The Complete Guide to Security Awareness Training
What actually changes staff behavior, why annual modules fail, how to run phishing simulations without destroying trust, and the metrics worth reporting.

Two things happen in the same week. The compliance dashboard shows 98 percent of staff have completed the annual security module, with an average quiz score in the nineties. On Thursday, an accounts payable clerk pays an invoice to a bank account that changed by email three weeks earlier, and nobody in the finance team mentions that four of them received the same message and deleted it.
Both facts are real and they do not contradict each other. The module measured whether people clicked through a presentation. The fraud depended on whether anyone picked up a phone, and whether anyone thought it was their job to say something. Those are different questions, and most awareness programs only answer the first one.
This guide is about building the kind that answers the second.
What security awareness training is actually for
The purpose of an awareness program is to change what people do under pressure, in the few seconds when a decision is being made and nobody is watching. Everything else is a means to that end.
That reframing matters because it changes what counts as success. If the goal is knowledge, the natural measure is a quiz. If the goal is behavior, the measures are things like how many suspicious messages reach the security team, how fast the first one arrives, and what proportion of new supplier bank details were confirmed by a call before money moved. NIST's guidance on building a cybersecurity and privacy learning program makes a similar distinction: a learning program is defined by the outcomes it is designed to produce for specific roles, not by the material it distributes.
It also sets an honest limit. Awareness training cannot make people into reliable detectors of malicious email. The NCSC has argued this directly, pointing out that telling users to avoid clicking bad links does not work as a strategy, because modern phishing is designed to be indistinguishable from legitimate work. A program that depends on everyone spotting everything is a program that has outsourced a technical problem to the least technical part of the organization.
What training can do is narrower and more valuable:
- Make reporting the default. One person reporting a message quickly protects everyone who received it.
- Install a small number of hard rules that work regardless of whether the message looked convincing, such as never acting on a payment instruction that arrived only in writing.
- Make it socially safe to pause, check and be wrong.
- Remove the friction that pushes people into unsafe workarounds, such as shared logins or personal accounts.
Why the annual module fails
The once-a-year compliance module persists because it is cheap, auditable and easy to buy. It fails for reasons that have nothing to do with its content quality.
Completion is a measure of administration, not capability
A completion rate tells you how well your reminder emails work. It says nothing about behavior. Worse, it becomes the number that gets reported upward, which means the program is managed toward a metric that cannot improve security.
The timing is wrong
People retain security information for a short period and then revert. A single session in March does nothing for a decision made in October. Spaced repetition is not a nice extra here, it is the mechanism by which anything sticks.
It is about the wrong threats
Generic modules cover a generic organization. If your real exposure is supplier bank-detail changes and mailbox takeover, a module about USB sticks and shoulder surfing wastes the time it takes and signals that the program is a formality. Train against what your own incident log and your own industry's fraud patterns actually show.
It treats everyone the same
A payroll administrator, a software engineer, a warehouse supervisor and a finance director face different attacks and have different powers to cause harm. One module for all of them has to pitch at the lowest common denominator, which is roughly nobody's actual job.
It often arrives as a punishment
When extra training is assigned only to people who clicked something, training becomes a sanction. People then optimize to avoid being assigned it, which means avoiding being noticed, which means not reporting.
Decide which behaviors you are buying
Before choosing a platform or writing a module, write down the specific behaviors you want. Keep the list short enough that a manager could recite it. Most organizations need some version of these:
- Report anything suspicious, immediately, through one obvious route, and do it even when you are not sure and even when you already clicked.
- Verify any request to move money or change payment details by calling a number you already hold, never a number in the message.
- Never approve an authentication prompt you did not personally trigger, and report it when one arrives.
- Use the password manager and the company's own tools rather than personal accounts, personal storage or shared logins.
- Raise your own mistakes fast, including clicking a link, entering a password or sending data to the wrong person.
- Know what data you hold and who is allowed to receive it.
Each of those is observable. You can count reports. You can audit whether verification calls happened. You can measure how long it took someone to admit a mistake. That is what makes them useful targets rather than slogans. Our article on human risk management covers how to organize a program around behaviors rather than topics.
