Insights / Risk and the budget
Risk and the budget

Risk that counts

Keep the heat map to sort the list. For the handful of risks that will decide this year's spending, add a rough dollar range from your own numbers, price the fixes and doing nothing, and fund what removes the most risk per dollar.

Ausper Technology · Nov 18, 2026 · Insight · Risk and the budget · 10 min read · Updated Nov 19, 2026

The short version
  • Keep the heat map to sort risks. For the few that decide the budget, add a rough dollar range: how often it could happen times what it would cost.
  • Price every fix three ways, with what you own, in-house or bought, and price doing nothing. Fund the fixes that remove the most yearly risk per dollar.
  • The stakes keep rising. IBM puts the average US breach at a record $11.5 million, and Allianz ranks cyber the top business risk for a fifth year.

A red square on a heat map sorts a risk. It does not tell a CEO what to spend. You do not need to price every cell, and no price will be exact. But for the handful of risks that will decide this year’s spending, a rough dollar range changes the conversation: how often it could happen, times what it would cost, built from your own numbers. Then price the fixes, including doing nothing, and fund the ones that remove the most risk per dollar.

The amounts at stake keep rising. IBM’s 2026 Cost of a Data Breach report puts the average breach at a record $4.99 million worldwide and a record $11.5 million in the US. AI-driven attacks added about $1 million per breach, and organizations took 247 days on average to identify and contain one (eSecurity Planet has a summary). In Allianz’s Risk Barometer, published January 14, the companies surveyed ranked cyber the top business risk for the fifth year in a row. AI jumped from tenth to second.

What a color can’t tell you

Our read: a heat map is a good way to sort a list and a poor way to spend money, and the research says the same. A 2008 study in Risk Analysis found that risk matrices “can assign identical ratings to quantitatively very different risks,” can correctly compare only a small fraction of randomly chosen pairs of hazards (less than 10 percent in its example), and that resources “cannot be based on the categories provided by risk matrices” alone (Cox, via PubMed). NIST’s current guidance says “quantitative methods are preferred” and encourages organizations “to quantify risk in financial terms wherever feasible” (NIST IR 8286 Rev. 1). In the budget meeting, a red cell argues with a quote from a vendor, and the loudest voice in the room settles it.

Here is what that looks like. Risk A is the live cloud key from the worked example below: 0.1 to 0.4 events a year, $2.5 million to $5 million each. Risk B, also illustrative, is a stale vendor integration that can take a customer portal down for a day: 0.1 to 0.3 events a year, $300,000 to $600,000 each. A common 5-by-5 matrix rates both “possible” and “major.” In dollars, A is about $937,500 a year at the middle of its ranges and B about $90,000. Same cell, about ten times apart.

Fig. 1
Same cell, about ten times apart

Two illustrative risks a common 5x5 matrix rates the same: possible, major. Their expected loss a year differs tenfold.

CriticalSevereMajorModerateMinor
AB
RareUnlikelyPossibleLikelyAlmost certain
Likelihood →
Impact runs up, likelihood runs right
  • A$937,500 a yearRisk A: a live cloud key in a vendor's code (the worked example below)
  • B$90,000 a yearRisk B: a stale vendor integration that can take a customer portal down for a day
Illustrative numbers, not a client. Expected loss a year at the middle of each range; our arithmetic, reproducible from the inputs in the text.
Data
MarkRiskLikelihoodImpactDollars
ARisk A: a live cloud key in a vendor's code (the worked example below)PossibleMajor$937,500 a year
BRisk B: a stale vendor integration that can take a customer portal down for a dayPossibleMajor$90,000 a year

Put a rough price on the risks that decide spending

The math is old and simple. Yearly loss is how often an event could happen in a year, times what one event would cost. It is the core of FAIR, the open standard for risk analysis, which defines risk as “the probable frequency and probable magnitude of future loss” (The Open Group; FAIR Institute). Carry both as ranges, low and high, because nobody knows either number exactly and a range says so; NIST describes the same idea as a three-point estimate (NIST IR 8286A Rev. 1). It is a way to compare options, not a forecast of the bill.

