Introduction
ERM can contribute value even when no large loss is prevented. It may help management choose a better strategy, sustain important services during disruption, or pursue an opportunity with clearer limits and monitoring. Evaluating these contributions requires more than listing ERM activities: the evidence should connect risk information to a decision or outcome that matters.
A claim that ERM “saved” a loss that did not happen is especially difficult to substantiate. The organization cannot observe what the same event would have cost under a different decision or without a control. A stronger evaluation identifies the decision, states the assumptions behind the comparison, and distinguishes measured results from estimates. These practices are aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.
Compare Decisions Under the Same Assumptions
For a decision, a useful starting point is to compare the expected outcomes of the selected alternative and a plausible alternative. Define the relevant outcome states, assign probabilities, and estimate each alternative’s payoff in each state. The comparison is informative only if both alternatives use the same states, probability assumptions, time horizon, and payoff basis.
Here, \(s\) indexes possible states, \(P(s)\) is the probability assigned to state \(s\), and \(U(a,s)\) is the payoff from alternative \(a\) in that state. “Payoff” must be defined for the particular analysis. It might be contribution over a stated period, service capacity, or another decision-relevant measure; it should not be presented as a complete measure of organizational value unless it actually includes all relevant effects.
Worked Example: Evidence From a Revised Expansion Decision
A hypothetical insurer is considering entering a new region. Management compares a full launch with a staged launch. For a one-year planning comparison, the favorable market state has probability 0.60 and the adverse state has probability 0.40. Payoffs are net contribution in millions of dollars over that year. A preliminary assessment had assigned probabilities of 0.80 and 0.20, respectively. Risk analysis then leads management to use the 0.60 and 0.40 assessment. Estimate the value of the revised decision compared with the decision management would have made using its preliminary assessment.
Method. Calculate the expected payoff of each alternative under the preliminary assumptions to identify the initial choice. Then calculate both alternatives under the revised assumptions and compare the revised choice with the initial choice using the same revised probabilities.
Under the preliminary probabilities, the full launch has expected payoff \(0.80(3.00)+0.20(-4.00)=2.40-0.80=\$1.60\) million.
The staged launch has preliminary expected payoff \(0.80(1.80)+0.20(0.60)=1.44+0.12=\$1.56\) million.
On those preliminary assumptions, management would choose the full launch. Under the revised probabilities, its expected payoff is \(0.60(3.00)+0.40(-4.00)=1.80-1.60=\$0.20\) million.
The staged launch’s expected payoff under the revised probabilities is \(0.60(1.80)+0.40(0.60)=1.08+0.24=\$1.32\) million.
The revised decision is the staged launch. Relative to continuing with the original full-launch choice, the modeled expected-payoff difference under the revised assumptions is \(\$1.32-\$0.20=\$1.12\) million.
Check. The staged alternative is higher in both revised-state calculations combined: its favorable-state payoff is $1.20 million lower than the full launch, but its adverse-state payoff is $4.60 million higher. Weighting those differences gives \(0.60(-1.20)+0.40(4.60)=-0.72+1.84=\$1.12\) million, matching the comparison above.
This is evidence of a potentially better-informed decision, not proof that ERM generated $1.12 million. The probability estimates and payoffs are hypothetical, and the outcome states are uncertain. The comparison also depends on the full launch being a credible counterfactual.
Resilience Is More Than a Loss That Did Not Occur
Resilience concerns the ability to continue or restore important operations when disruption occurs. It can be assessed using observable measures such as time to restore a service, the share of critical transactions completed during an interruption, or the size and age of a backlog. A test, exercise, or real disruption can provide evidence about these measures. A modeled scenario can help estimate their financial implications, but those estimates should be labeled as modeled.
Worked Example: Translating a Recovery Test Into Resilience Evidence
A hypothetical insurer tests its claims-processing recovery plan. In the test, the prior process would have taken 22 hours to restore service; after a revised recovery procedure, a repeat test restores service in 8 hours. For this illustrative calculation, assume interrupted processing would reduce contribution by $30,000 per hour, at a constant rate, with no backlog recovery effect or other consequential loss. Estimate the modeled reduction in interruption impact.
Method. Calculate the difference in restoration time, then multiply by the assumed hourly impact. The calculation treats the hourly impact as constant throughout the interruption.
The reduction in restoration time is \(22-8=14\) hours. The modeled interruption impact before the change is \(22\times \$30{,}000=\$660{,}000\). After the change it is \(8\times \$30{,}000=\$240{,}000\). The modeled difference is \(\$660{,}000-\$240{,}000=\$420{,}000\), also equal to \(14\times \$30{,}000\).
Check. The modeled reduction is positive because restoration time fell by 14 hours and the assumed hourly impact is positive. The calculation is not evidence that $420,000 was actually saved: it estimates the impact under specified assumptions. The repeat test directly supports the narrower claim that recovery time improved in that test. Evidence from a real disruption would help assess whether the improvement carried over to operating conditions.
The common mistake is to report the modeled dollar difference as a realized saving. The better-supported finding is that the tested recovery time improved from 22 hours to 8 hours; the dollar estimate is a conditional translation of that result.
Strategic Opportunity Can Also Be Evidence of ERM Value
Risk information can help an organization pursue an opportunity rather than simply avoid an exposure. For example, management may identify a customer need, set an initial exposure limit, and establish indicators that will inform whether to expand. The value assessment should compare the opportunity with a plausible alternative, state what outcomes are included, and avoid treating a favorable projection as a guaranteed result.
