ISO 31000 Risk Matrix: How to Build and Calibrate One (With Scales and Worked Example)

October 25, 2026

Search for "ISO 31000 risk matrix" and you will find a great many 5×5 grids with the ISO name nearby and almost nothing on how the grid was built. That is the wrong way round. The grid is the last ten minutes of the work; the scales, criteria and calibration behind it decide whether the register is usable. This guide builds a matrix from the criteria up, then applies it to a small regulated firm so you can see the numbers.

For the failure modes once a matrix is live, see our risk matrix best practices. This article is about getting the design right in the first place.

What ISO 31000 says about matrices — and what it does not

ISO 31000:2018 is a guidance standard. It is not certifiable, it mandates no tools, and it never mentions a 5×5 grid. What it does require, in the risk assessment part of the process, are three things that a matrix is one way of delivering:

  • Risk criteria (clause 6.3.4). Before assessing anything, the organisation defines how it will measure likelihood and consequence, over what time horizon, how the level of risk is determined, and what level is acceptable. The matrix is a visual expression of these criteria. If the criteria do not exist in writing, the matrix has no meaning.
  • Risk analysis (6.4.3). Understanding the nature of each risk, its sources, consequences, likelihood and the effectiveness of existing controls. The standard says analysis can be qualitative, quantitative or a combination, and that the level of detail should match the purpose.
  • Risk evaluation (6.4.4). Comparing the analysis with the criteria to decide whether anything further needs to be done. This is where colour bands and appetite enter.

The companion standard, IEC 31010, lists the consequence/likelihood matrix as one technique among around thirty and is candid about its weaknesses: ordinal scales, subjectivity, and the difficulty of aggregating results. ISO 31000 is therefore neither for nor against matrices. It asks that your method be defined, consistent, and fit for the decisions it informs. A matrix meets that bar only if it rests on explicit scales, is applied the same way by different people, and connects to decisions about treatment.

That framing answers a question we hear often: there is no "ISO 31000 compliant" matrix to buy. There is a matrix whose criteria you can explain to an auditor or supervisor, and one you cannot. Our risk regulations page covers how the standard interacts with sector rules.

Step 1: Fix the criteria before drawing the grid

Agree, in writing and ideally at board level:

  1. Time horizon for likelihood — usually 12 months for operational and compliance risk, the planning period for strategic risk. State it once and use it everywhere.
  2. Impact dimensions that matter — financial, regulatory, reputational and operational are the usual four; add safety, client harm or data protection where relevant.
  3. The anchor for financial impact — a percentage of own funds, annual revenue or profit before tax. Absolute numbers age badly; percentages travel with the business.
  4. The appetite statement the colour bands will encode. If the board has not said what it is unwilling to accept, nobody can colour the grid honestly.

Teams that skip this step produce a matrix in an afternoon and rebuild it the following year. Teams that do it spend two meetings on criteria and the matrix follows in an hour.

Step 2: The likelihood scale

Each level needs a descriptor, a probability band over the stated horizon, and a frequency equivalent for risks that recur. Give assessors both the probability and the frequency; some think in one, some in the other, and offering both reduces disagreement.

LevelDescriptorProbability in 12 monthsFrequency equivalent
1Rare< 5%Less than once in 20 years
2Unlikely5–20%Once in 5–20 years
3Possible20–50%Once in 2–5 years
4Likely50–80%Most years
5Almost certain> 80%At least annually, or already happening

Two design choices worth making deliberately. The bands are not equal: the scale is roughly logarithmic, because the difference between 2% and 10% matters more than the difference between 60% and 70%. And level 5 includes "already happening" — a control gap you know about today is not a 40% risk.

Step 3: The impact scale

Write one descriptor per level per dimension. A risk takes the highest level it reaches on any single dimension; never average across dimensions. The example below is anchored to a firm with own funds of €1.5m and revenue of €4m — scale the financial column to yours.

LevelFinancialRegulatoryReputationalOperational
1 Negligible< €15k (< 1% of own funds)Informal query, no findingInternal onlyMinor delay, one team, under a day
2 Minor€15k–€75k (1–5%)Finding requiring a remediation planSingle client complaint, resolvedOne function disrupted, 1–3 days
3 Moderate€75k–€225k (5–15%)Reportable breach; on-site inspection triggeredSeveral clients affected; local press or industry attentionCore process degraded up to a week
4 Major€225k–€450k (15–30%)Enforcement action, fine, conditions on licenceNational media; loss of a top-5 clientCore service unavailable 1–5 days
5 Severe> €450k (> 30%)Licence suspension or withdrawal; personal sanctions on managementSustained coverage; loss of several key clients or a banking relationshipCore service unavailable over 5 days; viability in question

The percentages in the financial column are the part that transfers between firms. A bank would anchor to a fraction of its capital headroom above requirements; a corporate service provider might anchor to months of operating cost. The regulatory column should be written in the vocabulary of your own supervisor — a CySEC-regulated firm, a Central Bank of Ireland-regulated firm and an FCA-regulated firm would phrase levels 3 to 5 differently.

