Articles

How to Quantify Cyber Risk in Dollars with FAIR: a Worked Example

Scenario, threat event frequency x vulnerability, PERT and lognormal inputs, the analytic expected loss and why you still simulate, worked on one ransomware scenario and then a portfolio.

A heat map tells a board that ransomware is "high" and payment fraud is "medium". It cannot say how much either is likely to cost next year, how bad a bad year could be, or whether a $1,100,000 control is worth buying. Quantification answers those questions in dollars. The method most risk teams use is the FAIR model, described in the Open FAIR standards published by The Open Group: risk is the probable frequency and probable magnitude of future loss. This guide works the method end to end on one scenario at Halden Marsh Industries, Inc., a fictional US-listed pump manufacturer with $1,850,000,000 of budgeted revenue, then rolls five scenarios into a portfolio. The ransomware scenario comes out at an expected annual loss of $1,404,218 on paper and $1,639,809 in simulation, and the gap between those two figures is the reason to simulate at all.

Worked in full in The Cyber Risk Quantification Handbook by Julian R. Sterling, with every figure reproduced in live Excel workbooks.Get the DRM-free PDF, $19.99 →

Step 1: write the scenario before you estimate anything

A number without a scenario cannot be challenged, so the first deliverable is a sentence. A usable scenario names the asset, the threat community, the effect (loss of confidentiality, integrity or availability) and the kind of loss. Halden Marsh's first scenario, S1, reads: ransomware encrypts the ERP and manufacturing execution systems and stops plants. Primary losses are lost production margin, response, recovery and overtime; secondary losses, which arise only when outside parties react, are customer penalties, litigation and reputation.

Keep one scenario per decision. A scenario called "cyber attack" is too broad to estimate and too vague to fund a control against. The five Halden Marsh scenarios each map to a decision the 2027 plan has to make:

IdScenarioTEF per yearVulnerability
S1Ransomware outage of plant and ERP systems1 / 3 / 80.010 / 0.020 / 0.050
S2Business email compromise and payment fraud8 / 20 / 450.010 / 0.030 / 0.070
S3Breach of employee and dealer personal data1 / 2 / 50.020 / 0.040 / 0.100
S4Outage of a critical cloud provider0.2 / 0.6 / 20.150 / 0.300 / 0.600
S5Insider theft of product design files0.5 / 1 / 30.050 / 0.150 / 0.350

Each input is a minimum, a most likely value and a maximum, not a point. Experts can defend a range; nobody can defend "2.7 attacks a year".

Step 2: frequency, or TEF x vulnerability = LEF

FAIR splits frequency in two. Threat event frequency (TEF) is how often a threat actor acts against the asset in a year. Vulnerability is the probability that a threat event becomes a loss event, given the controls in place. Their product is the loss event frequency:

LEF = TEF x Vulnerability

The split matters because controls act on different factors. Better phishing filters lower TEF; patching and endpoint detection lower vulnerability; immutable backups do neither and act on magnitude instead. A single "likelihood" score hides which lever you are pulling.

Each three-point estimate is turned into a PERT distribution, a smooth curve that peaks at the most likely value and stays inside the minimum and maximum. Its parameters follow directly from the three points:

alpha = 1 + 4 x (most likely - min) / (max - min)
beta = 1 + 4 x (max - most likely) / (max - min)
mean = (min + 4 x most likely + max) / 6

For S1, TEF (1, 3, 8) gives alpha 2.1429, beta 3.8571 and a mean of 3.5000 attempts a year. Vulnerability (0.01, 0.02, 0.05) gives alpha 2.0000, beta 4.0000 and a mean of 0.0233. The mean LEF is 0.0817: one successful ransomware event about every 12.2 years. That is a far more useful statement than "likelihood 3 out of 5".

Step 3: magnitude, from a calibrated 90 percent range to a lognormal

Ask the people who would pay for the loss, not only the security team. The finance lead, a plant manager and the head of legal give a 90 percent range: a low value they would be surprised to see undercut and a high value they would be surprised to see exceeded, each about 5 percent of the time. Before trusting those ranges, test the estimators. In the case, a plant manager captured only 6 of 10 true answers in a first calibration exercise, then 9 of 10 after two rounds of feedback. Ten questions prove nothing statistically, but they expose overconfidence quickly.

Cyber losses are skewed: most events are moderate and a few are very large. The lognormal distribution has that shape and can be fitted to a 90 percent range with two lines, where 1.6449 is the standard normal value that leaves 5 percent in each tail:

mu = (ln lo + ln hi) / 2
sigma = (ln hi - ln lo) / (2 x 1.6449)

S1's primary loss range of $4,000,000 to $38,000,000 gives mu 16.3275 and sigma 0.6843. The median event costs $12,328,828; the mean is $15,581,802, 26 percent above the median because the right tail pulls it up. Which figure you quote changes the answer, and the expected loss always uses the mean.

