One brand, for every market

Context

In 2016, PayPal’s marketing site received over 200 million unique visitors every month. It is not one website so much as a publishing surface that dozens of regional marketing teams write into, in their own languages, for their own markets, against their own campaign calendars.

The problem was not that those teams lacked pages. It was that every page was a static, custom instance. A hero built in Australia had nothing to do with a hero built in Germany, and the only thing holding the brand together across them was diligence.

The team that took this on was cross-functional and small for the size of the surface: a chief product owner and technical product owner, front-end and product design, visual design, two UX roles, and a content strategy function led by a manager. I worked as the content strategist executing day-to-day component naming, definitions, and copy across that system. Before the build started, the whole team went through Agile Scrum training together, setting the cadence the work ran on. Company-wide, the all engineering teams were being aligned on the same two week sprint schedule.

The output was a system called Canvas: a component library, a publishing framework, and a set of guidelines that told marketers what to reach for and why.

Act 1: What a page is made of

Canvas treated a page as something assembled, not authored. The model was atomic design: atoms combine into organisms, organisms into templates, templates onto a publishing framework called MPP that carried analytics, tagging, responsive grid, and interaction standards underneath every page.

That structure only works if people agree on what the pieces are called. A component library is a naming problem before it is a design problem. So the library was organized by what a marketer was trying to do rather than by what the component was made of. Twelve categories: Hero, Compare, CTAs and links, Featured merchants, Features and benefits, Fees and pricing, Offers, Page elements, Social & testimonial, Steps, Video, and Footers. Each one opened onto named organisms with their content slots spelled out, so Multimedia Panels Accordion listed image, panels, columns, CTA, device, video, accordion, links, and text before you committed to it.

At the point these screens were captured the library held 22 organisms with more than 150 configurations, generating over 300 organism variants.

Act 2: Locally relevant at scale

A global design system and locally relevant content pull against each other. Push too hard on consistency and every market gets a page that reads like it was written somewhere else. Push too hard on autonomy and the brand stops meaning anything.

Canvas resolved that by fixing the structure and freeing the content. The organism was standard. What went inside it was the market’s call. A regional team could not invent a new hero pattern, but they could decide what the hero said, which proof points earned space, and which offers were worth the page.

The guidelines had to carry that argument to people who did not report to the design system team and had campaigns to ship. Its navigation was built on what a marketer was trying to do, not on how the system was organized: Introduction, Design, Build, Evolve, Get help. The website is one of our most important customer touchpoints: use these resources to build pages that inform, engage, and motivate customers to sign up.

Every organism was created from the customer back, tested in market, and only then added to the global library. The team held the loop length between user tests to under two weeks.

Act 3: Writing the tool that writes the site

The part of this work that was most clearly content design never appeared on PayPal.com. It appeared inside Canvas.

Every organism needed a definition that a marketer could act on in one read: what the component is for, and when to reach for it.

The authoring view put the page composition beside the live preview, so the stack of organisms on a page was visible and editable as a list. Each organism entry carried a documentation scaffold behind it with fixed sections: required elements, style, best practices, options. Fixed sections are a content decision.

A third tool solved a more human problem: getting dozens of regional marketing teams and agency partners, most without engineering or analytics support on hand, to tag their own page events consistently for reporting and optimization. In collaboration with our SEO and analytics partners, I built a generator in Airtable to assemble structured IDs. The format was designed to be used consistently, so finding a pattern in a spreadsheet of thousands of click events could be more human-readable.

The same thinking produced the keyword planner. Search data on its own is a spreadsheet; what made it usable was the taxonomy laid over it. Every keyword was classified by country, type, sub-type, category, funnel stage, and sub-category, so a marketer could filter to consumer-facing acquisition terms for their unique market challenges. The instructions were written as four plain questions: which country, how much volume, which audience, where in the funnel.

Close

The through-line is that a design system is a language before it is a library. Twenty-two organisms only reduce anyone’s work if the labels are clear, the definitions agree with each other, and the tool explains itself in the moment someone is using it. That is content work, and almost none of it is visible on the finished page.

Regional marketing teams got a platform for publishing pages that were recognizably PayPal without being handed someone else’s words.