One warning. Do not train a behavior the organization has not made possible. Telling staff to verify supplier bank changes by phone is pointless if the supplier master file has no independently sourced phone numbers in it. Build the control first, then train the behavior that uses it.
Measuring behavior change rather than completion
This is where most programs can improve fastest, because better measurement changes what people do without costing anything.
The measures worth having
- Report rate. Of the people who received a suspicious or simulated message, how many reported it. This is the single most useful number in an awareness program, because reporting is the behavior that actually produces defensive value.
- Time to first report. How many minutes from delivery to the first report reaching the security team. Containment speed depends on this far more than on how many people eventually reported.
- Self-reported mistakes. How many people report that they clicked, entered credentials or sent something to the wrong recipient. A rising number here is usually good news about trust, not bad news about competence.
- Verification coverage. The proportion of new or changed payees in the period that have a documented out-of-band verification. This is the metric that maps most directly to prevented loss. Our guide to business email compromise explains why.
- Repeat-click concentration. What share of simulation clicks come from the same small group across several exercises. A concentrated pattern means targeted support for a handful of people, not another all-staff module.
- Response quality. How long the security team takes to acknowledge and close out a report. If people report and hear nothing, reporting decays.
Why click rate on its own is close to meaningless
A click rate is a function of difficulty. NIST developed the Phish Scale precisely because the same population will produce wildly different click rates depending on how plausible the cues in the email are and how well the premise aligns with the recipient's actual work. A program that wants its numbers to improve can simply send easier emails. A program that wants to learn something sends emails of known, recorded difficulty and tracks reporting alongside clicking.
If you report click rate at all, report it with the difficulty rating attached and never compare two exercises of different difficulty as though the change meant something.
Set a baseline and segment it
Measure before you intervene, then break results down by team, role and tenure. Aggregate numbers hide the thing you need to know, which is usually that one department with high-value access is well below everyone else. Our article on measuring security progress goes through building a small set of numbers you can actually sustain.
Report upward in terms of exposure
A board does not need click rates. It needs to know whether the organization would detect and contain a realistic attack, and whether the controls that stop the expensive outcomes are in place and tested. The NCSC's board toolkit frames these as governance questions, which is the right framing: leadership's job is to resource and endorse the program, not to admire its training statistics.
Phishing simulation done well
Simulated phishing is a legitimate and useful tool, badly used almost everywhere. The difference between the two versions is mostly about purpose.
A good simulation program exists to practice and measure reporting, to find out where the organization is blind, and to give people a safe rehearsal of a decision they will eventually face for real. It does not exist to catch people out or to generate a number for a report.
What that looks like in practice
- Announce the program, not the tests. Tell everyone, in plain language, that the organization sends practice phishing emails, why it does so, what is measured, what is not measured, and that nobody is disciplined for clicking. Then do not announce individual exercises.
- Make the report button the point. Every simulation should be an opportunity to use the real reporting mechanism, and reporting should give instant positive feedback. Our article on the phishing report button covers the mechanics.
- Calibrate difficulty deliberately. Mix easy, moderate and hard exercises, record the difficulty of each, and use the hard ones to learn rather than to score. The purpose of a genuinely difficult test is to show the organization that detection cannot be the only defense.
- Match the scenario to the role. Send finance a supplier bank-detail change. Send HR a candidate attachment. Send IT administrators a credential prompt. Send the executive assistant an urgent request from a traveling director. A generic password-reset lure teaches a payroll team nothing about the attack that will actually reach them, which is why payment fraud training needs its own scenarios.
- Keep the follow-up short and immediate. Someone who clicks should get a brief, specific, non-judgmental explanation of what the email did and what to do next time, delivered in under two minutes of their attention. Not a thirty-minute course.
- Run at a steady, moderate cadence. Frequent enough to be normal, rare enough not to become noise. Every four to eight weeks for most of the organization works.
- Test the process, not only the person. When someone reports, measure what happens next: did the security team pull the message from other mailboxes, did they block the domain, did they tell the reporter what happened. Half of what simulations expose is weakness in the response path.
Our article on how phishing simulation works goes into the operational detail of setting one up.
