---
title: "Compass Design System — Maximilian Dahl"
description: "Product design case study by Maximilian Dahl: Compass, a design system run as an internal product at Container xChange. Overall sentiment from 55% to 71%."
url: https://max.does.systems/en/work/compass-design-system/
lang: en
---

Product Design Case Study · Design Systems · Container xChange · container logistics · 2022 – 2025

# Running a design system as an internal product

Container xChange was growing faster than its UI kit: 79 different buttons, and fixing a single component could eat a week. I started Compass, built the team and ran the system as a product. It is still running today, outliving the team that built it.

Read the case · RJump to results

Role in the project · Design Systems Team Lead · and lead designer: strategy, architecture, team, rollout, measurement

Team · Guild, then systems team · First on my own, then with a direct report; later 2 design, 1 QA, 2 engineering

Responsibility · Compass as an internal product · Roadmap, OKRs per trimester, research and the sentiment report

Status · Live · +2 · As of September 2026 · Live · In the live product since T2 2023, after 11 months from zero to one · In use · Still in use and still evolving · Open · Documentation use stayed the weak spot at 2.2 · and 2 more states

As of September 2026

**Live · +2**

As of September 2026

- Live
   In the live product since T2 2023, after 11 months from zero to one

- In use
   Still in use and still evolving

- Open
   Documentation use stayed the weak spot at 2.2

A design system is not a kit you ship once. It is a product with users and upkeep, and it has to keep running without its team.

Compass Design System

## At a glance

- 55 → 71%
   overall sentiment in the Compass Sentiment Report
   **55 → 71%**overall sentiment in the Compass Sentiment Report · Internal survey, Q3 2022 → T2 2023

- 11 → 82%
   positive responses about Compass
   **11 → 82%**positive responses about Compass · Compass Sentiment Report, Q3 2022 → T2 2023

- 79 → 1
   buttons in the live product, then one Compass button
   **79 → 1**buttons in the live product, then one Compass button · Counted in the code; the Compass button live in T2 2023

1. Problem
   A product, not a UI kit
   79 different buttons in the live product. I set Compass up with users, a roadmap and OKRs.

2. Intervention
   Building blocks first, docs later
   250 tokens, one button instead of 79, live in T2 2023. The price: documentation use stayed at 2.2.

3. Outcome
   Outliving the team
   4 of 7 key results in T2 2023. Compass is still running, outliving the team that built it.

Role in the project**Design Systems Team Lead**and lead designer: strategy, architecture, team, rollout, measurement

Team**Guild, then systems team**First on my own, then with a direct report; later 2 design, 1 QA, 2 engineering

Responsibility**Compass as an internal product**Roadmap, OKRs per trimester, research and the sentiment report

02 · The system as a product

## Not another UI kit, but a product with users

The old UI kit could no longer be reworked. Every team solved problems on its own, flows and interfaces contradicted each other, and one-off solutions ate time that product work needed. The live product had 79 different buttons in use. We counted them in the code: a developer on the team had built a script that tracked down duplicates and custom code.

The old kit was broken at the root: nested spacers inside every component, with Bootstrap still lurking underneath. Every change turned into a repair job.

Instead of building yet another UI kit, I set Compass up as a product: its own strategy and roadmap, OKRs every trimester and three user groups, designers, engineers and product managers. I pitched it and represented it in the Product & Tech Leadership Circle. My design lead sponsored it, with backing from product and engineering leadership.

At first there was no official team for it. Compass began as an initiative and grew step by step:

From initiative to team

1. Me
   Guild
   First foundations alongside day-to-day work, laid by me

2. Me · Product designer
   Product designer
   Brought in with my design lead, reporting to me

3. Me · Product designer
   Systems team
   Me as Design Systems Team Lead and lead designer

4. Design · Design · QA · Engineering · Engineering
   With engineering
   Later 2 design, 1 QA, 2 engineering

Context

Users and roadmap first, components second

Users, a roadmap and OKRs came before the first component. The cost was time spent on research and buy-in before a single component existed. In return, Compass had users, sponsors and a goal to be measured against from day one.

