Usability: what it is, Nielsen's 10 heuristics and how to run a usability test
What usability is, how it differs from UX and interface, Nielsen's 10 heuristics, how to run a usability test step by step, where AI helps at each stage and which metrics to track.
Published on November 16, 2026. 12 min read
Usability is how easily a person can use a website, an app or a system to do what they came to do. When it is good, almost nobody notices: the sign-up finishes, the order goes through, the report is found. When it is bad, it shows up as abandoned carts, support tickets, rework and users who never come back.
This guide explains what usability is, which principles help you evaluate a screen, how to run a usability test step by step, how AI fits into each stage of that test and which metrics to track once the product is live. There is an FAQ at the end.
What is usability?
The ISO 9241-11 standard defines usability as the extent to which a product can be used by specified users to achieve specified goals with effectiveness, efficiency and satisfaction in a specified context of use.
In practice:
- Effectiveness: can the person complete the task? How many give up halfway?
- Efficiency: how much effort does it take? How many clicks, how much time, how many errors along the way?
- Satisfaction: how do they feel afterwards? Would they use it again or recommend it?
The definition also points to two things people often forget: who uses the product and in what context. A management system used eight hours a day by a trained team has different needs from an app a customer opens once a month, on a phone, in a hurry.
What is the difference between usability, UX and interface?
The three terms go together, but they are not the same.
| Term | What it covers | Question it answers |
|---|---|---|
| Interface (UI) | What the person sees and touches: screens, buttons, text, icons | How does the product present itself? |
| Usability | How easy it is to complete a task in that interface | Can the person do what they need without unnecessary effort? |
| User experience (UX) | Everything the person goes through with the product, before, during and after use | Does the product solve their problem, and is it worth coming back? |
A good-looking interface can have poor usability, and a product with good usability can still deliver a weak experience if it solves the wrong problem. To go deeper, see the guides on what an interface is and on the user journey.
Why does usability matter to the business?
Because usability problems cost money even when nobody complains. Common effects:
- Lower conversion. A confusing form, a hidden button or shipping costs that only appear at the end make people give up.
- More support. Every question the screen does not answer becomes a ticket, a call or a message.
- Operational errors. In internal systems, a poorly explained field produces wrong data that someone will have to fix later.
- Long onboarding. The less intuitive the system, the longer each new person takes to become productive.
- Product abandonment. With apps, people simply uninstall. Speed counts here too, as we cover in the post on loading and retention.
What are the principles of usability?
The most widely used set for evaluating an interface is Jakob Nielsen's 10 usability heuristics, published by the Nielsen Norman Group. They are general rules, useful for reviewing a screen before testing with users:
- Visibility of system status. The person knows what is going on: loading, saved, sent, failed.
- Match between the system and the real world. The screen uses the users' words, not the company's internal jargon.
- User control and freedom. You can go back, cancel and undo without losing everything.
- Consistency and standards. The same button does the same thing on every screen, and the product follows the conventions people already know from other apps.
- Error prevention. The best error is the one that never happens: constraints, confirmation before irreversible actions, good defaults.
- Recognition rather than recall. Options and information stay visible; nobody has to memorize what they saw on the previous screen.
- Flexibility and efficiency of use. Shortcuts for frequent users that do not get in the way of beginners.
- Aesthetic and minimalist design. Every extra element on the screen competes with what matters.
- Help users recognize, diagnose and recover from errors. Error messages in plain language, saying what happened and how to fix it.
- Help and documentation. When explanation is needed, help appears in context, with concrete steps.
Accessibility goes hand in hand. A screen that does not work with a screen reader, has low contrast or uses tiny fonts has poor usability for part of the audience. The WCAG guidelines are the most common reference.
Examples of usability problems
Some problems show up in almost every product that has never been tested:
- A form that clears everything when one field has an error.
- A primary button with the same visual weight as three secondary buttons.
- A menu with labels that only make sense to people who work at the company.