Build the cost from numbers your company already has:

  • Revenue per hour of downtime for the system at risk, times the hours it would take to recover.
  • Records held, times what each one costs you to notify, monitor and answer for.
  • Contract penalties: the SLA credits, liquidated damages and refunds you would owe.
  • Insurance limits: what the policy pays and where it stops. Check the high end of the range against it.

In our November 18 piece on outages, we gave a formula for what a day down costs. That is the cost half. This adds the other half, how often, and turns the answer into a yearly number you can set beside a budget line.

Price every fix three ways, and price doing nothing

For each risk, price the fix three ways: with what you already own, done in-house, and bought from any vendor. Give each one its first-year cost and the share of the yearly loss it removes. Doing nothing gets a price too: the yearly loss you keep.

Then rank. Divide the yearly risk each fix removes by what it costs, and fund from the top. Sometimes the cheapest fix wins. Sometimes the expensive one does, because it covers many risks at once. Either way the ranking shows its work, and anyone in the room can check it.

A worked example

Illustrative numbers, not a client. A review finds a live cloud key, with write access to a production storage bucket, in code a vendor ships to you. What would one event cost?

  • Downtime: $50,000 of revenue an hour, 24 to 48 hours to contain, rotate and restore. $1,200,000 to $2,400,000.
  • Records: 40,000 customer records the key can reach, at $25 to $50 each to notify, monitor and answer for. $1,000,000 to $2,000,000.
  • Penalties: $300,000 to $600,000 in SLA credits and liquidated damages.
  • One event: $2,500,000 to $5,000,000. The cyber policy’s $2 million limit covers less than half of the high end.

How often? While the key stays live, we would put it at 0.1 to 0.4 events a year, between once in ten years and once in two and a half. That makes the yearly loss $250,000 to $2,000,000, with a middle estimate of $937,500: the midpoint frequency times the midpoint cost.

Fig. 2
The whole calculation fits on one line

How often, times what one event costs, gives the expected loss a year. Ranges in, a range out.

0.1–0.4Events a year while the key stays live. Our estimate.
$2.5–5MWhat one event costs: downtime, records and penalties.
$0.25–2MExpected loss a year. Middle: $937,500.
Illustrative numbers, not a client. Our arithmetic, reproducible from the inputs listed in the text.
Data
StepValueWhat it is
10.1–0.4Events a year while the key stays live. Our estimate.
2$2.5–5MWhat one event costs: downtime, records and penalties.
3$0.25–2MExpected loss a year. Middle: $937,500.

What we calculate for a full risk

Multiplying the ranges gives a span. To see how a year can actually go, we treat each range as a 90 percent estimate, turn the cost of one event into a lognormal distribution and the frequency into a count of events, and simulate 200,000 years. That is the method Hubbard and Seiersen publish for cyber risk (Hubbard Decision Research), and NIST suggests the same Monte Carlo approach to take some guesswork out of the estimates (NIST IR 8286A Rev. 1). The result is a loss exceedance curve: the chance that one year’s loss passes any given amount. NIST uses that chart as its own example of a quantitative risk measure (NIST IR 8286 Rev. 1).

Fig. 3
One year in five, this key costs more than the $2 million insurance limit

Chance that one year's loss passes each amount, from 200,000 simulated years. Across: loss in a year, $ millions.

CHANCE PER YEAR, %05101520250246821.8%any loss21.8%past the $2M limit17.7%past $3M8.2%past $4M3.8%past $5M0.9%past $8MCHANCE PER YEAR, %05101520250246821.8%any loss21.8%past the $2M limit17.7%past $3M8.2%past $4M3.8%past $5M0.9%past $8M
Illustrative numbers, not a client. 200,000 simulated years (fixed seed): frequency drawn from 0.1 to 0.4 a year, cost per event lognormal on the $2.5 million to $5 million range. A loss exceedance curve, the standard picture in quantitative risk analysis.
Data
PointValueNote
any loss21.8%
past the $2M limit21.8%
past $3M17.7%
past $4M8.2%
past $5M3.8%
past $8M0.9%

Read it like this. In about four years out of five, the key costs nothing. In 21.8% of years, about one in five, the loss passes the $2 million insurance limit. In 3.8% it passes $5 million. The 1-in-20-year loss is about $4.6 million, and the average comes to about $0.9 million a year, close to the $937,500 shortcut. Those are numbers a CFO can plan around: the odds of blowing through the policy, and the bad year to budget for.