03 · Research

## One system, four roles, four different jobs

Before we touched a single component, I found out who needed Compass and for what. The Q3 2022 baseline survey had its own questions for every role: how does the old system feel, how satisfying is it, where does time get lost, what does documentation cost, who contributes. 35 people answered, 18 from engineering and QA, 12 from product and leadership, 5 from design. We added interviews with a pain-point analysis, a review of tools and processes, and benchmarks such as Shopify Polaris and IBM Carbon.

The baseline survey asked each role its own questions. Four jobs to be done came out of it, one per role:

- **Product designer**Job story
   When · I design a new feature,
   I want · proven building blocks,
   so · my time goes into the problem, not into rebuilding UI.

- **UI engineer**Job story
   When · I build an interface,
   I want · Figma and Storybook to show the same thing,
   so · I don't rebuild what already exists.

- **Product manager**Job story
   When · I plan a feature,
   I want · to know what the system already covers,
   so · I can decide faster and not wait for one-off solutions.

- **Brand and marketing**Job story
   When · I build a campaign,
   I want · the same colours and rules as the product,
   so · xChange looks the same everywhere.

Summarised from roles, research questions and sentiment answers.

> Absolutely chaotic, outdated and impractical.
> 
> Voice from the research, 2022

> I spent a week fixing a single component.
> 
> Voice from the research, 2022

To find the structure, we ran a card sort on the existing components: 46 building blocks, five people from design and engineering. It produced the taxonomy Compass still runs on today: foundations, actions, inputs, data display, containers, overlays, navigation, feedback, notifications, patterns, loaders, atomic elements and branding.

04 · Strategy and focus

## Building blocks first, the big vision later

The first plan was a research plan. We wanted to understand the problems, gather qualitative insights and derive a target picture we could plan backwards from. The models were scalable systems like Shopify Polaris and IBM Carbon: a foundation you can build several houses on.

The plan in October 2022 · Status update to the department

1. Original
   Vision first
   Research and learning, leading to a target picture for a system like Polaris or Carbon.

2. New
   Building blocks first
   Tools that let teams ship faster right away. The research results were kept.

Why the switch: the company and its products were in an experimental MVP phase.

Context

Building blocks before the vision deck

Lego bricks first, the vision deck later. In October 2022 I turned the focus around. The Polaris-style research vision stayed the goal, but the company needed tools that made teams faster right away. So first the right Lego bricks for every role, ones that work in the vast majority of cases, and the big documentation afterwards. The cost: the documentation came later, and the T2 2023 survey showed exactly that.

That set the goal for Q4 2022: the new UI kit for design by the end of the year, and for engineering from Q1 2023. At the same time we worked out with engineering how the route from design into code should look and what they needed for it.

For alignment we sorted everyone involved into three rings: who builds, who has a say and who stays informed. That made it clear whom to involve when.

Stakeholders in three rings

1. 01 · Core team · UX/UI design from product, front-end engineering from the marketplace, graphic design from marketing, QA as an open role

2. 02 · Involved · Brand lead, product managers, the front-end guild with a senior engineer

3. 03 · Informed · Head of engineering, head of product and further leadership

05 · Token architecture

## A name says what a colour is for

Compass had to carry five products: Leasing, Trading, Trucking, Slots and Wallet. So it was laid out as a hierarchy: tokens at the bottom, components, patterns and templates on top of them, the products at the very top. Whatever should look consistent at the top needs one shared source at the bottom. Without it, 79 different buttons had appeared.

