
Kredivo
Senior Product Designer
2025 — 2026
Building Flex: Kredivo’s Design System
Turning a component graveyard into a design system Kredivo’s designers actually use
Overview
When I joined Kredivo, the design team already had a “design system.” In practice it was a component graveyard: the same button in six variations, colors hardcoded everywhere, and nobody responsible for any of it. I rebuilt it as Flex, with a clear hierarchy, written rules, and a score that tells us whether it’s working.

The problem
Before building anything, I audited the existing Figma file. I went through every component, mapped the duplicates, and traced each inconsistency back to where it started.

Two problems stood out. First, there was no single source of truth. The file was shared storage, not a reference. People created and edited components on their own, with no owner and no canonical version, so a fix in one place never reached the others.
Second, there were no rules, so every decision came down to individual judgment. That works for a small team sitting together. It falls apart when the team grows, new people join, or two designers work on the same product without syncing, and the inconsistencies pile up over time.
Cleaning up the file piece by piece wouldn’t fix either problem. It needed a rebuild.
The approach
Three layers
I organized Flex around Atomic Design and split it into three layers, each with its own job:
- Core components are the building blocks: buttons, inputs, tags, icons. They show up everywhere, so they have to be exactly consistent.
- Composite components combine core pieces into recurring patterns like cards, modals, and navigation bars.
- Product implementations are where each team applies the system to its own context. Flex sets the rules, and the product applies them.

Without this split, the system would slide back into a pile of components with no hierarchy. I chose Atomic Design partly because engineers and PMs could follow it too. Atoms, molecules, and organisms map closely to how our engineers already thought about components, so we had a shared vocabulary from the start.
The name comes from the same idea. Flex is meant to be modular and bend with the product instead of boxing it in.
Deciding what to build first
Systematizing everything at once would have taken forever, so I scored every component on three questions:
- How often does it appear across products?
- How much visual debt does it create when it’s inconsistent?
- What breaks if it stays unstandardized?
The scores sorted components into four tiers by impact: Tier A (70% or more), Tier B (40 to 69%), Tier C (20 to 39%), and Tier D (under 20%).
Nine components made Tier A: Button, Checkbox, Shimmer, Navigation Bar, Divider, Section Title, Chips, Pagination, and Background Overlay. Together they cover most of the real design work across Kredivo’s products, so they got the most rigorous treatment. The rest split into Tier B (5), Tier C (6), and Tier D (35).
I would rather ship a small system that teams use than a complete one nobody opens.
Building it
One standard for every component
Every core component is built the same way, because a system is only trustworthy when its parts are consistent. Each one has:
- A defined anatomy: the container, label, icon slot, and state layer, and what each part does.
- Every interaction state documented: default, hover, pressed, disabled, loading, and error. Missing states are one of the most common causes of engineering rework.
- Variants organized by size, emphasis, and state, so adding one later doesn’t mean restructuring the component.
- Usage rules for when to use it and, just as important, when not to.
Trimming my own token system
The hardest feedback to act on is feedback about your own work. The color token system I had built grew past 140 tokens. Many were unused and several overlapped, so a system meant to reduce maintenance was creating more of it.

Instead of patching tokens one at a time, I rebuilt the set around meaning. One token, shape-brand-primary, now covers backgrounds, borders, and supergraphics, where each of those used to have its own. There are fewer tokens to maintain, the right choice is easier to spot, and a brand color change means updating one token instead of forty.
Spacing
The numbers used to be hardcoded, one screen at a time. Now they are tokens, so spacing and radius stay consistent across pages and designs.
Color
Semantic names stay put. A new brand or a new color mode only swaps the base and the brand tokens underneath, so the color system scales without a second set of names.
Impact
To know whether Flex was helping, I built a Design System Health Score. It weights each component by usage. Heavily used components count more, rarely used ones barely move the score, and a popular component that designers keep detaching pulls the score down.

For Q4 ’25 to Q1 ’26, Flex scored 84.5%, which puts it in the “Highly trusted & efficient” tier (80 to 100%). Behind that number:
- 110,159 component instances across all Figma files
- 94.22% stability, so designers mostly use components as they are instead of detaching and overriding them
- Button alone has 50,306 instances at 99.19% stability
- The 9 Tier A components drive most insertions
We now run the score every quarter and use it to decide what to improve next.
Day to day, design moves faster and needs less rework. New designers get up to speed sooner, and teams share one design language, which means fewer alignment meetings and cleaner handoffs.
Takeaways
The hardest part of a design system is deciding what not to build. The priority matrix kept my limited time on the components that helped the most people.
A design system is a product, not a file. It needs an owner, upkeep, and a way to measure itself. Defining the Health Score forced me to say what “working” means, and that shaped every priority call after it.
Measuring changes how you build. Once detaches lowered the score, I stopped treating them as neutral and started asking why people detached in the first place.
I knew Flex worked when a new designer joined, opened it, and found what they needed without asking anyone.