Phishing simulation done badly
The failure modes are consistent, and they do real damage. The worst of them cost more than they prevent, because they destroy the reporting culture that is the main thing you were trying to build.
Lures that abuse the employment relationship
Fake bonus announcements, fake redundancy consultations, fake payroll errors, fake news about health benefits. These work extremely well, which is why people reach for them, and they are the fastest way to make staff feel that the security team is their adversary. An employee who feels tricked about their salary does not become more careful. They become less cooperative, permanently. The topic is also usually irrelevant: the attack you need them to resist is a supplier payment request, not a bonus email.
Naming, shaming and leaderboards
Publishing lists of people who clicked, or team-versus-team scoreboards, converts the exercise into a social threat. The predictable result is that people stop reporting marginal cases, because reporting draws attention. Our article on phishing simulations and staff trust covers the damage in more detail.
Using results in performance management
The moment a simulation result can appear in an appraisal, you have stopped measuring behavior and started measuring fear. People will begin forwarding every internal email to colleagues for a second opinion, or refusing to open legitimate attachments, and the cost lands on the business rather than the attacker.
Impossible tests presented as failures
It is entirely possible to build a simulated email that essentially nobody can distinguish from real work, particularly if you use internal branding, a real ongoing project and a compromised-looking internal sender. That exercise is a useful demonstration of a technical gap. Presenting its click rate as a staff failure is dishonest.
Measuring only the click
Many platforms default to click and credential-entry statistics and treat reporting as an afterthought. If your dashboard leads with clicks, your program will drift toward punishing clicks. Reconfigure the reporting so report rate is the headline.
Simulation instead of controls
The most expensive mistake is to run a mature simulation program while leaving phishing-resistant multi-factor authentication unconfigured, external forwarding unmonitored and payment verification undocumented. Simulation is the last layer, not the first. MITRE ATT&CK catalogs phishing as an initial access technique precisely because its value to an attacker comes from what happens after the click, and that is mostly a technical question. Our guide to phishing and how to defend against it sets out the layers that should be in place first.
Building a reporting culture people trust
If you only do one thing from this guide, do this one. An organization where suspicious messages arrive at the security team within minutes has a genuinely different risk profile from one where they do not, and the gap is almost entirely cultural.
Make reporting take one click
A report button in the mail client, present on phones as well as laptops. If reporting means composing a new email, attaching the original with full headers and finding the right address, most people will delete the message instead and move on.
Acknowledge every single report
Automatically within seconds, and by a human within a working day for anything genuine. The acknowledgement should thank the person without qualification, including when the message was legitimate. "Thanks, this one was a real invoice from our stationery supplier, and you were right to check" is the sentence that produces the next report.
Say out loud that false alarms are the point
Tell people explicitly that you would rather have a hundred unnecessary reports than miss one real attack, and that nobody has ever been criticized for checking. Then behave that way, every time, including when the security team is busy.
Never punish a self-report
Someone who clicked a link, entered a password and tells you within five minutes has given you the best possible outcome. Treat them accordingly and make sure their manager knows that is the expected response. The practical guidance in what to do after clicking a phishing link is worth circulating so people know the first step is to speak up.
Close the loop publicly
When a report leads to something, say so. A short internal note saying a colleague's report led to a malicious domain being blocked before anyone else saw it does more for participation than any poster campaign. Keep it anonymous unless the person wants the credit.
Have leaders model it
When a director forwards a suspicious message and says publicly that they almost fell for it, the social cost of checking drops for everyone below them. When leadership instead treats verification as an insult to their authority, every payment control in the organization becomes negotiable, which is exactly the condition social engineering depends on.
Fix the friction people route around
If staff are using personal email to send files because the approved method fails on mobile, that is a security problem created by IT, not by staff. Collect the workarounds, publish what you fixed, and people will tell you about the next one.
Not sure about a message right now? Paste it into CyberWatch AI Scan for a free check, then report it to your security team either way.
Content that actually lands
Assume you have a few minutes of each person's attention every month or two. Spend it well.
- Short and single-purpose. One behavior per interaction. Three minutes, not thirty.