From token to product

1. Tokens
   Colours, type, elevation, spacing, shadows

2. Components
   Toggles, buttons, input fields, badges, checkboxes

3. Patterns
   Tables, cards, page headers, modals, lists

4. Templates
   Navigation, login, funnel, profile, dashboard

5. Products
   Leasing, Trading, Trucking, Slots, Wallet

The Compass library at a glance

Categories · 13, from a card sort in July 2022: 46 building blocks, five people from design and engineering · Foundations · Actions · Inputs · Data Display · Containers · Overlays · Navigation · Feedback · Notifications · Patterns · Loaders · Atomic Elements · Branding

Modules · Complex blocks such as page header, navigation and shell elements, with their own area on the board

Homes · Figma kit for design, Storybook for code, Zeroheight for the docs

Benchmarks · Shopify Polaris and IBM Carbon

The old kit knew colour values but not their jobs. So I named every colour by role, family and state: “interactive-primary-hovered” is the hover of a primary element, whatever tone sits behind it. Semantic names are longer and take discipline. In return, a tone changes in one place and reaches all five products, instead of being chased through every file.

The grammar of the tokens · Live in the lab

max.lab · Compass · Live · rebuilt · rebuilt · Rewritten for this site, not the original code. Where a detail is approximated, the docs say so.[Docs](https://max.does.systems/lab/#/compass/farbe?lang=en)

Pick a family and a state: name and value follow the same rule.

We kept the foundation deliberately small, because usability and maintainability were the first objective: 250 tokens, ten colour roles with up to six states per family, eleven type sizes on a 4px grid, twelve spacing steps, six radii and two shadows. During the rollout we cleaned the tokens up and scaled them so new products could adopt them instead of inventing their own. That is what the hierarchy was for.

The alert draws on the same families as the button: danger, warning, success, informative. Change a tone and both change. [See the alert in the lab](https://max.does.systems/lab/#/compass/komponenten/alert?lang=en).

06 · One button instead of 79

## Every edge comes from a token

The button shows how the system works at small scale: eight styles, four heights from 24 to 48 pixels, an icon on the left, the right or on its own, and a loading state. Spacing, radius, type and colour each come from a token, including hover, pressed and focus.

Button with token inspector · Live in the lab

max.lab · Compass · Live · rebuilt · rebuilt · Rewritten for this site, not the original code. Where a detail is approximated, the docs say so.[Docs](https://max.does.systems/lab/#/compass/komponenten/button?lang=en)

Switch on “Measure” and the inspector shows every spacing and colour with its token. The states sit in the matrix below.

For this page I rebuilt Compass: measured from the rendered original and written from scratch in React. I rebuilt 65 components this way and checked each against the original, measurement by measurement; whatever I had to approximate is listed on its page.

Measuring exposed a gap: in the original, a disabled toggle looks almost like an enabled one; only the icon colour and the cursor change. The rebuild shows it on purpose, the way it was. [See the gap in the lab](https://max.does.systems/lab/#/compass/komponenten/toggle?lang=en).

### Accessibility

Accessibility was a binding design principle in Compass: keyboard use, contrast, colour use and type sizes. The starting point was the old kit, whose colours lacked contrast. During the rollout we sharpened the contrast of the neutral tokens for placeholder and subtle text, made the type tokens responsive and checked all 57 components against the responsive criteria. There was no full WCAG review or audit.

The rebuild now measures against WCAG and shows where the original still had gaps: the disabled toggle is barely distinguishable from the enabled one, some contrasts fall below AA, and not every element shows a visible focus. In the rebuild these spots are documented on the component page and fixed where possible.

07 · Operations and governance

## Contribute, don't fork

A library alone doesn't carry a system. The Figma kit, Storybook and the documentation had to stay in sync, and teams needed a clear way to contribute instead of building next to Compass.

One source, four stations · on a weekly cadence

1. Figma kit
   Components and tokens as the source for design.

2. Storybook
   The same components in code, reworked during the rollout.

3. Zeroheight
   One home for rules, examples and usage.

4. Product teams
   Use, report and contribute through contribution guidelines.

The hardest route was the one from design into code. In autumn 2022 we worked out with engineering what they needed and planned a test library. In T2 2023 we reorganised and overhauled Storybook. That did not close the gap completely: in the survey, teams asked for Storybook, docs and Figma to line up better, and missed code documentation.