5×5 risk matrix: likelihood × impact
L \ II5I4I3I2I1
L154321
L2108642
L31512963
L420161284
L5252015105

Green: low · Yellow: medium · Orange: high · Red: extreme. Calibrate scales to your risk appetite.

Step 4: Appetite bands — colouring the grid

The product of likelihood and impact gives 25 cells with scores from 1 to 25. A naive approach colours by score bands alone. A better approach starts from the appetite statement and asks, for each cell, what the board would want to happen if a risk sat there.

BandCells (typical)What it meansWho decides
GreenScore 1–5Within appetite; monitor on the normal cycleRisk owner
AmberScore 6–11Review; treat, or accept with documented rationaleExecutive committee
RedScore 12–25, plus any risk at impact level 5 regardless of likelihoodOutside appetite; must be treated or formally escalatedBoard or board risk committee

The override in the last row is the important design decision. A likelihood-1, impact-5 risk scores 5 — the same as a routine 5-by-1 nuisance — but a rare event that threatens the licence is not within the appetite of any regulated firm. Colour the whole impact-5 row red, or at least amber, so the matrix reflects the appetite rather than the arithmetic. Conversely, some firms colour the 5-by-1 cell green: a daily irritation with negligible impact rarely merits board time.

Record the band definitions alongside the scales. Changing a band is a change to the appetite statement and should be approved as one.

Step 5: Inherent and residual scoring

Inherent (gross) risk is the level before the specific controls you rely on. Residual (net) risk is the level with those controls operating as designed. The gap between them is your control reliance — and the amount of risk that returns if the controls fail.

Three practical rules:

  • Do not score inherent risk in a fantasy world with no controls at all; score it without the controls listed against that risk. Otherwise every inherent score is 5×5 and the exercise is meaningless.
  • Use the same scales for both. Changing the scale between inherent and residual is a common way to make a register look better than it is.
  • Residual scores should be supported by control testing. A residual score that assumes controls work, when the last test of those controls failed, is a wish. Where a control has not been tested, score residual as if it operated partially.

Report both on the heat map. A board that sees only residual scores is seeing the risk function's opinion of itself.

Common mistakes

Compression. Seventy per cent of the register scores 3×3. The scales are too vague or assessors are hedging. Fix by putting numbers in the descriptors and challenging each 3 with "what would make it a 2 or a 4?"

False precision. Scoring 3.5, weighting dimensions to two decimal places, or multiplying ordinal numbers and treating the product as a quantity. A score of 12 is not twice as bad as 6; it is higher. Decimals and weighted sums on ordinal scales are arithmetic performed on labels.

Averaging across dimensions. A risk that is regulatory-severe and financially-negligible is severe. The mean of 5 and 1 is 3, and 3 is wrong.

Horizon drift. One assessor thinks in one year, another in five. Likelihood scores diverge by two levels and nobody notices because the horizon was never written down.

Self-scoring without challenge. Owners score their own risks and a red implies criticism. Build second-line challenge into the cycle, and compare two independent assessments on a pilot set before rollout.

Scoring categories instead of scenarios. "Cyber risk" is not scoreable. "Ransomware encrypts the file server and the last clean backup is 48 hours old" is. Write risks as cause, event and consequence.

A scale per department. Finance uses 1–5, IT uses 1–10, compliance uses high/medium/low. Nothing aggregates. One set of scales, enterprise-wide, with department-specific examples under each level if needed.

Heat map design

The matrix and the heat map are different objects: the matrix is the scoring instrument, the heat map is a report. Design the report for the reader:

  • Likelihood on one axis, impact on the other, with the highest-risk corner in the same position every time. Readers learn the shape; do not rotate it between meetings.
  • Show counts per cell or numbered dots, not a cloud of labels. Put the risk list beside the map.
  • Show movement: inherent-to-residual arrows for the top risks, and markers for risks whose score changed since the last report.
  • Use the same three or four colours as the appetite bands, and nothing else. Gradient fills and 3D effects add nothing and print badly.
  • Add a date and a scale version in the footer. A heat map from a register whose scales changed last quarter is not comparable with the previous one, and the board should be told.