Worked Example: Evaluating a Controlled Product Pilot
A hypothetical insurer considers a 100-policy pilot of a new small-business cyber endorsement. Management estimates that the pilot would produce net contribution of $140,000 in a favorable experience state and a net loss of $60,000 in an adverse state. The estimated probability of a favorable state is 0.70; the adverse-state probability is 0.30. The alternative is not to launch, with a contribution of $0 for this comparison. Estimate the pilot’s expected contribution and interpret what it supports.
Method. Weight each state’s net contribution by its stated probability, add the results, and compare with the no-launch alternative. Assume the two states are mutually exclusive and exhaustive for this simplified analysis, and that the amounts use the same time period and include the items management intends to count.
The pilot’s expected contribution is \(0.70(\$140{,}000)+0.30(-\$60{,}000)=\$98{,}000-\$18{,}000=\$80{,}000\).
The no-launch alternative has expected contribution of $0 under the comparison. Thus, the pilot’s modeled expected-contribution difference is \(\$80{,}000-\$0=\$80{,}000\).
Check. The favorable-state contribution exceeds the adverse-state loss, and the favorable state has the higher assigned probability. The calculation is positive: $98,000 of probability-weighted favorable contribution less $18,000 of probability-weighted adverse loss equals $80,000.
The estimate supports considering a limited pilot under the stated assumptions; it does not establish that the pilot will earn $80,000 or that ERM alone created that opportunity. Useful follow-up evidence would include actual claims experience, contribution, and whether results remain within the pilot’s exposure limit. Comparing those observations with the original assumptions helps management decide whether the opportunity merits further action.
Judge the Evidence, Not Just the Story
A credible evaluation links activity to a result through an explicit chain: risk information changed a decision or response; the change affected an outcome or a tested capability; and the evidence supports the connection. The claim should be proportionate to what was observed. For example, a documented decision change may support the statement that risk analysis informed the choice. It does not, on its own, establish the financial value of that choice.
| Evidence type | What it can support | What it does not establish by itself |
|---|---|---|
| Decision record showing alternatives, assumptions, and rationale | Risk information was considered and affected the decision process | The selected alternative caused a favorable result |
| Recovery test or disruption records | A measured change in restoration time or service continuity | A specific dollar saving unless the financial impact is also supported |
| Scenario-based estimate | A modeled difference under stated probabilities and payoffs | A realized saving or a certain future outcome |
| Pilot results against predefined indicators | Observed performance for the pilot scope and period | That the same performance will hold at a larger scale |
To make a value claim more credible, preserve the decision record, define a comparison that could reasonably have occurred, identify important assumptions, and track results over an appropriate period. Note material changes in the operating environment that could also explain an outcome. Where uncertainty is substantial, present a range or scenario comparison rather than a single precise figure.
A frequent exam trap is to call a loss “avoided” simply because it did not occur. Suppose a hypothetical event has a 5% annual probability and a $10 million severity. Its modeled expected loss before a control is \(0.05\times \$10{,}000{,}000=\$500{,}000\) per year. If the control reduces the estimated probability to 2%, with severity unchanged, the post-control modeled expected loss is \(0.02\times \$10{,}000{,}000=\$200{,}000\). The modeled expected-loss reduction is $300,000 per year under those assumptions.
The wrong calculation is to report “$10 million saved” because no event occurred. The correction is to report the $300,000 modeled annual expected-loss reduction, with its assumptions, and to say that no event occurred during the observation period if that is what was observed. A zero-loss period does not establish that the control prevented a $10 million loss; the event might not have occurred under either approach.
Check Your Understanding
For each problem, distinguish a calculated estimate from a claim that the estimate proves.
- A hypothetical control is estimated to reduce an annual event probability from 4% to 1.5%, with event severity unchanged at $2 million. Calculate the modeled annual expected-loss reduction. What kind of value evidence is this? (Answer: \(0.04(\$2{,}000{,}000)-0.015(\$2{,}000{,}000)=\$80{,}000-\$30{,}000=\$50{,}000\) per year, a modeled reduction rather than a realized saving.)
- Two strategies have payoffs of $800,000 favorable and \(-\$500,000\) adverse for Strategy A, and $400,000 favorable and $100,000 adverse for Strategy B. The favorable and adverse probabilities are each 0.50. Which has higher expected payoff, and by how much? (Answer: A has \(0.50(\$800{,}000)+0.50(-\$500{,}000)=\$150{,}000\); B has \(0.50(\$400{,}000)+0.50(\$100{,}000)=\$250{,}000\). B is higher by $100,000.)
- A recovery test shows restoration time falling from 15 hours to 9 hours. Assume interruption impact is $25,000 per hour, constant, with no other effects. Calculate the modeled reduction in interruption impact. (Answer: \(15-9=6\) hours; \(6\times \$25{,}000=\$150{,}000\). This is a conditional estimate based on the test and hourly-impact assumption.)
- A 100-policy hypothetical pilot has a net contribution of $200,000 in the favorable state and a net loss of $100,000 in the adverse state. Their probabilities are 0.65 and 0.35, respectively; not launching has contribution $0. Calculate the expected difference and name one observation that should be tracked. (Answer: \(0.65(\$200{,}000)+0.35(-\$100{,}000)=\$130{,}000-\$35{,}000=\$95{,}000\) above no launch under the assumptions; track actual pilot claims experience or contribution against the estimates.)
- A hypothetical event has a 3% annual probability and $4 million severity before a control, and a 1% probability after it, with severity unchanged. No event occurs during the year. Calculate the modeled expected-loss reduction and state whether $4 million was demonstrably saved. (Answer: Before, \(0.03(\$4{,}000{,}000)=\$120{,}000\); after, \(0.01(\$4{,}000{,}000)=\$40{,}000\); modeled reduction, $80,000 per year. No: the absence of an event does not demonstrate that a $4 million loss was prevented.)