Now the four options, ranked by yearly risk removed per dollar:

  • Use what you already own: $10,000. Revoke and rotate the key today, move the integration to short-lived credentials from a role you already have, and turn on the secret scanning you already license. A few days of work across your team and the vendor’s. We estimate it removes 90 percent of the risk, $843,750 a year at the middle. About $84 of yearly risk removed per dollar.
  • Buy from a vendor: $120,000 a year. A secrets-management platform with managed scanning of every vendor delivery. An estimated 95 percent, $890,625 a year. $7.40 per dollar.
  • Do it in-house: $180,000 in the first year. Stand up a vault, scanning in every pipeline and rotation automation. Also 95 percent. $4.90 per dollar.
  • Do nothing: $0. You keep the whole $937,500 a year at the middle, and up to $2,000,000.
Fig. 4
Using what you already own removes about $84 of yearly risk per dollar

Yearly risk removed per dollar of first-year cost, at the middle estimate. Illustrative numbers, not a client.

Use what you own$10,000
$84
Buy from a vendor$120,000 a year
$7.40
Do it in-house$180,000 first year
$4.90
Do nothingkeeps $937,500 a year
$0
Illustrative numbers, not a client. Our arithmetic from the inputs listed in the text.
Data
FixRemoved per $1Note
Use what you own$84$10,000
Buy from a vendor$7.40$120,000 a year
Do it in-house$4.90$180,000 first year
Do nothing$0keeps $937,500 a year

The cheap fix wins by more than ten to one. That does not make the platform a bad buy. Count every key it would cover across your whole register and it may rank first next quarter. What matters is that the ranking decides, and that it is written down.

Our forecast

Our forecast: in January 2027, Allianz’s Risk Barometer will rank cyber the top business risk for a sixth year, and AI will stay in the top three. If either misses, we will say so here.

Our question: what is your biggest cyber risk worth in dollars a year, and what is the cheapest fix that removes most of it?

What we’d do first

  • Take the ten reddest cells on your heat map and price each as a range: how often, times what it would cost, from your own revenue, record counts, contract terms and policy limits.
  • Price every fix three ways, plus doing nothing, and rank by yearly risk removed per dollar.
  • Start with what you own. Turn on and tune what is already in your licenses before you buy anything new.
  • Check the high end against your insurance. If one event could pass the policy limit, that gap belongs in front of the board.
  • Re-price every quarter, and whenever a finding, a contract or a policy changes.

Ausper is positioned to build this register with you and keep it current: cyber metrics and risk visualization, executive reporting, independent security assessment, and second opinions before major spend. We’re paid for the answer, not for what you buy. If the best fix is a setting you already own, that is what we will tell you.

Want a first finding to price? Our free 30-minute AI exposure review looks for the kind of risk in the example above: keys in code, shared packages and what your vendors ship you.

Common questions

Why not keep the heat map?
Keep it to sort the list. A color cannot be compared with a price, and two red cells can differ in cost by ten times. For the few risks that decide spending, add a rough dollar range and rank the fixes by risk removed per dollar.
Where do the numbers come from?
From your own books: revenue per hour of downtime for the system at risk, records held and what each costs you to handle, the penalties in your contracts and the limits in your insurance policy. Where a number is uncertain, carry a low and a high.
What does risk removed per dollar mean?
The yearly loss a fix takes away, divided by what the fix costs in its first year. A fix that removes $843,750 of yearly expected loss for $10,000 scores about 84. Fund from the highest score down.
How do you get from two ranges to a loss curve?
Treat each range as a 90 percent estimate. The cost of one event becomes a lognormal distribution between its two ends, and the frequency becomes a count of events each year. A simulation draws many years from both and sorts the results; the share of years above any amount is the loss exceedance curve. Our worked example uses 200,000 years and a fixed seed, so anyone can rerun it.
Are the worked example’s numbers real?
No. They are illustrative, chosen to show the method, and they come from no client. Every input is listed in the text, so you can redo the arithmetic with your own.
What do the real breach numbers say?
IBM’s 2026 report puts the average breach at a record $4.99 million worldwide and $11.5 million in the US, with about $1 million added when the attack was AI-driven and 247 days on average to identify and contain a breach.

