Charts: data visualization component library
A scalable chart system for an incident management platform. It started as one component and ended as four chart families with dozens of responsive variants.
- Role
- Senior Product Designer
- Timeline
- Platform
- Web (Desktop)
- Team
- Solo designer · engineering team
Overview
ecoPortal needed a consistent, reusable way to show incident data across the platform. I designed a chart library with Area, Bar, Line and Combo charts, each with variants for different levels of data complexity and responsive breakpoints. The process ran end to end: briefing, lo-fi validation, hi-fi design and implementation QA.
The problem
ecoPortal is an incident management platform for companies, and data drives the decisions there. Safety managers, compliance teams and executives need to read trends, compare categories and spot patterns over time. But the platform had no component system for data visualization.
Charts showed up here and there, each designed on its own, with its own visual language, different axes and no shared tokens. The same kind of data looked different depending on the screen.
How do we design a chart system flexible enough for any data scenario the product needs, with enough rules that everything still feels like one product?
The scope was big from day one: four chart families (Area, Bar, Line and Combo), each with several complexity variants and several breakpoints.
Briefing
The product team kicked off the project because of a recurring pain: every new screen with data meant engineering building a chart from scratch. There was no reusable component, no agreed visual language and no guidance on which chart to use when.
The briefing mapped the families we needed, based on current and planned features:
- Trends: how incidents change over time (Area and Line charts)
- Category comparison: breakdowns by type, department or status (Bar charts)
- Geographic distribution: incidents by location or site (Bar chart, Locations)
- Combined metrics: several data dimensions in one view (Combo charts)
Questions set in the briefing:
- How many series should each chart type support, and when should you switch to a different type?
- How do we handle segmented data: stacked, grouped or proportional?
- How does a chart behave at each breakpoint, and what can go at smaller sizes?
- How does a Combo chart mix two chart types without becoming ambiguous?
Ideation & brainstorming
With the four families defined, I explored widely before locking any component structure. I looked at how analytics tools (Tableau, Looker, Grafana, Power BI) build their chart libraries and how design systems like Carbon, Polaris and Atlassian's handle data visualization at scale.
The main hypotheses:
- A variant only exists if a real data scenario in the product justifies it.
- Consistent axes, grid and legends matter more than making each chart distinct. The visual grammar has to stay the same even when the shapes change.
- Chart responsiveness isn't layout responsiveness. Reflowing isn't enough; you have to decide which labels, axes and legends stay at smaller sizes.
- Combo charts need hierarchy. One type reads as primary and the other as a supporting layer, instead of two equals fighting for attention.
I mapped every combination: per family, the complexity variants (1 series, multi-series, segmented, dataset applied) and, per variant, the breakpoints.
Lo-fi usability testing
For a component library, lo-fi testing is less about task flows and more about comprehension and scale. I built lo-fi versions of the main charts and tested them with the people who consume this data every day: analysts and managers who read incident reports.
What I tested:
- Whether people could tell chart types apart quickly and knew what each one was for
- Whether the segmented (stacked) variants were readable right away or needed explaining
- Whether the data stayed readable at smaller breakpoints
- Whether a Combo chart (bars with a line on top) read as one visualization or as two charts stacked together
What I found:
- Multi-series Area charts were mistaken for Line charts. The fill needed more visual weight for Area to have its own identity.
- Segmented Bar charts needed a legend at every breakpoint, even the smallest. Without it, nobody could read them.
- Combo charts with both layers at the same weight confused people. The test confirmed one layer had to sit clearly behind the other, with a thinner stroke and slightly lower opacity, to set the reading order.
- The Locations bar chart was the easiest to understand. Horizontal bars labeled with place names were the most natural way to compare regions.
Flow refinement
With the results in, I revised the component structure before hi-fi. The changes were structural (how the components were organized and which rules drove their behavior), not visual polish yet.
What changed
Area chart with more identity: I raised the fill opacity and thinned the line so Area and Line read as different chart types, not style variations.