- A search that cannot find the product if the word has an accent or a typo.
- A sign-up step that asks for data the product does not need yet.
- An internal system screen with twenty fields, of which the team uses five.
- An "Error 500" or "Operation failed" message that does not say what to do.
What is a usability test?
A usability test means watching real people from the product's audience try to complete real tasks. Nobody asks "did you like it?": the participant gets a task ("find the finance course and enroll") and the facilitator watches where they hesitate, make mistakes or give up.
The most common formats:
- Moderated: a facilitator runs the session live, in person or over video, and can ask questions.
- Unmoderated: the participant does the tasks alone, with screen recording, in a remote testing tool.
- Heuristic evaluation: specialists review the interface against the heuristics, without users. It is fast and good for clearing obvious problems before testing with people.
- A/B test: two versions go live to different groups and the data shows which works better. It suits products that are already live and have traffic.
On the number of participants, the Nielsen Norman Group shows that tests with about 5 users already reveal most problems for one audience profile, and recommends running several small tests with fixes in between rather than one large test. When the product has very different audiences (for example, the buyer and the approver), test each group.
How to run a usability test, step by step, and where AI fits in
The test is a cycle: plan, test, fix and test again. AI helps with the repetitive work at every stage, and decisions stay with the people who know the product and its users.
- Set the goal and tasks: AI suggests tasks based on usage data; the team picks the ones that matter to the business.
- Recruit participants: AI helps build the profile screener; the team confirms they belong to the real audience.
- Run the sessions: Automatic transcription and moment tagging; the facilitator observes and asks questions.
- Analyze and group: AI groups similar quotes and errors; the team reviews the videos and sets severity.
- Prioritize and fix: Problems ranked by impact; product and design decide what goes first.
- Measure and retest: Agents track usage metrics; a new test confirms the fix worked.
AI agents handle the repetitive work at each stage; people decide what to test, what to fix and when it is good enough.
1. Set the goal and the tasks
Start with a business question: "why do so many people abandon sign-up?", "can the team enter an order without help?". Then write 3 to 6 tasks that represent what users really do.
Where AI fits in: an agent can read product usage data, support tickets and app store reviews and point out where people get stuck most. That helps you choose tasks based on what actually happens instead of a meeting-room guess. The product team decides the focus of the test.
2. Recruit participants
Participants need to come from the product's audience. Testing a management system with colleagues from the development team almost always hides the problems.
Where AI fits in: AI helps draft the screening questionnaire and filter responses against the defined profile. A person confirms that each participant really belongs to the audience.
3. Run the sessions
Ask participants to think aloud, do not help when they get stuck and record what they do, not only what they say. Record screen and audio, with consent.
Where AI fits in: automatic transcription, tagging of moments where the person hesitated or made a mistake and a summary per session. The facilitator is free to observe and ask the right questions instead of taking notes on everything.
4. Analyze and group the findings
Bring together what repeated across participants. A problem that came up with three out of five people is serious; one that came up with a single person may be specific to them.
Where AI fits in: an agent groups similar transcript excerpts and errors and proposes a list of problems with the evidence for each. The team reviews it against the videos, merges or splits items and sets severity. Automatic grouping gets it wrong when two similar comments have different causes, so human review is mandatory.
5. Prioritize and fix
Rank problems by impact (how much they affect the main task and how many people they affect) and by effort to fix. Fix first whatever prevents people from completing the task.
Where AI fits in: AI helps rewrite button labels and error messages, generate screen variations for discussion and, in development, speed up fixes with agents reviewed by the team. Product and design decide what goes in now and what waits.
6. Measure and test again
After the fix, test again with different people from the same profile and track the live product's numbers. It is common for a fix to solve one problem and reveal another that was hidden behind it.
Where AI fits in: agents watch the funnels and raise an alert when a step gets worse after a release. We cover this in metrics from day zero.
Can AI replace testing with users?
