Nobody funds consistency.
Petz had no design system, an ageing interface, and two commercial goals. I built the system as the answer to the goals.
Petz is one of Brazil's largest pet retailers.
When I arrived in 2020 there was no design system. The app and the website had drifted apart, and the interface was years behind.
The obvious pitch was consistency. I didn't make it.
Consistency is a designer's complaint. The business had two goals of its own: increase conversion, and grow fast. So I translated the system into those terms before I built any of it. Increased conversion became frictionless trust. Aggressive growth became speed to market. The system wasn't a cleanup project. It was the way to get both.
Then the pandemic arrived and online stopped being one channel among several. It became the priority. The system got more urgent and a lot more visible.
I started with no formal authority over the designers I needed. In 2021 the team grew and the function became mine. The system is the thing I used to earn that, and the thing I left behind: after Petz merged with Cobasi in December 2025, the successor organisation built its system using mine as the base.
Petz sold through a website, an app and physical stores. The digital products had been built at different times by different people, and it showed.
The same component existed in several versions. Type sizes were chosen per screen. Buttons had no agreed size, radius or padding.
That's the version a designer tells. It's true and it convinces nobody with a budget.
The version that matters is the second-order one. Every new screen started from scratch, so every new screen took longer than it should have and arrived slightly different from the last. A checkout that looks assembled from parts asks the customer to trust it anyway. A campaign page that has to be built from nothing can't go out in the week the campaign needs it.
Nobody at Petz was going to fund a consistency project. They were funding conversion and growth. The design system had to be the answer to those, or it wasn't going to happen.
I joined in 2020 to lead UX.
At the start that was a description of the work, not of my authority. The designers I needed were not mine, and neither were the engineers.
In 2021 the team grew and the function became formally mine, including recruiting, mentoring and setting how the team worked.
That order matters. I didn't get a team and then build a system. I built the system without a mandate, and the mandate followed.
Petz had two commercial goals. I took both and rewrote them as design problems.
Increase conversion → frictionless trust. The goal underneath was sales, and specifically checkout completion. A customer abandons a checkout that feels uncertain. Inconsistency reads as uncertainty: an unfamiliar button, a form field that behaves differently from the last one, a price displayed in a shape you haven't seen before. So the system's first job was to make the buying path feel like one product built by one company.
Aggressive growth → speed to market. The goal underneath was launching seasonal campaigns faster, testing new product verticals, and reacting to the market inside the week rather than the quarter. You cannot do any of that if every page is built from nothing. So the system's second job was to make new surfaces cheap.
Two goals, two arguments, both in the language the people funding it already used.
This is the part I'd point at if you asked me what changed at Petz because I was there rather than someone else. The components would have been drawn either way. What I changed is what the system was for, and therefore who paid for it and who cared whether it got used.
The failure mode of a design system is a single person who owns it. It grows while they push it and stops the day they're busy.
So the governance came before the component library was finished. Three things:
A contribution model. A defined path for a designer to propose a component, get it reviewed, and have it added. Contributing was a normal part of the work, not a favour to me.
Named owners. Parts of the system belonged to people. Not to a committee, and not to me by default.
A review cadence. The system was looked at on a schedule, not when something broke.
The cost is speed. A contribution model is slower than deciding it yourself, every single time, and for a while you are watching people make choices you would have made differently. I think that trade is correct and I'd make it again, but it is a real cost and it is paid immediately, while the benefit arrives much later.
The documentation lived in Confluence, where the whole company already worked, not in a design tool where only designers could reach it.
And it covered more than component specs.
Writing standards. How CTAs are written, how form fields are labelled. The words in an interface are part of the interface, and they were drifting the same way the visuals were.
Accessibility, at the point of implementation. I worked with developers in the presentation layer, in the HTML and CSS, to understand where accessibility was actually breaking. A specification that only describes appearance doesn't survive being built.
Both design tools. The framework was applied in Adobe XD and Figma, so the system didn't fracture along the line of which tool a team happened to use.
The test I was designing for: someone who wasn't in the room should be able to use this correctly. That test is also the reason the system was still legible to another organisation five years later.
The app went first. The site followed.
Doing both at once would have been slower in the only way that counts. A system nobody has used yet is a set of opinions. Getting it into one real product turns the opinions into something with evidence attached, and the second platform inherits a system that has already been argued with.
The cost is that for a period you are running two realities, and the site is visibly the older one. You have to be willing to let something look unfinished in public while you finish it.
I had no authority in 2020 and a team in 2021, and the way I worked had to change with it.
Without authority, the only currency is usefulness. I made the system the fastest way for a designer to do their own work. Adoption that comes from a component being genuinely quicker to use is more durable than adoption that comes from a rule.
With the team, the job became setting the bar rather than holding it myself. That meant critique on a schedule, decisions that had a named owner, and documentation of the reasoning rather than just the outcome, so a decision could be re-opened by someone who wasn't there when it was made.
The recruiting and mentoring came with that.
For the first stretch, every decision came through me.
I was reviewing every component, resolving every disagreement, and approving every addition. It felt like quality control. It was a bottleneck, and it had a second cost I didn't see at the time: while everything routes through one person, nobody else builds the judgement to decide it themselves.
The contribution model above is the fix. I should have built it earlier, before the system was good, when it was still small enough that other people's mistakes in it would have been cheap.
The lesson I took is that a design system is a change management problem wearing technical clothes. The components are the easy half.
Adoption. The app first, then the site.
Speed. This is the result I'd defend hardest. Campaign and seasonal pages stopped being built from nothing. Work that used to take a sprint went out in days, which is the difference between a campaign that lands in its own week and one that lands late.
Checkout completion improved. I'll say clearly what I can't prove. This was 2020 and 2021 in Brazil, when the pandemic pushed retail online and Petz prioritised digital over the physical stores. Everything about online performance moved in that period, for every retailer. I'm not going to claim a design system caused a number that a global event was also pushing. What I'd stand behind is narrower: the buying path became consistent, checkout completion improved over the same period, and the system was part of why.
What outlived it. Petz merged with Cobasi in December 2025. The successor organisation built its design system using mine as the base. That's the outcome I had the least control over and the one I'd put the most weight on. Five years and a merger later, the work was still the sensible thing to build from.
The components were the easy half.