Qualitative Risk Matrix, Decoded

The oldest and most common way to rate a risk: describe how likely it is and how bad it would be in words, then let a table turn the two words into one. Here is the version NIST publishes, how it really behaves, and how to stop two assessors giving the same risk different answers.

5levels, Very Low to Very High, for every factor
2lookup tables: one for likelihood, one for risk
0arithmetic — words are combined by table, not multiplied
25cells in the risk matrix, each a decision made in advance

A qualitative risk assessment rates each factor in words — Very Low, Low, Moderate, High, Very High — and combines them with a lookup table instead of a formula. NIST SP 800-30 Rev. 1, the US federal guide to conducting risk assessments, defines it as assessing risk “based on nonnumerical categories or levels”. Its strength is communication: a board understands “High” at once. Its weakness, in NIST’s own words, is that the range of values is small, so ranking risks within a level is hard, and “unless each value is very clearly defined … different experts … could produce significantly different assessment results.”

Approach (NIST SP 800-30 §2.3.2)What the values areBest at
Qualitative — this pageWords: Very Low … Very HighCommunicating risk to decision makers
Semi-quantitativeBins, scales or representative numbers that mean nothing outside the assessment (1–5, 0–100)Ranking many risks against each other — see Semi-Quantitative Risk Scoring
QuantitativeNumbers that keep their meaning outside the assessment (dollars, events per year)Cost-benefit decisions on controls — see Inherent vs Residual Risk

1 · Five Levels, Three Questions

NIST does not rate “likelihood” in one go. For an adversarial threat it asks three separate questions, each answered on the same five-level scale:

  1. How likely is the adversary to initiate the threat event? (Table G-2) — from “highly unlikely” (Very Low) to “almost certain” (Very High). For accidents and natural events, Table G-3 adds a frequency: Very High is more than 100 times a year, Moderate 1–10 times a year, Very Low less than once every 10 years.
  2. If it happens, how likely is it to cause adverse impact? (Table G-4) — this is where existing controls show up: a well-defended system makes a successful attack less likely.
  3. How bad would the impact be? (Table H-3) — from negligible (Very Low) through limited (Low), serious (Moderate) and severe or catastrophic (High) to multiple severe or catastrophic effects (Very High).
LevelImpact (Table H-3), in brief
Very HighMultiple severe or catastrophic adverse effects on operations, assets, individuals, other organisations or the Nation
HighA severe or catastrophic effect — e.g. unable to perform one or more primary functions, major damage to assets, major financial loss, loss of life
ModerateA serious effect — primary functions still performed but significantly less effective, significant damage or financial loss
LowA limited effect — effectiveness noticeably reduced, minor damage, minor financial loss, minor harm
Very LowA negligible effect

Each of these NIST tables also carries a semi-quantitative column (e.g. Very High = 96–100 or 10). A purely qualitative assessment ignores it; the semi-quantitative page uses it.

2 · NIST’s Two Lookup Tables

The words are combined in two steps. First, Table G-5 turns the two likelihood answers into an overall likelihood. Then Table I-2 combines overall likelihood with impact into the level of risk:

Likelihood ↓  /  Impact →Very LowLowModerateHighVery High

Table I-2, NIST SP 800-30 Rev. 1, Appendix I. Every cell is NIST’s published value.

▸ Read the table closely and it is not symmetric. A Very High likelihood of a Very Low impact is Very Low risk; a Low likelihood of a Very High impact is Moderate. The High and Very High likelihood rows are identical. NIST’s matrix lets impact drive the answer — a design choice your own matrix should make deliberately, not inherit.

The overall-likelihood table (G-5) behaves the same way: it combines “will they try?” with “will it work?”, and a Very High chance of an attempt that is very unlikely to succeed comes out Low.

3 · Worked Example: Ransomware via Remote Access

The same risk used across this series: ransomware delivered through a compromised remote-access account. Each answer is annotated, as NIST recommends (“this value is high because …”).

QuestionAnswerBecause
Likelihood of initiation (G-2)HighInternet-facing VPN; ransomware crews routinely buy and spray stolen remote-access credentials
Likelihood of adverse impact (G-4)ModerateMFA on the VPN, but EDR coverage is partial and backups are not tested
Overall likelihood (G-5)ModerateHigh × Moderate in Table G-5
Impact (H-3)HighCore systems down for days: unable to perform a primary function, major financial loss
Level of risk (I-2)ModerateModerate × High in Table I-2

Now remove MFA from the VPN. The chance an attempt succeeds rises to High, overall likelihood becomes High, and the level of risk becomes High. That one control moves the risk a full level — exactly the kind of statement a qualitative assessment makes well.

Try it: NIST SP 800-30 qualitative risk

Overall likelihood (G-5)
Level of risk (I-2)

The selected cell is outlined in the Table I-2 matrix above. Both lookups use NIST’s published tables exactly.

4 · Writing Levels People Agree On

The failure mode of every qualitative scale is that words mean different things to different people. Sherman Kent showed it at the CIA in the 1960s (“Words of Estimative Probability”): analysts reading the same report put very different odds on the word “probable”. US intelligence now pins its words to ranges in Intelligence Community Directive 203 — “likely” means roughly 55–80%, “very likely” 80–95%. Do the same in your risk methodology:

  • Anchor likelihood to frequency or odds — “High: expected once or more in the next 12 months”, as NIST’s Table G-3 does for non-adversarial events.
  • Anchor impact to your own thresholds — a revenue figure, hours of downtime, number of records, a regulator notification. “Major financial loss” means nothing until you put a number on it.
  • Annotate every rating — one line of “because”. NIST credits annotation with raising repeatability and reproducibility.
  • Calibrate on old incidents — have two assessors rate three past events independently and compare. Disagreement means the definitions need work, not the assessors.
  • Fix the matrix in policy — decide once whether impact dominates (as in NIST’s I-2) and publish the table; do not let each assessment improvise.

5 · When Qualitative Is the Right Call

Use it whenMove on when
You are building a first risk register and need coverage fastYou need to rank 40 risks that all came out “Moderate” — go semi-quantitative
Data is thin and an honest answer is a range of opinionA control costs real money and you must show it pays for itself — go quantitative
The audience is a board or a regulator who wants a level, not a modelYou are rating individual application flaws — use OWASP Risk Rating or CVSS and SSVC

IEC 31010:2019, the international catalogue of risk assessment techniques, lists the consequence/likelihood matrix among its techniques and notes the same trade-off: easy to use and communicate, but coarse and dependent on well-designed scales.

Bottom line

  • Words, combined by table — qualitative assessment never multiplies; a published matrix decides every combination in advance.
  • Likelihood is two questions — will they try, and will it work. Controls live in the second.
  • NIST’s matrix is impact-weighted — choose that deliberately if you copy it.
  • Define every level with a number or an example, and annotate every rating, or two assessors will disagree.
  • Great for communication, weak for ranking — graduate to scores or money when the decisions need it.

Sources & method. Definitions, scales and both lookup tables from NIST SP 800-30 Rev. 1, Guide for Conducting Risk Assessments (September 2012): §2.3.2 and Tables G-2 to G-5, H-3, I-2 and I-3, transcribed from the published PDF; short quotations are from §2.3.2. Risk matrix as a technique: IEC 31010:2019, Annex B (cited, not quoted). Sherman Kent, “Words of Estimative Probability”, Studies in Intelligence (CIA, 1964); probability ranges from ODNI Intelligence Community Directive 203, Analytic Standards (2015). The worked example is illustrative. Original analysis; no third-party graphics reproduced.

← Back to Cyber security