Above the components sat the patterns. We ran complex building blocks such as page headers, navigation and shell elements as modules, with their own area on the board: 42 of the 181 tickets in T2 2023. A pattern library for recurring flows came on top.

One flow instead of many parts · Live in the lab

max.lab · Compass · Live · rebuilt · rebuilt · Rewritten for this site, not the original code. Where a detail is approximated, the docs say so.[Docs](https://max.does.systems/lab/#/compass/komponenten/pattern-form?lang=en)

Stepper, fields and validation as one pattern. Change the step or trigger an error: the notice, the field and the stepper show it together.

Rules turn building blocks into a system. The survey asked designers how they use the kit, on a ladder from one-off solutions to global components only. In T2 2023 all seven designers stood on the same rung.

How designers use the kit · Q3 2022 → T2 2023 · How designers use the kit. Q3 2022, 5 designers: 3 at one-offs, 2 at adapt and give back. T2 2023, 7 designers: all 7 at extend by the rules.

Extend by the rules · 0%**100%**+100 pp

One-offs · 60%**0%**−60 pp

1 · One-offs · One-offs · Every design a one-off, so-called snowflakes.2 · Mix · Mix · One-offs and global components, adapted personally.3 · Adapt and give back · Adapt and give back · Adapt global components and give the change back to the system.4 · Extend by the rules · Extend by the rules · Use global components, extend them by set rules or request new variants.5 · Global components only · Global components only · Without any adaptation.

One answer per designer.

Two principles sat behind it. Before anyone builds a new component, they check whether an existing one can be extended. And Compass provides building blocks that fit 80 per cent of the time, without anyone having to reinvent them.

Context

A one-week sprint instead of two

At the product teams' request we cut the sprint from two weeks to one, so design and engineering no longer ran in separate, consecutive weeks. We brought marketing and brand on board so Compass would hold outside the product, too.

### Open to everyone, with set routes

Compass was open to contributions from every team. To keep that from turning into chaos, every contribution needed a set route into the system, and every role knew what it was responsible for. Every release got a version, and anything being replaced was marked deprecated before it was removed. We planned to retire the old UI kit and its legacy code over two trimesters. That was harder than expected, because nobody knew exactly where old code was still hiding.

From need to release · every week

1. Request or contribution
   In the #design-systems channel or by the contribution guidelines

2. Parking lot and backlog
   Everything lands on the Compass board, prioritised by the OKRs

3. Weekly sprint
   Design and engineering in the same week

4. Quality gate
   My frontend developer checks every contribution before release

5. Release and changelog
   As a version, with the changelog in Confluence, Zeroheight and the channel

Who takes care of what

1. Systems team
   Plans, builds, releases and measures the system, prioritised through OKRs every trimester.

2. Design
   Uses the building blocks, extends them by the rules and requests new variants.

3. Engineering
   Keeps the components in code identical to the kit and watches over consistency.

4. Product
   Asks for existing components first and plans MVPs from what exists.

5. Marketing and brand
   Brings the brand guidelines into the docs, since T2 2023.

Split from the target picture at the start; marketing and brand joined in T2 2023.

Context

Open to everyone instead of built centrally

Compass was open to all instead of built centrally: we led, but invited everyone to contribute. The cost: some found the rules too relaxed, and quality assurance had to grow with it. Both came back as requests in the survey.

A system doesn't adopt itself. We presented Compass at all-hands and tech town halls, started an AMA for designers, set up a channel for questions and a Confluence page with a changelog, and opened the system to contributions from every team.

Learning

If you think you are over-communicating, you are probably doing it right. A design system lives on its messages landing again and again.

08 · Rollout and measurement

## Measuring what tickets can't show

In T2 2023 the new UI kit went live, and the team rebuilt the whole platform on Compass, right up to the frame around every page.

