UX/UI design that turns visitors into customers

Most projects cannot fund a study and most do not need one. What they need is to stop guessing about five or six specific things.

The cheap sources are usually already in the building. Support tickets name the screens people fail on. The sales team can recite the three questions every prospect asks. Search queries on your own site read as a list of things people could not find. Session recordings from a single week show where a form is abandoned. None of that is a controlled experiment. All of it beats an opinion.

Five conversations with real users surface most of what matters in an interface, provided the questions are about what happened rather than what someone would prefer. Ask about the last time they tried to do the thing. Preferences are easy to answer and rarely predict behaviour.

A design review shows the screen full of the right data, on a fast connection, with a name that fits its column. Then it ships and meets the world.

• the empty state, before any data exists
• the loading state on a slow connection
• the error when a payment provider times out
• a list of two items, and a list of ten thousand
• a product name in German that runs twice as long
• a user with permission to see half the page

These are not edge cases in the statistical sense. They are most of the first hour a new customer spends with a product, and a missing empty state is why an onboarding looks broken on day one. We draw them alongside the main screens rather than after, because the alternative is a developer inventing them at the end of a sprint, alone, at speed.

A design file is a set of pictures. Code is a set of rules. Most handoff friction is the distance between those two.

What a build needs beyond the artboards: how each element behaves between the two widths that were drawn, which text may wrap and which must truncate, what is fixed and what stretches, the hover and focus and disabled versions of every control, and the tap target on a phone. Then the dull ones. What the layout does when the browser font size is doubled. Whether a component appearing in two places is one component or two that happen to look alike.

The cheapest fix is a short review together while the first screens are being built. It takes an hour. It prevents the version of the project where design and development each did what they were asked and the result matches neither.

Most interface problems are language problems in a visual costume. A button nobody presses is rarely too small; it usually promises something the person did not want, or conceals what happens after. An empty state that feels broken is a paragraph written by whoever was nearest the deadline.

So the words come first. In a document, before the layout. Every screen has a job that fits in one sentence: what the person came to do, what they need to know in order to do it, what happens next. If that sentence cannot be written, the screen is not ready to be drawn. Drawing it only hides the gap under something attractive.

The text written first is the text nobody demos. Error messages that name the field and the fix rather than announcing that something went wrong. A confirmation that says what was charged and when it arrives. The day-one empty state, where there is no data and the screen has to teach instead of display. The wording of a destructive action, where delete and delete permanently are two different promises. These are the moments people remember, and they are routinely written last, at speed, by whoever has the file open.

Working this way also shortens the argument about visuals. Once the words are agreed the layout has a job — make this readable in this order — and taste occupies a smaller share of the conversation. Copy written to fit a box that already exists gets trimmed until it stops being true. A box built around a sentence that has to be said tends to fit. That is the whole trick.

It also changes who sits in the review. A screen described in sentences can be checked by the people who know whether its promise is true — support, legal, whoever owns the number on display — before the pixels make disagreement expensive. Send them a picture instead and you get comments on the colour, which is the one thing they were not needed for.