enough to inform a decision. For every percentage, specify the numerator and denominator. For every count or rate, specify which policies, claims or transactions enter the calculation and the period to which they belong. Without those details, two reports can display different values while both appear to use the same metric name.
The following specifications illustrate how a monitoring set can connect several views of underwriting activity. Thresholds are deliberately identified as hypothetical and should not be transferred to another portfolio without validation.
| KRI | Calculation and population | Source and owner | Illustrative threshold and response |
|---|---|---|---|
| Referral override rate | Bound policies with a recorded referral override divided by all bound policies in the month | Policy administration and referral log; underwriting operations manager | Green at 4% or below; amber above 4% through 7%; red above 7%. Amber: review causes within five business days; red: notify the underwriting executive the same business day. |
| Early reported claim frequency | Claims reported within 90 days after inception, per 1,000 earned exposure-years for policies that have completed the full 90-day observation window | Claims system linked to policy exposure records; portfolio actuary | Compare with a defined plan level and maturity rule. An amber signal prompts validation and segment review; a red signal prompts prompt senior underwriting review. |
| Quote-to-bind rate | Bound quotes divided by eligible quotes issued in the period, using a consistently defined quote cohort | Quote platform linked to policy records; distribution or underwriting manager | A lower-than-planned rate can signal changed selection, pricing or distribution conditions. The owner checks the same cohort and investigates with other indicators before recommending action. |
The source must support the stated definition. For example, a referral override rate cannot be trusted if overrides are entered inconsistently or if the log excludes some binding channels. Data lineage and reporting lag also matter: a monthly number based on incomplete records may be less useful than a slightly later, reconciled number. Assign one accountable owner for reviewing the result, even when another team produces the data.
Worked Example: Specify and Classify an Override KRI
A hypothetical commercial property portfolio has 400 policies bound in April. The referral log records 20 bound policies that proceeded with an underwriting referral override. The approved illustrative thresholds are green at 4% or below, amber above 4% through 7%, and red above 7%. Calculate the indicator, classify it, and state the appropriate response.
Method. Divide the count of overridden bound policies by all policies bound in the same month, then apply the thresholds to the resulting percentage.
The numerator is 20 overridden bound policies. The denominator is 400 bound policies. Thus:
A rate of 5% is greater than 4% and no greater than 7%, so it is amber. The underwriting operations manager owns the review. Under the illustrative response rule, the manager reviews the reasons for the overrides within five business days and reports whether the pattern is concentrated in a particular rule, team or product feature. The calculation checks: 5% of 400 is 20, which matches the observed override count.
Common mistake avoided. Dividing by only the policies referred would answer a different question: the proportion of referrals overridden, not the share of all bound policies with an override. The denominator must match the stated KRI definition.
Set Thresholds and Escalation Before the Signal Arrives
A threshold turns a measure into a monitoring trigger, but a threshold is only useful when its meaning and response are clear. A good design records which comparison is being made, whether the boundary is inclusive, what period applies, and whether the threshold is a warning or a formal limit. Thresholds may be based on approved risk appetite, business-plan expectations, historical variation or expert judgment; the rationale should be documented and reviewed rather than adjusted informally after an unfavorable result.
An escalation rule should identify the recipient, timing and expected next step. For example, an amber reading might require the owner to validate the data and prepare a short explanation for the underwriting manager within five business days. A red reading might require same-day notification to the underwriting executive, followed by a documented review. These actions do not presume that the KRI is correct or that underwriting must immediately change; they establish who must assess the signal and when.
Before escalating a surprising result, check that the reporting period, portfolio definition, denominator, data completeness and threshold version are correct. If the metric is valid, investigate plausible explanations and relevant segments, and document the conclusion. Do not silently change the definition or threshold to make the result appear acceptable.
Evaluate Indicators Together
Individual KRIs show different parts of underwriting performance. An override rate can signal a change in adherence to referral rules, early claim frequency can signal a change in observed claim experience, and quote-to-bind rate can signal a change in the flow of business. A movement in one is not proof of a particular cause. Consider the indicators together, check whether they describe comparable cohorts and periods, and seek supporting evidence before drawing a conclusion.
One important comparison is between segment-level rates and a portfolio-wide rate. A portfolio loss ratio is incurred losses divided by earned premium for the same defined portfolio and period. When segments have different earned premium, the overall result is premium-weighted, not the simple average of segment percentages. This follows directly from the definition:
The identity holds because each segment’s incurred losses equal its loss ratio multiplied by its earned premium. Summing those segment losses and dividing by total earned premium gives the overall ratio. A simple average gives each segment equal weight regardless of premium, so it generally answers a different question.
Worked Example: Calculate and Interpret Three Monthly Signals
In a hypothetical April monitoring report, 27 of 450 bound policies had a referral override. For a fully matured 90-day observation cohort, 36 claims were reported on 1,200 earned exposure-years. The portfolio also issued 600 eligible quotes, of which 144 bound. For this example, override thresholds are green at 4% or below, amber above 4% through 7%, and red above 7%. Early claim frequency is green at 27 claims or fewer per 1,000 earned exposure-years, amber above 27 through 33, and red above 33. Quote-to-bind is green at 25% or above, amber from 20% up to but not including 25%, and red below 20%. Classify all three and state what their joint pattern does and does not establish.