Sources

Last verified Nov 19, 2026

  1. IBM, Cost of a Data Breach Report 2026Global average a record USD 4.99 million, up 12 percent; US average a record USD 11.5 million; AI-driven attacks added an average of USD 1 million per breach; mean time to identify and contain a breach 247 days. Accessed Nov 18, 2026.
  2. eSecurity Planet, IBM 2026 Cost of a Data Breach Report: Key Findings (Jul 30, 2026)The same figures, summarized: $4.99 million global, $11.5 million US, about $1 million added by AI-driven attacks, 247 days to identify and contain. Accessed Nov 18, 2026.
  3. Allianz Commercial, Allianz Risk Barometer 2026: Cyber remains top business risk but AI fastest riser at #2 (press release, Jan 14, 2026)Cyber ranks first for the fifth time in a row (42 percent of responses); AI jumps from tenth to second (32 percent). Accessed Nov 18, 2026.
  4. Cox, L. A. Jr., What's Wrong with Risk Matrices? Risk Analysis 28(2):497-512 (2008), abstract on PubMedIdentical ratings for quantitatively very different risks; correct comparisons for only a small fraction of random pairs (less than 10 percent in the example); resource allocation cannot rest on matrix categories. Abstract read through NCBI's E-utilities. Accessed Nov 19, 2026.
  5. NIST IR 8286 Rev. 1, Integrating Cybersecurity and Enterprise Risk Management (Dec 2025)"Quantitative methods are preferred"; quantify risk "in financial terms wherever feasible"; Fig. 6, a loss exceedance curve as an example of a quantitative risk measure. Supersedes the withdrawn 2020 edition. Accessed Nov 19, 2026.
  6. NIST IR 8286A Rev. 1, Identifying and Estimating Cybersecurity Risk for ERM (Dec 2025)Three-point estimation; Monte Carlo simulation "perhaps several hundred thousand cycles" to reduce subjectivity. Accessed Nov 19, 2026.
  7. The Open Group, Technical Standard: Risk Taxonomy (C081, 2009)"Risk is the probable frequency and probable magnitude of future loss" (loss event frequency and probable loss magnitude). Accessed Nov 19, 2026.
  8. FAIR Institute, FAIR Model Standard v3.0 (Jan 15, 2025)Risk derived from loss event frequency and loss magnitude, mostly quantified in financial terms. Accessed Nov 19, 2026.
  9. Hubbard Decision Research, How to Measure Anything in Cybersecurity Risk (book site and its one-for-one substitution model)90% confidence intervals, a lognormal impact (mean of the logged bounds, log range divided by 3.29) and a loss exceedance curve; "the quantitative estimates avoid some of the errors introduced by risk matrices." Accessed Nov 19, 2026.

Updates

  • Nov 19, 2026. Added a heat map, the whole calculation on one line and a simulated loss curve, with the research behind the method (Cox 2008, NIST IR 8286 Rev. 1 and 8286A Rev. 1, FAIR, Hubbard). The IBM figures moved into the text; the earlier numbers did not change.
  • Nov 18, 2026. Retitled from the draft title to “Risk that counts”: a rough dollar range for the few risks that decide spending, alongside the heat map, not instead of it.

Get Insights by email

Thanks for subscribing.

Share on LinkedInFollow AusperSubscribe by RSS

Related reading

Free 30-minute check

Find the first risk worth pricing.

Thirty minutes with our team on your code, your vendors and your keys, and what we would fix first. No code, credentials or access needed.

Book the AI exposure review