Good design is not decoration. It is how a company saves time and money.

When the word design comes up in a company, most people picture something visual. Colours, a logo, a modern looking login screen. Design becomes the last line item on the project, the thing you add once the functionality is finished and the budget is almost spent.

At the same time, that same company is paying for something else. For the minutes people spend every day hunting for a value in an internal system. For fixing an order somebody entered incorrectly because the form allowed it. For two weeks of onboarding instead of two days. These are design costs, they simply appear in the accounts as salaries, support and error rates.

In digital products and internal systems, design is not the visual layer at the end. It determines how long people need for a task, how many mistakes they make and how much training they require. All three can be converted into money.

Why design still gets mistaken for aesthetics

Aesthetics is the only part of design that shows up in a screenshot. Good design, meanwhile, reveals itself as the absence of a problem, and nobody gets praised for a mistake that never happened. On top of that, design usually sits under marketing on the org chart, in other words under the department responsible for how things look.

The main reason, though, is a different one: most companies never measure usability, so there is no figure to set against the cost of improving it. ISO 9241-11 defines usability as the extent to which specified users can achieve specified goals with effectiveness, efficiency and satisfaction in a specified context of use. Maximilian Speicher's technical report points to a detail that tends to get overlooked: the word specified appears three times in that definition. Usability is not a property of software, it is the result of software, a person and a task meeting. A supplier who has never watched your people work cannot assess it.

According to MeasuringU's review, efficiency in that standard is measured mainly by how long a task takes to complete, in other words by time-on-task. Usability is measured in time, not in aesthetics.

Five different things that are all called design

A designer and a managing director will often cover five separate disciplines in a single sentence. The difference between them is the difference between cosmetics and economics.

• Visual design covers how it looks: typography, colour, contrast and hierarchy, in other words what the eye notices first.
• Interaction design covers what happens when somebody does something: button behaviour, states, feedback, the ability to undo a step.
• Information architecture covers how information is organised and labelled so that people find it even when they have no idea where to look.
• UX design covers the whole experience of completing a task, from first contact to outcome, including what happens away from the screen.
• Service design covers how the entire service fits together across channels and departments, including the work of the person on the other side.

A company that buys only the first item gets a prettier wrapper around an equally expensive process.

Every point of friction has a price

Friction is any step that costs time or attention and delivers no value. In practice we see six kinds: unnecessary clicks, re-entering the same data, searching for information, errors and the work of correcting them, support requests and onboarding time.

Nielsen Norman Group measured exactly this on corporate intranets. Across three waves of research they tested 42 intranets, and in the last of them the average task completion rate was 74 per cent, effectively the same as ten years earlier. Finding information about a department head took an average of 2 minutes and 19 seconds in the first wave and 2 minutes and 46 seconds in the third. On its own, 27 seconds is negligible. Multiplied by the number of people and the number of searches per year, it is not.

Cognitive load, in Nielsen Norman Group's terms, is the amount of mental resource needed to operate a system. Part of it belongs to the task and cannot be removed, part is added by the interface itself: inconsistent labels, styling with no meaning, information nobody needs. Once the load exceeds capacity, performance drops: slower comprehension, a missed detail, an abandoned task.

Error prevention is Nielsen's fifth heuristic: an error message is useful, but the best design stops the error from happening in the first place. Slips caused by inattention are countered by constraining the input and setting a sensible default, mistakes caused by a wrong mental model are countered by clear labelling.

Learnability describes how quickly somebody completes a task on first contact with an interface and how many repetitions they need before they are efficient. Nielsen Norman Group measures it by tracking time-on-task across five to ten attempts and plotting the learning curve. The slope of that curve determines the cost of training.

Cost of Friction: a model you can work out yourself

For friction to be something you can make decisions about, it has to carry a price. That is why we use the Cost of Friction model:

time × frequency × number of people × cost

For an internal system that means: number of people × minutes lost per day × working days × hourly cost of labour. A worked example, meaning a hypothetical company with estimated inputs rather than measured data: 40 people in operations, each losing 12 minutes a day retyping data between two systems, 250 working days, a fully loaded labour cost of 25 euros an hour.

40 × 12 minutes is 480 minutes a day, or 8 hours. Eight hours times 250 days is 2,000 hours a year, times 25 euros is 50,000 euros a year. For one single point of friction. The equivalent of one full-time employee who spends the entire year doing nothing but working around a bad interface. And if observation shows the loss is not 12 minutes but 4, the figure comes out at roughly 16,700 euros a year. That is still a sum worth tracing back to its cause.

Nielsen Norman Group runs the same calculation on a larger scale, using an hourly cost of 30 dollars. For a company with 10,000 employees that works out at savings of roughly 4 million dollars a year for moving from the worst quartile of intranets to the average, and roughly 2.4 million for moving from the average into the best quartile. These are estimates rather than invoices, but the mechanism holds just as well with twenty people.