- Specific to the work. Use the systems, job titles and suppliers people recognize. Generic stock scenarios get discounted immediately.
- Built on stories. A described incident with a timeline and a decision point is remembered. A list of rules is not. Use real public cases or your own anonymized near misses.
- Honest about difficulty. Saying "this one was genuinely hard to spot, here is the process that catches it anyway" earns more credibility than pretending every attack has obvious tells. Our article on why phishing works is a useful framing for this.
- Delivered where people are. Mail, chat, team meetings, printed cards for non-desk staff. Do not assume a browser-based module reaches everyone.
- Just in time. A short prompt attached to the moment of risk, such as a banner on external email or a reminder inside the payment system when a payee is new, outperforms anything delivered weeks earlier.
Role-based paths
Everyone gets the core behaviors. These groups need more, and should get it as a short, specific path rather than extra volume of the same material.
- Finance, accounts payable and procurement. Supplier bank-detail changes, invoice interception, urgency from senior names, out-of-band verification, dual authorization. See finance team fraud training.
- HR and payroll. Payroll diversion, candidate attachments, bulk personal data requests, pressure near pay-run deadlines.
- IT administrators. Credential phishing against privileged accounts, authentication fatigue attacks, pretexted helpdesk calls requesting resets, malicious application consent.
- Executives and their assistants. Targeted impersonation, voice cloning, calendar and travel-based pretexts, and the specific instruction that they will never be asked to approve a payment by message alone.
- Developers and engineers. Credentials in code, dependency and package risks, access to production data, tooling that bypasses corporate controls.
- Customer-facing and frontline staff. Caller verification, refund and chargeback pretexts, physical access requests, and what they are allowed to confirm to an unknown caller.
Governance, ownership and budget
Awareness programs die from diffuse ownership. Name one accountable person. Give them the authority to change a process, not only to produce a report, because most of the behaviors on your list depend on a process existing in the first place.
A workable split is: security or IT owns content, scenarios and measurement; HR owns enrollment, onboarding, record keeping and anything that touches employment terms; leadership owns the visible endorsement and the resourcing. Agree in writing what simulation results may and may not be used for, and have HR sign that off before the first exercise. Doing that afterwards, during a dispute, goes badly.
Map the program to whatever framework you already use so it is not reinvented every audit cycle. The NIST Cybersecurity Framework has awareness and training as a named category, which makes it straightforward to show that the program is a managed capability rather than an annual event.
A realistic first year
This sequence assumes a small team and no dedicated budget beyond a simulation tool.
- Month one. Write the short list of target behaviors. Check which of them the organization has actually made possible, and fix the gaps. Deploy a one-click report button everywhere, including mobile. Agree the no-punishment rule in writing with HR and leadership.
- Month two. Baseline. Run one moderate-difficulty simulation across the organization and record report rate, time to first report and click rate with its difficulty noted. Do not publish individual results.
- Month three. Announce the program properly, in leadership's own words, including what is measured and what will never be used against anyone. Deliver one short core session, under fifteen minutes, covering reporting and payment verification.
- Months four to six. Start the simulation rhythm at one exercise every six weeks, with role-appropriate scenarios. Begin the monthly three-minute nudge. Publish one anonymized near miss internally.
- Months seven to nine. Build the role-based paths for finance, HR, IT administrators and executives. Run a tabletop exercise with the people who would handle a real incident. The NCSC's Exercise in a Box and CISA's tabletop exercise packages both provide ready-made scenarios, which saves writing one from scratch.
- Months ten to twelve. Re-measure against the month-two baseline using an exercise of comparable difficulty. Review where repeat clicking concentrates and offer those people direct support rather than more modules. Report to leadership in terms of exposure and tested controls, then set next year's behaviors.
Objections you will hear, and honest answers
"Our people are smart, they do not need this." Intelligence is not the variable. Attention, time pressure and plausible context are, and attackers choose the moment. The point of the program is to install a process that works when attention is low.
"We cannot afford to lose working hours." Then do not spend thirty minutes a year per person on a module nobody remembers. Spend three minutes every six weeks on one behavior, and put the saved time into the payment controls that stop the expensive outcome.