Method. Calculate each indicator using its own stated numerator and denominator, apply its own threshold bands, then interpret the pattern as a set of signals rather than as a causal finding.
Referral override rate:
Since 6% is above 4% and no greater than 7%, the override rate is amber.
Early reported claim frequency: 36 claims divided by 1,200 earned exposure-years gives \(36/1{,}200=0.03\) claims per exposure-year. Multiplying by 1,000 gives \(0.03 \times 1{,}000=30\) claims per 1,000 earned exposure-years. Since 30 is above 27 and no greater than 33, the frequency is amber.
Quote-to-bind rate:
Since 24% is at least 20% but below 25%, the quote-to-bind rate is amber. The calculation checks: 24% of 600 is 144.
All three indicators are amber. The owners should follow the specified amber review process and check data completeness and cohort comparability. The combination supports a more focused review of underwriting and distribution conditions than any one indicator alone. It does not establish that overrides caused the reported claims or that a particular pricing action is warranted; further analysis is required.
Common mistake avoided. Treating three amber readings as proof of one underlying cause overstates what monitoring data show. Joint movement is a reason to investigate, not by itself evidence of causation.
Worked Example: Reconcile Segment Results to the Portfolio KRI
A hypothetical portfolio has two segments for the quarter. Segment A has earned premium of $900,000 and incurred losses of $540,000. Segment B has earned premium of $700,000 and incurred losses of $560,000. The illustrative portfolio loss-ratio threshold is amber above 65% through 75% and red above 75%. Calculate both segment ratios and the portfolio ratio, classify the portfolio result, and explain why the unweighted average is misleading.
Method. Divide each segment’s incurred losses by its earned premium. Then add losses and premiums separately before calculating the portfolio ratio.
Segment A’s ratio is \(540{,}000/900{,}000=0.60=60\%\). Segment B’s ratio is \(560{,}000/700{,}000=0.80=80\%\). Total incurred losses are \(540{,}000+560{,}000=\$1{,}100{,}000\); total earned premium is \(900{,}000+700{,}000=\$1{,}600{,}000\). Therefore:
The portfolio ratio is above 65% and no greater than 75%, so it is amber. The arithmetic average of the segment percentages is \((60\%+80\%)/2=70\%\), which is not the portfolio ratio because the two segments have different earned premiums. The weighted calculation checks: \(60\% \times \$900{,}000=\$540{,}000\) and \(80\% \times \$700{,}000=\$560{,}000\), matching the given losses.
Common mistake avoided. Averaging segment percentages equally ignores the amount of premium supporting each segment. Use total losses divided by total premium for the portfolio result, and retain segment results to help investigate what is driving it.
Common Design and Interpretation Traps
- Leaving the denominator implicit. “Override rate” is incomplete unless the population in the denominator is stated.
- Mixing populations or periods. A claims numerator from a mature cohort cannot be compared directly with exposure from a different or immature cohort without explaining the difference.
- Using a threshold without an owner. A color on a dashboard is not an escalation process unless someone is accountable for reviewing it.
- Changing a definition after a signal. Validate a suspected data issue and document any approved correction; do not redefine the KRI simply to remove a breach or warning.
- Inferring cause from correlation or co-movement. Multiple adverse signals justify investigation but do not identify the causal mechanism on their own.
- Reporting only an aggregate. A portfolio rate can obscure different segment results. Use segment views to investigate, while ensuring the aggregate is calculated with the correct weights.
Check Your Understanding
For each problem, use the stated definitions and thresholds; the scenarios are hypothetical.
- A portfolio binds 280 policies in a month, of which 14 have a recorded referral override. The threshold is green at 3% or below, amber above 3% through 6%, and red above 6%. Calculate and classify the override rate. (Answer: \(14/280=5\%\), amber.)
- A mature cohort has 30 reported claims on 1,200 earned exposure-years. The plan is 26 claims per 1,000 earned exposure-years; amber is above 26 through 32, and red is above 32. Calculate and classify the frequency. (Answer: \(30/1{,}200 \times 1{,}000=25\) claims per 1,000 exposure-years, green, below plan.)
- An underwriting team issues 480 eligible quotes and binds 96. Its rule is green at 24% or above, amber from 18% up to but not including 24%, and red below 18%. Calculate and classify quote-to-bind. (Answer: \(96/480=20\%\), amber.)
- Segment A has $600,000 earned premium and $360,000 incurred losses; Segment B has $900,000 earned premium and $720,000 incurred losses. Calculate both segment loss ratios and the overall ratio. (Answer: A is 60%; B is 80%; overall is \((\$360{,}000+\$720{,}000)/(\$600{,}000+\$900{,}000)=72\%\), not the unweighted 70% average.)
- A 90-day claim KRI is intended to include only policies whose full observation window has elapsed. The report contains 50 claims and 2,000 exposure-years in total, but 20 claims and 800 exposure-years belong to policies that have not completed 90 days. What are the matured-cohort frequency and the key interpretation point? (Answer: \(30/1{,}200 \times 1{,}000=25\) claims per 1,000 exposure-years; exclude both the immature claims and their exposure, and interpret the result using the defined mature cohort.)