One token system for several brands and products, built without a dedicated team between March and October 2026.
At a glance
Context
Three B2B products (an order management back office, a B2B e-commerce platform and a field sales app) with separate libraries, and customers who want the interface in their own brand.
Problem
Every new brand was done by hand, in Figma and in code; design and development drifted apart; the old token structure couldn’t handle some of the customisations customers asked for.
My role
Project owner on the company side, from choosing the partner to handing over to the team. I designed 60% of the components.
Team
Me as the only in-house designer, a specialist external consultant, two front-end developers.
Timeline
March – October 2026
Outcome
Nile now supports 7 brands and is used by two products, with an automated design-to-code pipeline and a team that worked independently once the consultancy ended.
7brands on a single architecture
83public components, about 1,800 variants
60%of the components designed by me
Some references, such as names, data and brands, are made up or left out because they’re covered by non-disclosure agreements (NDA).
The problem
The existing design system had been built for a single product. Next to it sat three separate component libraries, one for the back office, one for the B2B e-commerce platform and one for field sales, sharing nothing but the foundations. Designers ended up creating local components, and designs lost consistency.
Every customer who wanted the interface in their own colours meant reworking components by hand, in Figma and in code, with design and development constantly at risk of drifting apart. And the token structure had a limit that weighed heavily on customers: it allowed a single brand colour. When a customer asked for two, there was no clean way to say yes.
On top of that, we had a goal that was still unusual in March: make the components consistent and structured enough for AI tools and agents to use them too, so they could generate interfaces and prototypes that stay true to the system.
Constraints
No dedicated team. I was the only in-house designer on the project, working with an external consultant and two front-end developers: one for technical decisions, one for implementing the components.
A consultancy with agreed timing and scope, renegotiated when the project had to take on the integration of a whole UI kit for data tables.
Products already in use by customers, built in Angular.
How it started
In March, together with the head of the UX team, I wrote the brief and met two specialist consultants, walking them through the state of our existing design systems. We kicked off at the end of May, with weekly office hours for the whole project.
March 2026Brief and partner selectionState of the existing design systems, two proposals compared.
End of MayKickoffAudit of the previous design system and of the products, foundations, token structure.
SummerPipeline and componentsPipeline to development, components and patterns; at the end of August, the call to rebuild instead of patching.
October 2026HandoverFour training sessions, documentation, an independent team.
Project milestones, from the brief to the handover.
Key decisions
1. A partner for the architecture, not for the components
Decision
We chose the proposal that built a token system able to carry every brand, with responsive variables, adaptive typography and an automated path from design to code.
Ruled out
A consultancy focused on drawing components.
Why
The problem wasn’t the individual components, it was the structure that had to carry their variations across brands and products.
2. Three token collections, not a three-tier chain
Decision
Primitives (20 colour families, spacing in 4 px steps, radii, borders, typography) are identical for every brand. Each customer’s personality lives only in the semantic tokens, which draw from the primitives. A third collection, responsive, holds the values that change with the breakpoint. Component-specific tokens sit inside the semantic collection, in their own group.
Ruled out
A separate token tier for components.
Why
With this many components, an extra tier multiplies complexity and erodes consistency.
What it meant
Adding a brand means adding a new mode to the semantic tokens. From the customer’s colours we generate the palettes and map them to roles, without touching a single component.
PrimitivesThe same for every brand20 colour families, spacing in 4 px steps, radii, borders, typography
→
SemanticOne mode per brandPrimary colour, backgrounds, text, borders, brand radiiComponent tokens: their own group inside the semantic collection
ResponsiveOne mode per breakpointSizes and line heights that change with the screen
Nile’s three collections. Semantic and responsive tokens both start from the primitives, along two different axes: brand and breakpoint.
Brand previewMouse · size mdTouch · size xlOrder now
The same primary · filled button in three invented brands, Ardesia (industrial), Velluto (fashion) and Brezza (services), on three devices, using Nile’s token names. Semantic tokens follow the brand; the device picks the size (md with a mouse, xl with touch) and the breakpoint; text size and disabled opacity stay the same. The component is never touched. Values are made up.
One button, three brands
01 · Ardesia
The same primary · filled button in the Ardesia brand: the brand’s semantic tokens set colour, radius and font.
02 · Hover
On hover a single token changes: the background switches to specific/button/primary/filled/backgroundHover.
03 · Disabled
Disabled: the opacity comes from layerOpacity/disabled, the same for every brand.
04 · Velluto
Switching brand only changes the semantic values: Velluto has another colour, another radius, another font.
05 · Brezza
Brezza: same component, same token names, different values. The component is never touched.