Legend rule: legends became mandatory at every breakpoint for segmented variants. At the smallest sizes labels can be truncated, but the legend never goes away.
Combo chart hierarchy: the primary metric uses bars (solid, full opacity) and the secondary uses the line (thinner stroke, slightly lower opacity). I documented this as a usage rule for the component, not just a design decision.
Breakpoints defined: large (default desktop), medium (less padding, smaller axis labels), small (only essential labels) and an even smaller thumbnail variant for dashboard summary cards.
Segmentation scope: segmented versions were defined for Bar and Combo, based on real product scenarios. Segmented Area made it into scope; segmented Line was left for a later cycle.
Hi-fi design
With the structure validated, I moved to hi-fi, building every variant, state and responsive size with ecoPortal's design system tokens.
The final library has four families:
Area chart: three variants. 1 series (one trend line with fill), multi-series (overlapping fills in different colors) and segmented (stacked areas). Used for trends over time.
Bar chart: four variants. 1 series (one vertical bar per category), multi-series (grouped bars), segmented (stacked bars) and Locations (horizontal bars with site or location names on the Y axis). Used to compare categories and regions.
Line chart: three variants. 1 series, multi-series (lines in different colors with point markers) and segmented. Used to compare trends when exact values matter more than volume.
Combo chart: four variants. 2 series (main bars with a line on top), Dataset applied (bars colored by dataset dimension), Segmentation (stacked bars with a line on top) and a fourth variant for custom combinations. Used to lay a trend over a categorical distribution.
Each variant comes in four sizes, with explicit rules for what appears at each breakpoint: which labels, axis ticks, legends and interactive states.
Design decisions
Unified tokens: every chart uses ecoPortal's data visualization tokens (color ramps, axis color, grid opacity, label type). A new chart type later on needs no new tokens, just a new component assembly.
Dark background by default: charts use a dark canvas with light data, matching the rest of the platform's dashboards. Beyond the look, a dark background is easier on the eyes in the monitoring contexts where these charts show up most.
Thumbnail variants: each type has a thumbnail version (the smallest responsive size) made for dashboard summary cards, where the chart shows trend direction rather than exact values.
Handoff & design QA
As in all my ecoPortal projects, I owned implementation QA on top of the handoff. I followed every sprint of the library and checked each delivered component against the spec.
The handoff docs covered the anatomy of each variant, a token map per element (fill color, axis line, grid opacity, label style), responsive behavior per breakpoint and usage rules saying when to use each type and variant.
We ran two handoff sessions with engineering: one for Area, Bar and Line, and one just for the Combo chart, which needed a more detailed interaction spec because of its layered behavior.
What QA caught:
- Area chart fill opacity not matching the token in the multi-series variant
- Gap between bars in the grouped Bar chart slightly off spec
- Line chart point markers at the wrong size at the small breakpoint
- Combo chart line rendering above the bars instead of below (a z-index issue)
- Legend colors not matching series colors in the dark theme
Everything was fixed before the library was released for use.
Results & learnings
The library became the foundation for every data screen in ecoPortal, current and future. Any new screen that needed a chart could use the library instead of starting from scratch, which cut design and engineering time per screen considerably.
What worked:
- Putting usage rules inside the component, not only in a separate doc, let engineering pick the right chart on their own without checking with design every time.
- The breakpoints validated in lo-fi held up through implementation, with only a few small QA fixes.
- Tokens built for a dark background made the library work with the existing dashboards right away, with no visual adaptation.
What I'd do differently:
- Audit every existing screen before designing. Mapping each improvised chart that already existed would have revealed more edge cases before lo-fi.
- Spec the interactive states (hover tooltip, active point) in the same cycle instead of deferring them. They were the first request after launch.
- Create a formal decision tree ("which chart should I use for this data?"). It would have saved engineering a lot of questions during implementation.
Next steps: spec interactive states (hover, tooltip, active selection), add a donut/pie family for part-of-whole data, and bring the library's tokens into ecoPortal's broader design system documentation.