Security Technology People Can Actually Understand
A verdict nobody can act on has failed, however good the analysis behind it. The design rules that decide whether a non-specialist can use a security tool.

Here are two descriptions of the same email.
The first: SPF soft fail with DKIM d= misalignment, sender domain registered within 48 hours, URL reputation unknown, display name mismatch on From header, nine correlated recipients in tenant.
The second: the sender's domain was registered two days ago, the link opens a copy of the Microsoft sign-in page, and the same email reached eleven other inboxes at your company this morning.
Identical findings. One of them causes somebody to stop and report it. The other causes somebody to forward it to IT and wait, which is a different and slower outcome, or to close the tab.
The gap between those two paragraphs is where most security technology fails, and it is not a presentation problem. It is a design question about who the tool thinks it is talking to.
The assumption buried in most security output
Security tooling grew up inside security teams, and it inherited the audience it was built for. The output of a traditional product is a measurement handed to an analyst who will interpret it: a severity, a score, a protocol result, a reputation verdict. The interpretation step is real work, and in that setting it is somebody's job.
Now move the same output to a small organization. The person receiving it is an operations manager who also handles IT, or a practice administrator, or the one person in a thirty-person firm who is slightly better with computers than everyone else. The interpretation step has not disappeared. It has simply been assigned to somebody who cannot perform it.
The result looks like disengagement and gets described as an awareness problem. It is not. It is a tool producing something its reader cannot convert into a decision.
NIST maintains a whole program on human-centered cybersecurity on the premise that this is a technical discipline rather than a matter of tone, and the same premise drives the secure design principles published by the NCSC and the Secure by Design work at CISA. Security that fights the person using it loses, and the clearest historical example is password policy: NIST's own digital identity guidance abandoned mandatory periodic password changes largely because the human response to them, predictable minor variations, made accounts less secure rather than more.
Six rules we design against
1. End with an action, not a reading
The output of a check should be the next step. Do not click this. Call the supplier on the number you already had. Report this to your security team. Change this password from a device you trust.
A risk score alone is a reading. It tells somebody how worried to be without telling them what to do, which is the exact state in which people do nothing. Our scan results carry a score, a short summary of what the message is and why, the specific warning signs, and then what to do next, in that order, because the order matters when somebody is reading it in a corridor.
2. Name the thing, not the category
"Impersonation detected" is a category. "This is not docusign.com: there is a zero in place of the o" is the thing.
Categories are efficient for people who already know the taxonomy and useless for everyone else. The specific version does two jobs: it convinces, because the reader can verify it with their own eyes, and it teaches, because the next time they see a zero in a domain they will catch it themselves.
This constraint also has a disciplining effect on the analysis. If a finding cannot be stated as a concrete fact about the thing in front of the user, it is usually a signal rather than a conclusion, and signals should stay in the engine until they add up to something worth saying.
3. Give three reasons, not thirty
An investigation can produce a long list: authentication results, delivery route, link destinations, attachment hashes, call-back numbers, how many other people received it. All of that is worth keeping and worth showing to someone who asks.
What reaches the person deciding should be the strongest few. Three is roughly the number someone reads under pressure. The rest belongs one layer down, available to whoever wants it, which is a different audience even inside the same organization.
4. Say when you are not sure, in words
This is the rule with the highest cost and the clearest payoff.
A tool that produces confident verdicts is more satisfying to use and more dangerous, because the first time it is confidently wrong it teaches somebody that checking is pointless. A verdict that states what it could not determine keeps the person inside the decision, where they belong, particularly when money is about to move.
The equivalent at organization scale is refusing to invent numbers. Our security score is weighted across measured areas, and any area with no data behind it is excluded rather than filled in with a plausible value, with the gap stated in plain words: nobody has been sent a simulation yet, so this cannot be measured. Less impressive on a first viewing. Considerably more useful in a meeting where somebody asks what the number is based on.
5. Make the safe path the easy path
Every design decision sets a default, and the default wins.
If reporting a suspicious email means composing a forward to a shared mailbox and attaching the original with headers intact, the realistic default is deletion. If it is one step from the inbox and the email disappears from the person's view immediately, the default becomes reporting. Nothing about the person changed.
The NCSC's observation that telling users to avoid clicking bad links still is not working lands here. Advice is a weak lever. Changing which action is easiest is a strong one.