No. AI tools can run a quick heuristic review of a screen, flag contrast and accessibility issues and simulate navigation paths. That is useful as a first pass, before recruiting participants. But AI does not know what your users expect, which words they use at work, how rushed they are when they open the app or what they tried before getting there. That only shows up when you watch real people.
What works is combining both: AI cuts the time spent on preparation, transcription and analysis, and testing with people remains the source of truth.
Which usability metrics should you track?
A few simple metrics cover the three ISO dimensions:
| Metric | What it shows | Dimension |
|---|---|---|
| Task completion rate | How many people manage to finish | Effectiveness |
| Time on task | How much effort the task takes | Efficiency |
| Errors per task | Where the interface confuses people | Effectiveness and efficiency |
| SUS (System Usability Scale) | A 10-question survey on perceived ease of use | Satisfaction |
| Drop-off by funnel step | Where people give up in the live product | Effectiveness |
| Support tickets per feature | Where the screen does not explain itself | Efficiency |
The first four come from testing. The last two come from the product in use and show whether the improvement holds.
Usability in practice: UEEK projects
On Itaú Mulher Empreendedora, the platform had hidden content that was hard to navigate, and the design did not follow a current standard, which hurt usability and engagement. The new site got categories, filters, search and structured pages, with an interface designed for women entrepreneurs and following the WCAG guidelines. With a more intuitive journey, access to the program's initiatives became smoother.
On the launch site for Lexus's first fully electric vehicle in Portugal, we ran usability tests with personas and wireframes to validate the flows before the final design. The goal was to highlight the electric model without creating noise around the hybrid line.
On the Bom Controle app, we started with prototyped wireframes to validate interactions and flows before the visual concept, and each module was validated by the client's team before moving on.
In all three cases, validation came before code. Fixing a flow in a prototype is much cheaper than fixing it after it has been built.
When to get help with usability
Some signs that an outside review is worth it:
- Users give up halfway through an important flow and nobody knows why.
- Support keeps answering the same questions about the same screen.
- The product grew without a visual standard and every screen looks like a different system.
- Interface decisions turn into debates about taste, without data.
UEEK's Design and UX comes in two formats: a UX assessment with problems ranked by impact, or a design team working alongside your product, with decisions backed by usage data. The work is done by hybrid squads, where AI agents handle the repetitive parts (transcription, grouping, metric tracking) and specialists run the tests and decide what changes. That is 14 years and more than 500 projects delivered for more than 200 clients in 6 countries.
Frequently asked questions
What is usability in a few words?
It is how easily a person can use a product to do what they need, with effectiveness, efficiency and satisfaction.
What is the difference between usability and UX?
Usability measures whether a person can complete a task without unnecessary effort. UX is broader: it includes everything they go through with the product, including whether it solves the right problem.
What are Nielsen's 10 heuristics?
Visibility of system status, match between the system and the real world, user control and freedom, consistency and standards, error prevention, recognition rather than recall, flexibility and efficiency of use, aesthetic and minimalist design, helping users recognize, diagnose and recover from errors, and help and documentation.
How many people do you need for a usability test?
About 5 participants from the same profile already reveal most of the problems, according to the Nielsen Norman Group. Several small tests with fixes in between work better than one large test.
How does AI help with usability testing?
In preparation (choosing tasks from usage data), in the sessions (transcription and moment tagging), in analysis (grouping similar findings) and in follow-up (alerting when a funnel gets worse). Choosing what to test and deciding what to fix stay with people.
Does AI replace testing with real users?
No. It works as a first pass and speeds up the work, but your users' behavior, vocabulary and context only show up when you watch real people.
When should you run a usability test?
Before development, with a prototype, to validate the main flows. After launch, whenever a funnel metric gets worse or an important new feature goes live.
Does usability matter for internal systems too?
Yes, and the gain is often larger: the team uses the system all day, so every click saved and every error avoided repeats hundreds of times a month.
Topics
- usability
- usability testing
- Nielsen heuristics
- UX
- interface design