If you work in insurance, you have seen this chart many times. Insurers and catastrophe modelers have used the loss exceedance curve for decades to price hurricanes, earthquakes and floods.[1] If you own cyber risk, it may be new. Cyber risk quantification (CRQ) brings it into security programs, where it is still new to many teams and often replaces a familiar heat map.
The curve answers one plain question: what is the chance that we lose more than a given financial amount in a year? Pick any amount on the chart, and the curve tells you how likely it is to be exceeded.
That single question is why the curve is so useful. It lets a board set a risk appetite in money terms. It lets a security team show what a control is worth. And it lets a risk owner talk to an insurer in the insurer's own language.
This article walks through the curve step by step. By the end you should be able to read any loss exceedance curve, compare two options with it, and use it in an insurance conversation.
A loss exceedance curve has only two axes, and each answers a simple question.
The curve always starts high on the left and falls to the right. That is because small losses are easy to exceed, and very large ones are hard to exceed. A curve can never rise as you move right.
Reading across. Start with an amount you care about. Find it on the horizontal axis, go up to the curve, then read left to the vertical axis. In the chart, the dashed guide starts at a loss of 40. It meets the curve at about 5%. So there is about a 5% chance, each year, that losses will exceed 40.
Reading up. You can also start from a probability. Pick a level the board cares about, such as 5% per year. Go across to the curve, then down to the horizontal axis. That amount, 40 in this example, is the loss the organization should expect to exceed only about one year in 20.
Both readings use the same curve. Reading across answers "how likely is a loss this big?" Reading up answers "how big a loss should we plan for?"
Most reports call out a handful of points. Each one answers a different question, so it helps to know them by name.
Return periods. Insurers often describe a point on the curve as a "1-in-N-year" loss. The math is simple: a 1-in-20-year loss has about a 5% chance of being exceeded in any year (1 ÷ 20 = 5%). A 1-in-100-year loss has about a 1% chance.[1] It does not mean the loss happens exactly once every 100 years. It can happen twice in a decade, or not at all for a century.
| Point | Chance of exceeding per year | What it tells you | Example curve (relative scale) |
|---|---|---|---|
| Most likely annual loss | Varies | The most common result: a typical year | About 7 |
| Median (1-in-2-year) | 50% | Half of years are better than this, half are worse | About 10 |
| Expected annual loss (EAL) | Varies | The long-run yearly cost; useful for budgets and comparing options | About 14 |
| 1-in-10-year loss | 10% | A bad year you should plan for; also the top of the 10th–90th band | About 26 |
| 1-in-20-year loss | 5% | A severe year; a common reference for insurance limits | About 40 |
| 1-in-100-year loss | 1% | A very rare, severe year; a stress test for survival | About 95 |
Notice the order. The most likely loss sits below the median, and the median sits below the average. That is normal for cyber losses. A few very large years pull the average up, so the average is often higher than what most years actually cost.[2]
No single point tells the whole story. The average is good for budgets, but it hides the tail. The 1-in-100 point shows the tail, but it says nothing about a typical year. Good decisions use several points together.
The shape of the curve tells you what kind of risk you face, before you read a single number.
Each row shows the same risk two ways. On the left, a ridge shows how annual losses are spread: where most years land and how far the bad years reach. On the right is the matching exceedance curve. The grey bar in the bottom row marks the years with no loss at all, which is why that curve starts low.
Two curves can share the same average and still describe very different risks. One may be a steady cost of doing business. The other may be quiet for years and then threaten the company. The shape is what tells them apart.
Our article on how certain a CRQ result is explains how to tell frequent losses from rare, severe ones, and how much of a curve's spread comes from missing data.
The most practical use of a loss exceedance curve is comparison. Run the analysis twice: once for things as they are, and once with a proposed control in place. Then draw both curves on the same chart.
The gap between the two curves is the risk reduction. You can read it in two ways:
Controls work in two ways, and the curve shows both. A control that prevents incidents lowers frequency, which pulls the curve down. A control that limits damage, such as faster recovery, lowers magnitude (severity), which pulls the curve to the left. Many controls do some of each.
Look at where the gap is widest. A control that mostly narrows the gap near the left helps in typical years. A control that narrows the gap in the tail protects against the rare, severe year. Both have value, but they answer different board questions.
The reduction in average annual loss, compared with the control's cost, gives a simple return on investment. The change in the tail shows how much safer the organization becomes against the worst case.
This is where the loss exceedance curve feels most at home. Insurers use it to decide what to cover, where coverage starts and stops, and what to charge.[1] A risk owner who can read it can have a far better conversation with a broker or underwriter.
Annual versus per-event curves. There are two common versions of the curve, and it matters which one you are looking at.
The two differ when several incidents can happen in one year. Policy terms often follow one view or the other. A per-incident retention, for example, applies to each event. An annual aggregate limit caps the total paid in a year. Ask which view a curve shows before comparing it with policy terms.
Placing retention and limit on the curve. Mark the retention (what you pay before cover starts) and the top of the limit on the loss axis. The curve then tells you three things at a glance:
This turns insurance buying into a clear trade-off. A higher retention lowers the premium but leaves more of the typical year with you. A higher limit costs more but protects against the tail. The curve shows exactly which years each choice affects.
A few misreadings come up again and again. Each one is easy to avoid once you know it.
You do not need to build a loss exceedance curve to use one well. You need to ask the right questions of it. Before you act on any curve, check these five things:
With those answers, the curve becomes a shared language. The board can set risk appetite as a point on the curve. Security can show what a control moves. Finance and insurance can see where to retain risk and where to transfer it.
The OT gap. A curve is only as complete as the losses it includes. Industrial (OT) incidents bring costs that IT-focused models often leave out: damaged equipment, lost production, safety incidents, and environmental cleanup. Leave those out, and the tail of the curve looks much smaller than it really is.
The same is true of the data behind each loss. A curve built mostly on public data and industry estimates has to stay wide to stay honest, and its tail is the least certain part. Each step of evidence from your own sites replaces an assumption, so the curve becomes narrower and more accurate, and the tail becomes something you can plan around. Our article on how certain a CRQ result is shows how that works.
DeNexus™ focuses on filling that gap for industrial operators and the insurers who cover them. As the pacesetter in OT cyber risk quantification, DeNexus builds OT-specific losses into DeRISK™ CRQ, so the tail of the curve reflects how industrial sites actually lose money. To learn more, read What Is OT Cyber Risk Quantification?
Put your OT losses on the curve
DeRISK CRQ builds OT-specific losses into the loss exceedance curve, so the tail you plan and insure against reflects how industrial sites actually lose money. No OT network data needed to start.
See DeRISK CRQ in action