Evolving wetter.com's website and apps — a high-traffic product carrying 30+ features and a lot of legacy — toward a calmer, research-led experience that surfaces the right feature at the right moment without adding noise.
wetter.com's website and two mobile apps carry more than 30 genuinely useful features — pollen forecast, sunrise & sunset, vacation weather, garden and grill weather, UV index and many more — but most people never find them. At the same time, the screens risked overwhelming users with data they didn't ask for.
Because these are high-traffic products with millions of users, nothing ships as a big-bang redesign — a sudden change invites complaints, so we evolve step by step, measuring as we go, building every new page in the new design system (and modernizing the tech underneath: Material 2 → Material 3, the site moving to React). My focus: make the product calmer and the right features discoverable at the right moment. Parts have shipped incrementally. The home-screen redesign shown here is a personal concept I’m exploring — it’s still evolving, and the shipped version will differ.
Several problems at once — powerful features nobody could find, and screens that already felt overwhelming to read.
30+ useful tools sat buried in menus — and you can't promote them all at once (too many, and many are seasonal: UV and ski in summer, pollen in spring). Meanwhile the screens people did see felt cluttered: dense data they hadn't asked for, bright unremovable ads competing with content, and accessibility gaps that made the noise harder still. Discoverability, overload, visual noise and accessibility were all part of the same problem.
Before redesigning anything structural, I ran a series of deep, open-ended interviews across ages, life stages and geographies — exploratory research that reveals how people really behave, not what they think they should say. No rigid script: a warm-up, then concrete, experience-based questions to get past polished answers to real behaviour. I used AI to help synthesise the transcripts — clustering recurring patterns and surfacing themes faster — then validated every finding myself.
The core insight: people don't want weather for its own sake — they want it to plan. Two scenarios dominate — a frequent daily check, and an occasional, more deliberate one for a weekend or trip. Everything else (wind, humidity) is secondary: available, but never in the way.
What to wear, getting the kids to school — it comes down to temperature and whether it'll rain. Often checked on a watch, widget or smart speaker.
Comparing several days ahead for a getaway or event — people weigh a range of days, not a single moment.
“I never open the daily view — a huge, endless table, like a portal to hell. Air pressure in hPa? Who even looks at that?”
“If I see a little sun, that’s it — I don’t even read the temperature. A cloud and I’m already down; rain and I just know I need to prepare.”
“I love ‘feels like’ — that feels like the real temperature, not the number the sensors show in the shade.”
“When we head out I always check the wind — without a ‘feels-like’ I have to. If it’s windy she needs a warmer layer, and her hands freeze, so on go the mittens.”
“I wake up and the widget already shows what the day holds — I don’t open anything. But planning a trip, I compare every site, tab by tab.”
“The ideal? Look through the window and see not a number but how it feels — ‘cozy’, or ‘perfect for a run’.”
The redesigned home leads with current and today's weather — high/low and feels-like temperature. Feels-like matters more than it looks: one father dressing his three-year-old was reading wind and humidity himself to decide on mittens; feels-like folds all of that into a single perceived temperature, so nobody has to hunt. Alongside it, a plain-language description that's warm and anticipatory — "colder by noon, grab a layer" — caring, never pushy (no directive pressure), and refreshed through the day as conditions change. The friendly text is generated with AI, in the user's context.

Under the main weather sit small info cards — rain and warnings — readable without reading: the rain card glows blue when rain is coming and stays calm when it isn't; the warnings card shows a warning color or stays green for all-clear. They're tappable and open the right product — rain into the radar, warnings into the warnings map. Seasonal, contextual cards join when relevant — a UV-index card showing the current level with a plain label, a pollen card flagging an elevated count — always unobtrusive. Each one serves the user's goal and quietly drives discovery of a feature at the same time.

Rather than one overwhelming feature dump, I designed a reusable section — "Heute relevant in [city]" (with a new badge, because "Features" means nothing to a user) — that leads with three large promoted tiles (seasonal and contextual) above the full grid of all ~30 features, which fades after the first row behind a "show all" control, plus feature search on mobile. Promoted features also remain in the full list, and positions never shift — so people can always re-find what they saw (a principle I share with our CPO). Beyond that, bigger feed-style feature cards can be dropped onto any page where they're contextually relevant, so promotion follows relevance instead of living in one fixed spot — and in the new concept, people can customize which features they see (some, like warnings, stay fixed).



The icon was a flat raster image — a gradient sun with rays. I rebuilt it fully in vector and, when iOS introduced Liquid Glass icons, remade it as layers in Apple's Icon Composer (and did the same for Android), handing the file to engineering. The mark has no single glyph — a rayed sun behind clouds, full of gradients — yet it had to hold up across every new mode: clear glass, tinted (any single user colour), colourless, and Dark Mode. Dark mode was the hardest — a sun and clouds read oddly at night — so after many iterations I moved it into sunset tones that keep the brand instantly recognisable. Modern, still unmistakable, and with no drop in App Store performance.


The interviews reframed the whole product. Nobody opens a weather app to admire data; they open it to make a decision — what to wear, whether to bike or drive, how to dress the kids, whether the weekend is worth a trip. That single shift decided what leads (temperature, rain, feels-like), what supports (glanceable warnings and seasonal cards), and what steps politely aside (wind, humidity, and the long tail of features) until the moment it's relevant.
"Plan my day." Temperature and rain, read in seconds — often from a watch, widget or smart speaker. This is what the home screen is built around.
"Plan my weekend or trip." A multi-day overview to compare days — surfaced when wanted, not forced onto everyone all the time.
Smaller, concrete improvements delivered incrementally in the new design system:
A section under today's weather showing next-2-hour precipitation as a bar chart, linking straight into the radar — "Regenradar Berlin", "Regenradar Rom" — surfacing what research showed people care about most: will it rain.
Rebuilt the article and daily weather-video pages (produced by the in-house Munich media team) on the new typography and color — calmer, more readable, on-system.
Replaced an outdated animated radar banner in the ad column with an on-brand newsletter promo — logo included so readers know it's our content, not an ad.
Shipped the initial feature-visibility section on desktop and mobile web, with feature search on mobile — the first step of the wider concept.
Gave ad blocks a distinct grey-blue background and kept the surrounding content calm and functional — so our content reads as trustworthy and visually distinct from unremovable advertising.
A high-traffic product moving, step by step, from noisy and buried to calm and discoverable.
A deliberately incremental approach for a product with millions of users and heavy legacy — ship small, measure, and migrate users gradually rather than forcing a big-bang redesign. The home-screen redesign is a research-validated concept, aligned with product and fellow designers, not yet in code.