"Training does not work, so let us just buy technology." Buy the technology first, genuinely. Then recognize that the residual risk after good technical controls is concentrated in human decisions about money and access, and that reporting speed is the only lever you have on detection time for attacks that carry no malicious payload at all.
"Our click rate is already low." Check the difficulty of your tests and your report rate. A low click rate with a low report rate usually means people are deleting suspicious mail silently, which leaves the security team blind.
Where CyberWatch AI fits
CyberWatch AI lets an organization run its own realistic practice emails, including the payment and supplier scenarios that matter most to finance teams, and see who verifies, who reports and who pays. Reporting is a single step for the employee, and the follow-up training is short and specific to the pattern they met rather than a generic annual module. Anyone unsure about a real message can check it and pass it to the security team at the same time. Our article on security awareness training that works covers the day-to-day running of a program like this.
The short version
Stop measuring completion. Pick five or six behaviors that attackers depend on, make each of them possible before you train it, and measure reporting speed and verification coverage instead of quiz scores. Run phishing simulations to practice reporting rather than to catch people, never with lures that abuse the employment relationship and never with a leaderboard. Acknowledge every report, protect anyone who self-reports a mistake, and have leadership say publicly that checking is expected.
A program built that way produces something a module never will: an organization where the first person to see an attack tells somebody, within minutes, because they expect to be thanked for it.
For the wider program this sits inside, see our guides to business cybersecurity and incident response.
Frequently asked questions
What is security awareness training?
It is the set of activities an organization uses to change how its staff behave around security risks: recognizing and reporting suspicious messages, verifying unusual requests, handling data properly and raising mistakes quickly. The goal is changed behavior, not completed modules.
How often should security awareness training happen?
Short, frequent contact works better than one long annual session. A sensible rhythm is a brief interaction every four to eight weeks, plus immediate training when something relevant happens, plus a longer onboarding session for new joiners.
Does security awareness training actually reduce risk?
It reduces risk when it targets specific behaviors that attackers depend on, especially reporting speed and out-of-band verification of payment requests. Training that only transfers knowledge, measured only by completion, has little measurable effect.
What is the right phishing simulation click rate?
There is no right number, because click rate depends almost entirely on how hard you made the email. A low rate usually means the test was easy. Report rate and time to first report are far more useful measures.
Should employees be punished for failing a phishing test?
No. Punishment teaches people to hide mistakes, which is the opposite of what you need. The one case for a formal conversation is repeated deliberate circumvention of a control, and that is a management issue rather than a training one.
What should we measure instead of completion rates?
Report rate, median time from delivery to first report, the share of new or changed payees with a documented verification call, how quickly people report their own mistakes, and how concentrated repeat clicking is in a small group.
How do we train staff who do not sit at a desk?
Use the channels they already use. Short messages, team briefings, printed one-pagers in break areas and a phone number they can call. A web module that assumes a corporate laptop will not reach a delivery driver or a shop floor team.
Is it fair to send simulated phishing emails without warning?
Tell staff in advance that simulations are part of the program and explain why, then do not announce individual tests. The surprise is in the timing, not in the existence of the exercise, and that distinction is what keeps trust intact.
Who should own security awareness training?
One named person with the authority to change processes, not only to report numbers. It usually sits with security or IT, with HR owning delivery logistics and leadership owning the visible endorsement that makes checking socially safe.
Sources
- SP 800-50 Rev. 1, Building a Cybersecurity and Privacy Learning Program, NIST
- Telling users to 'avoid clicking bad links' still isn't working, UK National Cyber Security Centre
- Phishing attacks: defending your organisation, UK National Cyber Security Centre
- A Phish Scale: Rating Human Phishing Message Detection Difficulty, NIST
- Exercise in a Box, UK National Cyber Security Centre
- CISA Tabletop Exercise Packages, CISA
- Avoiding Social Engineering and Phishing Attacks, CISA
- Phishing, Technique T1566 - Enterprise, MITRE ATT&CK
- Cyber Security Toolkit for Boards, UK National Cyber Security Centre
- Cybersecurity Framework, NIST


