An app people opened once a month to check their bill.

• Walk-in retail sales down about 30%, while total revenue kept growing.
• You couldn't buy a phone in the app. Only on the website or in a store.
• Germany, Portugal and every other market looked and behaved differently.

A shop for connected-home and work-from-home tech.

• Vodafone's own devices beyond phones: GigaCube, TV, Wi-Fi extenders, smart plugs.
• Partner products from Samsung, Siemens and Philips, sold through Vodafone.
• Bundled with data, like a GPS bike light with its own SIM.

The next screens were built on Source.

Source was Vodafone's design system from 2016 to 2024. Judge the decisions on the next few slides.

For visual craft, judge Unity, the system that replaced it.

Should Vodafone and partner products feel like one shop?

The brief asked for a seamless marketplace. The team leaned towards one merged catalogue. My worry: when a partner's order is late, people blame Vodafone.

I argued against upsell. Testing proved me wrong.

After adding to basket, the team wanted a You might also like sheet. I thought our catalogue was too small and it would feel pushy.

In a rapid test, people liked it and didn't find it salesy. We shipped it with one rule: the sheet only appears when there's something genuinely related.

We offered flexibility nobody wanted.

Workshops gave every item its own delivery method: home, one of ~400 stores, or a pickup point.

They expected one choice for the whole order. Some picked options for the wrong item.

Designing for failures we don't own.

Partners kept stock and support details in our CMS. When they couldn't fulfil an order, the fault was theirs, but it happened inside our app.

1. What went wrong, in plain words
2. Who is responsible for fixing it
3. What happens next, and when

Giving people a reason to come back.

Rewards existed, but fewer than 10% engaged with it. I proposed small games with different rhythms.

• Instant: tap to reveal, spin, dice, scratch card
• Memory: match the cards
• Streak: collect a letter a day to spell a prize

You can do it right and still not get the ending you planned.

The late-2023 downturn hit. Partners paused, inflation cut demand, and Vodafone sold its Italian and Spanish businesses. Marketplace launched only where demand held.

Ten markets, ten design systems.

Germany had Brix, the UK and Ireland had Source, Spain had Resource, Hungary had Aurora. Patterns drifted, and every team rebuilt the same button.

BrixSourceResourceAurora+6 more Unity

One system for every market. Hence the name.

My part: components, icons, illustrations, and the accessibility process towards the 2025 EAA.

"50% off", and still more than I paid three months ago.

During Covid, prices rose far faster than the discounts. And the inbox didn't help.

People arrived mid-purchase. If the first track failed, they deleted the app.

They weren't exploring. They wanted a fast answer to one question: did it work?

Don't ask people to set up a scraper.

Websites change constantly, so rigid scrapers break. We built change detection instead: point at what matters, get told when it changes.

• Reliable over precise. Keep working without babysitting.
• Any meaningful change. Price, stock, wording.
• Fewer decisions. No rules or logic to configure.

Inside, a change tracker. Outside, a price tracker, because that's who would pay.

I tested demand. Shervin tested feasibility.

Before building properly, a landing page asked one question: can we explain this well enough for someone to leave their email?

A lifetime price for early sign-ups: a thank-you, and a reason to commit.

Any web page to a tracked item, in a few taps.

Paste a link or share from the browser, tap the price on the live page, and the app watches it from then on.

No marketing team, so I ran the experiments.

I kept testing beyond the screens: how the store page sells the app, and when to ask people to pay.

No ad budget. People did the talking.

• Reddit first. Posts about the problem, and the design, travelled.
• A public feedback board, Discord and Telegram. Users watched their ideas ship.
• Viral in Argentina, then Türkiye, where high inflation made it essential.

Android requests arrived every day. We said not yet.

The hard part wasn't the UI. It was tracking reliably across thousands of websites. Two platforms would have halved our speed on the thing that mattered.

Not everyone wants a subscription.

People kept asking to pay per item. Alongside the plans, we added pay-as-you-go tracking credits.

Also after launch: Normal, Extended and Ultra tracking for tricky pages, and follower tracking that brought in creators.

Revenue lags. We watched what comes before it.

1. Time to first tracker
2. Return visits to check prices
3. Alerts that mattered, a real drop on something wanted

Only 3% finished onboarding. It "worked", so nobody wanted to touch it.

Research showed people skimming, and a big drop at the permissions step. We made it visual, shorter, asked for permissions in context, and added a reward at the end.