The secondary loss is a second lognormal, fitted to $1,000,000 to $12,000,000, with a mean of $4,607,734, and it only occurs on 35 percent of S1 events. FAIR's six forms of loss (productivity, response, replacement, fines and judgments, competitive advantage, reputation) serve as a checklist so the ranges do not quietly omit a category.

Step 4: the analytic expected loss

With frequency and magnitude in hand, the annualized loss expectancy (ALE) follows by hand:

Repeat for the other four scenarios and the ranking is already more informative than the heat map it replaces:

IdMean LEFMean loss per eventAnalytic ALE
S10.0817$17,194,509$1,404,218
S20.7389$900,016$665,012
S30.1089$6,201,958$675,324
S40.2492$3,302,375$822,842
S50.2083$5,959,405$1,241,543

The total is $4,808,939. Note S5, insider theft of design files: frequent enough and with a heavy secondary loss, it carries 25.8 percent of the total, while the old heat map scored it 8, the same as payment fraud. A score of 8 covered both a scenario worth $617,294 a year in simulation (S2) and one worth $1,194,297 (S5). Multiplying ordinal scores is not arithmetic.

Step 5: why the average is not enough, and why you simulate

An ALE of $1,404,218 suggests a steady annual cost. S1 never behaves like that. In most years nothing happens; in a few years a single event costs eight figures. The average describes no actual year, and a board that budgets to it will be wrong almost every time. To see the shape, run a Monte Carlo simulation: each trial is one simulated year, with TEF, vulnerability, the number of loss events (a Poisson count with the trial's LEF) and each event's magnitude drawn from their distributions.

Trial 2 is the first S1 year with a loss. TEF draws 5.2482 attempts and vulnerability 0.0123, so LEF is 0.0644. The probability of zero events is 0.9376; the count uniform of 0.961500 exceeds it, so one loss event occurs. The primary draw gives $19,360,813 and the secondary uniform of 0.723554 is not below 0.35, so no secondary loss. S1 loss for the year: $19,360,813.

Over 1,000 trials, S1 has no loss in 90.7 percent of years, and the simulated ALE is $1,639,809, 16.8 percent above the analytic figure. That is sampling error, not a contradiction: with a loss in only 93 of 1,000 years, a handful of large draws moves the mean. The simulation's real output is the distribution, which is the subject of the guide on how to build and read a loss exceedance curve. The free Cyber Risk Monte Carlo Model runs exactly these 1,000 trials for S1, so you can replace the ranges with your own and watch the results move.

Step 6: roll the scenarios into a portfolio

Simulated together, the five scenarios give a portfolio ALE of $5,016,300, which is 0.27 percent of revenue and 2.9 percent of pre-tax income. The median year costs $1,505,535, and one year in four has no loss at all. These are the first numbers a CFO can put next to a $14,500,000 security budget and an insurance premium.

From there the analysis turns into decisions. Controls are priced by rerunning the same trials with new inputs, and insurance is tested against the tail, as shown in the guide on how to choose a cyber insurance retention and limit. The full method, from scenario workshops to the board report, is laid out in The Cyber Risk Quantification Handbook.

Questions practitioners ask

What is FAIR in cyber risk quantification?

FAIR is a model, set out in the Open FAIR standards of The Open Group, that defines risk as the probable frequency and probable magnitude of future loss. Frequency is threat event frequency x vulnerability; magnitude is primary plus secondary loss. The output is a dollar distribution, not a color.

How do you turn an expert's range into a distribution?

Ask for a calibrated 90 percent range and fit a lognormal: mu is the average of the logs of the two bounds, sigma is their difference divided by 2 x 1.6449. For S1, $4,000,000 to $38,000,000 gives a median of $12,328,828 and a mean of $15,581,802.

Is the analytic ALE wrong if the simulation gives a different number?

No. The analytic ALE is the exact mean of the model; the simulation estimates it from a sample. For S1, $1,639,809 simulated against $1,404,218 analytic reflects only 93 loss years in 1,000 trials. The simulation is needed for percentiles, not for the mean.

Do I need special software to quantify cyber risk?

No. A spreadsheet with stored random numbers, PERT and lognormal inverses and a Poisson count reproduces the method. The free Cyber Risk Monte Carlo Model does it for one scenario over 1,000 trials, prefilled with S1.

Read the full handbook

This guide is one arc of The Cyber Risk Quantification Handbook. The book follows one company from its heat map to five quantified scenarios: calibrated estimates, the FAIR model, Monte Carlo simulation, the portfolio loss exceedance curve against board tolerance, the return on controls, the cyber insurance renewal and an incident materiality decision.

Paperback PDF edition, $19.99 Book page and companion workbooks

Free download: Cyber Risk Monte Carlo Model (Excel).

Also on this site

Illustrative figures computed by the book model; the company, its scenarios and every estimate are fictional. Rules described in principle as read in November 2026. Education only, not insurance, legal or security advice.