A bad internal system versus a well designed workflow

The strongest evidence comes from healthcare. A simulation study published in the Journal of the American Medical Informatics Association tested emergency department physicians at four sites running software from only two vendors. On some tasks there was a ninefold difference between sites in time to completion and an eightfold difference in the number of clicks. Error rates across tasks and sites ranged from zero to 50 per cent.

The same software. A different rollout, a different configuration, a different workflow. Productivity is therefore not decided by the licence but by how the work around it is designed. And this is not cosmetic: in an analysis of roughly 9,000 safety event reports from three paediatric sites, summarised by the AHRQ, 5,079 of them involved the system and medication, and in 3,243 of those events usability was a contributing factor.

In an ordinary company it looks like this. A bad internal system mirrors the structure of the database: one screen per table, a form with dozens of fields, half of them mandatory, and the only way to learn the status of a process is to ask somebody. A well designed workflow mirrors the flow of the work: it asks only for what the person knows at that step, pre-fills the rest, shows the current status and the next step, and refuses any combination that somebody would later have to correct by hand.

The difference is not visible in how it looks. It is visible in time-on-task, in the error rate and in whether people work around the system using a spreadsheet on the side. What that leakage costs is something we cover in the piece on why the most expensive system is the unused one.

Why an interface redesign is not enough when the process is wrong

If the process is bad, an interface redesign turns it into a prettier bad process. Eight unnecessary steps with good typography are still eight unnecessary steps. That is why we work in this order: eliminate, simplify, automate, and only at the very end extend with decision support tools. Automating a step that should have been removed is the most expensive way to set it in concrete.

The same holds outside internal systems. Baymard Institute reports that the average number of checkout fields fell from 12.7 in 2019 to 11.3 in 2024, while eight would be enough for most online shops. Their important finding is that the number of fields affects checkout usability more strongly than the number of steps. What decides the outcome is how much a person has to process, not how many screens you spread it across. According to Baymard, 17 per cent of users have abandoned a checkout at some point because it was too complicated, which is precisely what happens when you have a website visitors do not understand.

Which leads to something we tell clients before they sign anything. If the friction can be removed by dropping a step, changing a configuration or reaching an agreement between departments, there is no sense in paying for custom development. That only pays off where the process is genuinely specific. Which is why you have to look at the whole: a company is one product and every bad process is its technical debt.

The order in which we do design

Process, user, information, interface, visuals. This order is not an aesthetic preference, it is sorted by the cost of getting it wrong. A mistake in the process is the most expensive, a mistake in the colour of a button the cheapest.

1. Process. First we establish what actually happens, who does what, why and in what order. We do not design a step that should not exist.
2. User. Then we establish who handles the task, what they already know and what distracts them while they work. Without that, usability cannot even be defined.
3. Information. Then we organise and label the content according to the company's own vocabulary and the order in which decisions are made. A large share of UI problems are really naming problems.
4. Interface. Only then do we design screens, states, error messages and error prevention.
5. Visuals. Finally typography, colour and brand, whose job is to make the hierarchy readable and reduce cognitive load, not add to it.

One note on cost, so that none of this reads as a sales pitch for research. Quantitative measurement is expensive: Nielsen Norman Group recommends around 20 users per design, and a study of that kind costs roughly four times as much as a qualitative one. Most companies only need to measure their one or two most expensive tasks and handle the rest by observation. Be careful with return figures as well: an older Nielsen Norman Group analysis of 42 redesigns reported an average improvement in metrics of 135 per cent, while a newer analysis from 2008 put it at 83 per cent. Anybody promising 135 per cent today is quoting old data.

Where to look for UX debt in your company

UX debt is accumulated friction that nobody has ever booked. These questions bring it to the surface without an audit.

• Which task recurs most often and how many people perform it? Calculate the Cost of Friction there first, not wherever the loudest complaint is coming from.
• Where does the same piece of data get entered a second time? Every repeated entry is lost time and an opportunity for error.
• Which spreadsheet lives alongside the system that was meant to replace it? A shadow tool is a map of what the system is missing.
• What generates the most support requests? A recurring question is an unpaid invoice for design.
• How many days does it take before a new joiner works unsupervised, and which step slows them down most?
• Which error gets corrected by hand at least once a week, and what in the interface made it possible?

If nobody can answer them, that is a finding in itself: the cost exists and it is nowhere to be seen.

Key takeaways

• Pick the task the largest number of people perform every day and calculate the Cost of Friction: number of people × minutes lost per day × working days × hourly cost. One figure moves the design conversation from a question of taste to a question of budget.
• Measure three things instead of relying on impressions: time-on-task, the error rate and the number of repetitions before a new joiner works unsupervised. A stopwatch and observing five people is enough to start with.
• Eliminate first, then simplify, and only then automate. An automated unnecessary step is the most expensive kind of unnecessary step.
• If the problem can be solved by changing a configuration, removing a field or reaching an agreement between departments, do not buy custom development.
• Before you commission a redesign, ask to be shown what will change in the numbers, not on the screen. If the supplier cannot say which metric they intend to improve, they have not yet diagnosed the problem.