6. Do not make starting a project
Usability includes the first hour. A measurement that requires mail-server configuration, directory integration and a change approval never happens in an organization without an IT team, which means the organizations with the least visibility stay that way.
Removing that gate was a deliberate design choice in our console: no mail-system integration, invite people by email, group them into departments, run a first campaign the same afternoon. That is less a feature than the removal of a reason to give up.
What this does not mean
Writing for non-specialists is sometimes read as dumbing down, so it is worth being precise about what stays untouched.
The depth of the analysis is not the thing being simplified. Behind a three-line verdict about a reported email there can be exact, code-level forensics: authentication and alignment, reply-to and display-name tricks, whether a visible link matches its real destination, redirects unwrapped, attachment types and hashes, call-back numbers, how many other people in the organization received the same message, and threat intelligence lookups that record which sources answered and which could not. None of that gets removed. It gets summarized by whatever subset is load-bearing, and the rest stays available.
Nor does it mean avoiding all technical vocabulary. It means introducing a term once, in use, where the reader needs it, and never assuming a term carries meaning because it is standard inside the field. Multi-factor authentication can be explained in a clause. "SPF soft fail" cannot be explained in a clause, which is a sign it should not be the headline.
A test you can run on any security tool. Show its output to someone in your organization who is not technical, and ask what they would do next. If they cannot answer, the tool has not finished its job.
How to tell whether you got it right
Usability claims are cheap, so the measures have to be behavioral.
- Does anyone use it when it matters? Not in a demo. On a Thursday, with eleven minutes until a meeting, about an invoice.
- Does the recommended action get taken? A tool whose advice is read and ignored has produced a reading, not a decision.
- Does reporting go up? If making a safe action easier does not increase how often people take it, the action was not actually made easier.
- Can a non-specialist explain the result to somebody else? This is the strongest signal, because it means the explanation transferred rather than just landing.
The underlying position is simple enough to state in one sentence, and it is the reason this gets treated as a design problem rather than a copywriting one: a verdict that a non-specialist cannot act on has failed, however sophisticated the analysis behind it.
For a worked example of the output side, see how to read a scan result. For the single highest-leverage default in most organizations, making reporting one step. For documentation that people might actually read, the one-page security policy, and for where this sits in a wider program, the small business cybersecurity guide. You can try the plain-language output yourself with CyberWatch AI Scan.
Frequently asked questions
What makes a security tool unusable?
Output that requires expertise the reader does not have. A finding expressed as a protocol result, a severity code or a reputation score leaves the interpretation to someone who was never there.
Is plain language just a presentation layer?
No. It constrains the analysis. If a finding cannot be stated in a sentence a non-specialist can act on, it usually means the tool has reported a measurement rather than reached a conclusion.
How many reasons should a verdict give?
Around three. Enough to be convincing and to teach something, few enough to be read in the ten seconds the person actually has.
Should a tool show uncertainty?
Yes, in words. A verdict should state what it could not determine, because a confidently wrong answer teaches people to stop checking at exactly the wrong moment.
Why does a single score help if it loses detail?
Because a non-specialist needs one handle on a complicated subject. The score is only honest if the evidence behind it is available underneath and any unmeasured part is reported as unmeasured.
What is the cost of a difficult setup?
Abandonment. An integration project is a fortnight for an organization with an IT team and the end of the evaluation for one without, so the measurement never happens at all.
Does designing for non-specialists weaken the analysis?
It should not. The analysis can be as deep as you like. What changes is that the conclusion, not the working, is what reaches the person who has to act.
How do you know a security tool is usable?
Measure whether people use it under pressure, and whether the action it recommends gets taken. Adoption and action are the only honest usability metrics.
Sources
- Human-Centered Cybersecurity, NIST Computer Security Resource Center
- Secure design principles, UK National Cyber Security Centre
- Secure by Design, US Cybersecurity and Infrastructure Security Agency
- NIST Special Publication 800-63B, Digital Identity Guidelines: Authentication and Lifecycle Management, NIST
- Telling users to 'avoid clicking bad links' still isn't working, UK National Cyber Security Centre
- Secure Our World, US Cybersecurity and Infrastructure Security Agency
- Phishing scams: how to spot and report them, UK National Cyber Security Centre
- Awareness and Cyber Hygiene, European Union Agency for Cybersecurity