Linking the matrix to KRIs and treatment actions

A matrix reviewed annually is a historical document. Two links keep it current:

  • Key risk indicators are the live feed that should move a likelihood score. Each amber or red risk needs at least one KRI with thresholds; when the indicator turns red, the risk is re-scored, not just noted. Our KRI examples library gives concrete indicators by risk type.
  • Treatment actions are what a red or amber score obliges. Every risk outside appetite needs a treatment option (treat, tolerate, transfer, terminate), a named owner, a dated action and a target residual score. When the action closes and the control is tested, the residual score changes and the heat map changes with it.

In a risk platform such as ActiveERM, the scales, appetite bands, KRIs and actions sit in one place, so a breached indicator or a closed action updates the heat map rather than waiting for the next workshop. The mechanism matters less than the discipline: a score should never change without a reason you can point to, and a reason should never occur without the score being reconsidered.

Worked example: a small regulated firm

The firm: an investment firm with 25 staff, €1.5m own funds, portfolio management and reception and transmission of orders, outsourced IT, one compliance officer. Scales as above; 12-month horizon.

Risk scenarioInherent L × IKey controlsResidual L × IBandTreatment and KRI
Client-money reconciliation fails to detect a shortfall for more than one day3 × 5 = 15Daily internal reconciliation, four-eyes sign-off, monthly external reconciliation2 × 5 = 10Red (impact-5 override)Tolerate at red with board acknowledgement; KRI: unexplained reconciliation differences, red if unresolved after one day
Ransomware encrypts the file server; backups are 48 hours old4 × 4 = 16Immutable daily backups, endpoint detection, MFA, quarterly restore test2 × 3 = 6AmberTreat: move to 4-hourly snapshots, target 2 × 2; KRI: restore test failures, critical patch lag
Compliance officer leaves; role vacant more than 90 days3 × 4 = 12Deputy appointed, documented procedures, retained external compliance consultant3 × 2 = 6AmberTreat: cross-train deputy to regulatory-approval standard; KRI: critical vacancy duration
Outsourced IT provider ceases operations2 × 4 = 8Contractual exit clause, documented environment, escrow of configuration2 × 3 = 6AmberTreat: test the exit plan within 12 months, as DORA expects; KRI: critical providers without a tested exit plan
Suitability assessments become stale across managed accounts4 × 3 = 12Annual review workflow, system block on trades for overdue files2 × 3 = 6AmberTolerate; KRI: suitability file exceptions

Three things the example shows. The client-money risk stays red after controls because the impact override says so — and that is correct, because the board should see it every quarter. The ransomware risk's residual impact drops from 4 to 3 because restore testing changes the consequence, not just the likelihood; controls can act on either axis. And the register has one red, four ambers and, in the full version, a long green tail, which is what a calibrated matrix for a firm of this size should look like.

When this firm first ran the exercise, eleven of thirty risks came out red. Two changes fixed it: the financial scale had been copied from a template built for a firm ten times larger, so a €1m top band was replaced with €450k; and inherent scores were being compared against residual bands. Neither was a scoring error. Both were calibration errors, which is the point of the whole exercise — and the kind of problem discussed in our piece on the challenges of enterprise risk management.

Frequently asked questions

Does ISO 31000 require a 5×5 matrix?

No. It requires defined risk criteria and a consistent method of analysis and evaluation. A 3×3, 4×4 or 5×5 matrix, or a quantitative method, can all satisfy that. 5×5 is popular because it discriminates better than 3×3 without implying more precision than most firms have.

Should the matrix be symmetrical?

Not necessarily. Most regulated firms treat high-impact, low-likelihood risks more seriously than low-impact, high-likelihood ones, and the colouring should reflect that asymmetry.

Is a risk matrix enough for regulatory purposes?

For a small firm's ICARA, ORSA or operational risk framework, a well-documented matrix with written scales, appetite bands and evidence of use is normally accepted. Supervisors object to matrices that are undocumented or static, not to matrices as such. Larger firms will need quantification alongside it.

How often should scales be recalibrated?

Review annually; change rarely. Recalibrate when own funds or revenue have moved enough that the financial bands no longer represent the stated percentages, or when the register shows compression or an implausible colour distribution.

What is the difference between a risk matrix and a heat map?

The matrix is the scoring tool — the scales and bands. The heat map is the picture of where the risks currently sit on it. One matrix produces a new heat map each reporting period.

To see the scales and bands above implemented as a live register, explore Risk Management or request a demo.

Explore ActiveERM

See how ActiveERM helps you with governance, risk, compliance, and audit in one platform.