06 · Touch
With touch, the device picks the xl size and the breakpoint; the text size stays the same.
3. A pipeline instead of manual exports
Decision
Tokens leave Figma through a plugin in the W3C standard format, land in the repository, and a pipeline generates the CSS and publishes the package the front end uses. The commit convention decides the type of release.
Ruled out
Exporting the CSS and handing it to development by hand, as we did before.
Why
Design and development read from the same source of truth, and no value is ever copied by hand again. I set up the plugins as private tools for the organisation, so every designer can use them without a developer setup.
FigmaVariables, one mode per brand
PluginExports tokens as standard W3C JSON
RepositoryA single source of truth
PipelineGenerates the CSS and publishes the package
Angular libraryProducts pick up the updated tokens
How tokens travel from design to code, with no manual steps.
4. Rebuild instead of patching
Decision
At the end of August, facing the trickiest call of the project, we chose a real refactor of the missing components with the in-house team. Most were rebuilt; only a few were reconnected to the new tokens.
Ruled out
Wiring the old components to the new tokens as they were.
Why
It looked faster on paper, but it would have carried the previous system’s debt into Nile. In parallel, we renegotiated the scope of the consultancy.
5. Density by pointer type, not by width
Decision
Five breakpoint ranges shared by every product. Component density depends on the pointer, not on screen width: touch means comfortable density, mouse means compact, at any size.
Why
The same components have to work in a back office used on desktop, in an e-commerce platform used mostly on smartphones and in a tablet app used in the field.
My role, day to day
Coordination: timing, milestones and Gantt with the consultant; sign-off on architecture decisions.
Design: I designed 60% of the components and migrated the most complex ones from the old system, such as the grid version of the card, with slots and variants by device and state.
Quality: systematic component audits with Nile’s AI assistant, finding and fixing recurring violations: opacity-based tokens, placeholders left in files, inconsistent property names.
Governance: I set up work tracking for design and development from scratch, built for a federated model where both designers and front-end developers contribute.
An AI-ready Nile
Ramses is the AI assistant that keeps Nile’s rules: it knows the architecture, naming conventions, component properties and documentation rules. We use it to check a component before release, to generate spec grids and to migrate legacy components to the new tokens. Every new rule goes into the assistant with a version and a changelog.
The documentation is structured, and the components live in a well-organised Figma library that AI agents can read and rebuild through MCP. That keeps the system readable by people and machines alike.
Handover and adoption
Before the consultancy ended in October, I designed a four-session knowledge transfer, with plenaries and separate sessions for design and front end. The recorded lessons became topic-based documentation that works as a manual, along with an installation guide for the plugins.
Since October, developers have been implementing the components in the Angular library, documented in Storybook, and marking them as done straight from Figma’s Dev Mode. Nile’s documentation, covering foundations and components, is being published as an internal site.
Results
7 brands on a single architecture. Adding a customer means creating a new token mode, not changing components.
Two brand colours instead of one. The previous system gave each customer a single colour, one of its heaviest limitations. Now every brand has two, a primary and a secondary, each with its own range of tones.
Icons are customisable too. We introduced four icon libraries, one of them the standard, and a plugin that swaps them in design files in a few steps. It’s a customisation option designed to be sold to customers.
One web library instead of two. The back office and B2B e-commerce libraries became a single library: 75 component sets, about 1,800 variants, 83 public components.
Foundations shared by every product. Today Nile powers the back office and the B2B e-commerce platform; the field sales app is next, with a new mobile library.
From design to code without copy and paste. One source of truth for design and development.
An independent team. The project closed on schedule, and developers now implement components without outside support.
Next steps: measuring the impact
Now that the system has been handed over, the next step is to prove its value with numbers. We’ll measure the time it takes to launch a new brand, the share of components fully tokenised and implemented, and developer questions per component as a signal of how clear the specs are. The first figure will come from the comments already in Nile’s files; the others with the next releases and the first new brand launch.
What I learned
A federated design system with no dedicated team only holds up if its rules are written so that anyone, person or tool, can apply them without asking. And the moments when it pays to slow down, like rebuilding a component instead of just tokenising it, are the ones that save the most later.
Why it’s called Nile
The company I built it for is named after the Egyptian god of the sun disc, whom the pharaoh Akhenaten placed at the centre of worship. Hence Nile: the river that crosses different lands while staying one, the right metaphor for a system that serves many brands.
The tools follow the same mythology. Hapi, the god of the Nile, is the plugin that carries tokens from Figma to the repository; Kemi, the ancient name of Egypt, is the plugin for switching icon families; Ramses is the AI assistant that keeps the system’s rules; Delta, like the river’s mouth, is the pipeline that turns tokens from JSON into CSS.