
Publication Date
November 10, 2026
Category
Cybersecurity Awareness
Reading Time
7 Min
Author Name
Shubh Arya
Security Awareness Metrics: From Counting Training to Measuring Behaviour
Security awareness metrics measure how well a programme changes what people do. The useful ones track behaviour, such as report rate, time to report and consistency across channels, rather than activity such as course completion.
Most dashboards still lead with completion. It is easy to collect and easy to defend, and it says almost nothing about how your people will act when a message lands at the wrong moment. This guide offers a ladder for thinking about metrics, a scorecard you can start on this month, and a table of the statistics people most often cite, with what each one cannot tell you.
The metrics ladder
Think of metrics as four rungs. Most programmes stop on the first or second.
| Rung | It answers | Example metrics | The trap |
|---|---|---|---|
| 1. Activity | Did people do the programme? | Completion rate, hours of training | Shows effort, not change |
| 2. Outcome | Did they fall for the test? | Click rate, credential entry | Depends heavily on how hard the lure was |
| 3. Behaviour | Did they do the thing you want? | Report rate, time to report, consistency | Needs realistic, repeated testing |
| 4. Condition | Under what circumstances? | Results by channel, team, region, pressure | Needs enough data to split safely |
Boards tend to see rung one. NACD's 2026 director handbook on cyber-risk reporting recommends concise indicators of maturity and trend, and the only people-related metric it lists is the percentage of employees completing security training. That is a fine place to start and a poor place to finish.
Why one number never works
The Human OS white paper makes the point bluntly: human risk is contextual and dynamic, so no universal score captures it. Its reasoning is the 4:58 PM problem. The same employee, shown the same request, can respond differently at 10 in the morning and at the end of a long Friday, because workload, emotion and attention differ even when knowledge does not. A metric that ignores context will mislead you.
The paper lists seven things worth tracking: reporting behaviour, response speed, repeat patterns, role and exposure, scenario difficulty, channel resilience and an improvement trend. The scorecard below turns those seven into calculations, and adds an eighth of our own.
The scorecard: eight metrics
| Metric | Human OS measure | How to calculate | Better when | Watch out for |
|---|---|---|---|---|
| 1. Report rate | Reporting behaviour | People who reported a simulated suspicious message ÷ people who received it | Higher | Pair with metric 8, or people report everything |
| 2. Time to report | Response speed | Median time from delivery to report | Lower | Needs timestamps. Use the median, not the mean |
| 3. Click rate by lure type | Scenario difficulty | Clicks ÷ delivered, split by lure category | Lower, within each type | Never average easy and hard lures |
| 4. Channel coverage and report rate per channel | Channel resilience | Report rate separately for email, text, voice, QR and chat | Higher, and even across channels | Coverage alone is not behaviour |
| 5. Repeat rate | Repeat patterns | People who acted on a lure again after feedback ÷ people who acted | Lower | Needs at least three exercises to read |
| 6. Concentration by role | Role and exposure | Share of risky events from the top teams, weighted by access to money or data | Lower, or at least understood | Use teams first. Targeting individuals can feel punitive |
| 7. Quarterly change | Improvement trend | Change in metrics 1 to 3 against the previous quarter, on matched scenarios | Moving the right way | Compare like with like. A harder lure can hide progress |
| 8. Report precision (ours) | Not in the paper | Reports that were real or simulated threats ÷ all reports | Higher | Low precision means the reporting habit needs coaching, not praise |
A platform that tests across channels and shows results by individual, team and region makes metrics 4 and 6 straightforward.
The statistics people cite, and what they cannot tell you
| Statistic | Source | What it can tell you | What it cannot |
|---|---|---|---|
| Text and voice attacks had a success rate 40% higher than email phishing | Verizon 2026 DBIR release, 19 May 2026 | That channels differ, so a score from email tests alone does not cover them | How large the baseline or sample is. The release does not say, so treat it as a direction |
| No significant link between recency of annual training and failing a simulation; training after a failure had an effect of about 1.7 percentage points | UCSD Health, 19,500+ employees, 8 months | That one annual session may not carry far, and post-failure training had a small effect | That training is useless. It was one health system and one approach |
| Click rates of 1.82% and 30.8% for two lures in the same study | UC San Diego news release on the same study | That the lure drives the result | Which lure is closest to your real threats |
| Warnings and reporting helped; voluntary embedded training did not; 14,700 employees, 15 months | ETH Zurich (Lain, Kostiainen, Capkun), 2022 | That reporting and warnings are worth building around | Whether mandatory training would perform differently |
How to read a click rate
In the UCSD Health study, one simulated email, a message about an Outlook password, drew a click rate of 1.82 percent. Another, about a vacation policy, drew 30.8 percent. Same employees, same eight months. The click rate was describing the email, not the people.
This is why NIST developed its Phish Scale. It asks practitioners to rate how hard a phishing email is to detect, looking at the cues in the message and how well it fits the recipient's context, so that two click rates can be compared fairly. You do not need the full method to take the lesson. Report click rates by lure type, never as one average, and treat a falling average with suspicion until you have checked that the lures stayed equally hard.
Evidence on training that follows a failure is mixed. UCSD found a small effect, and ETH Zurich found that voluntary embedded training did not help. The practical answer is to measure whether coaching moves your own numbers, particularly report rate and repeat rate.
Build your first scorecard in an afternoon
Pick three metrics: report rate, time to report and click rate by lure type.
Run one exercise per channel you actually use, so you have a baseline in each.
Write down the definition of each metric before you look at the results, so the numbers cannot be reshaped afterwards.
Set a review date a month out. Look at the change, not the level.
See SimuPhish in actionFAQs
Related Articles

How Habits Become Risk: What Repetition Does to Security Decisions
Security culture is the habits people share when nobody is checking. See how repetition turns routine into risk and how leaders can reset it.
Nov 6•8 Min read

Why Awareness Training Alone Isn't Reducing Human Error
Annual awareness training alone is not reducing human error. Organizations need simulations, behavioral data, and adaptive interventions to measurably lower workforce cyber risk.
Jul 9•6 Min read

Security Awareness Training for SAMA Compliance: A Guide for Saudi Enterprises
Learn how Saudi enterprises can build a SAMA-compliant workforce with an AI-driven Human Risk Management Platform for cybersecurity awareness training, phishing simulations, and audit ready reporting.
Jun 16•9 Min read