181 tickets by type · Rollout in T2 2023

1. Components_**41%**Components · Building and maintaining the building blocks, from the button up._75 tickets

2. Modules_**23%**Modules · Composed parts made of several blocks, such as tables and forms._42 tickets

3. Docs_**14%**Docs · Documentation pages: usage, variants and rules per block._25 tickets

4. Strategy_**12%**Strategy · Roadmap, OKRs, research and alignment with the teams._21 tickets

5. Foundation_**7%**Foundation · Tokens: colour, type, spacing and radii._12 tickets

6. Other_**3%**Other · Everything not assigned to one of the types._6 tickets

- 15
   components and modules documented to our MVP definition

- 100%
   of components meet the responsive criteria

From token to product screen · Live in the lab

max.lab · Compass · Live · rebuilt · rebuilt · Rewritten for this site, not the original code. Where a detail is approximated, the docs say so.[Docs](https://max.does.systems/lab/#/compass/komponenten/pattern-marketplace-list?lang=en)

An order list built only from Compass parts. Switch data state and density: filled, loading, empty, error.

Context

Measure sentiment instead of counting tickets

We measured sentiment, not tickets. How much faster a design system makes a team is hard to measure honestly without consistent tickets and estimates. So I designed the Compass Sentiment Report and ran it twice: a baseline in Q3 2022 with 35 responses and a repeat in T2 2023 with 33, on a 1-to-5 scale, across design, engineering, product management and leadership.

Measurability was a key result in its own right: at least two qualitative and quantitative metrics for the impact on product and business. For the quantitative side we compared notes with teams at booking.com, Figma and Pinterest; all of them were wrestling with the same question. My proposal was a shared label for design-system work. Until then, the sentiment report carried the load, with its own scoring across all answers.

All answers compared · Q3 2022 → T2 2023 · Share of answers from 1 to 5. Q3 2022: 17, 16, 46, 17 and 5 percent. T2 2023: 5, 9, 30, 41 and 15 percent.

Satisfied · 4 and 5 · 21%**56%**+35 pp

Dissatisfied · 1 and 2 · 33%**14%**−19 pp

Share of answers per rating, across all questions.

Averages in T2 2023, scale 1 to 5

- 3.9 · +0.9
   Compass overall · before 3.0
   **3.9**Compass overall · before 3.0 · Average of all answers about Compass overall, on a scale from 1 (very dissatisfied) to 5 (very satisfied). From 3.0 in Q3 2022 to 3.9 in T2 2023.

- 4.0 · +1.4
   Figma kit (design) · before 2.6
   **4.0**Figma kit (design) · before 2.6 · How designers rate the Figma kit. From 2.6 to 4.0: the kit went from obstacle to tool.

- 4.0 · +2.2
   UI kit used as intended (design) · before 1.8
   **4.0**UI kit used as intended (design) · before 1.8 · How consistently designers use the UI kit by its rules. At +2.2 the biggest jump among these values.

- 4.0 · +0.9
   Code library (engineering) · before 3.1
   **4.0**Code library (engineering) · before 3.1 · How engineers rate the code library. From 3.1 to 4.0; positive answers from 28 to 69 per cent.

- 3.5 · +0.4
   Code library usage · before 3.1
   **3.5**Code library usage · before 3.1 · How much the code library is used day to day. The smallest rise: from 3.1 to 3.5.

- 2.2 · newly measured
   Docs: usage
   **2.2**Docs: usage · Asked for the first time in T2 2023, so there is no earlier value. At 2.2 the lowest value, the price of the decision: building blocks first, docs later.

That weak spot was the price of my October 2022 call to ship the bricks first and the docs later: documentation usage scored 2.2, well below everything else.

09 · Outcome

## Trimester scorecard: 4 of 7 key results

When xChange had to restructure, the systems team was disbanded. I took over Financial Services. Compass is still in use and still evolving; the designer who stayed on has made it tool-agnostic for AI-assisted coding. My design lead at the time later wrote that I had built the initiative from the ground up.

Compass Sentiment Report, Q3 2022 → T2 2023

- 55 → 71%
   overall sentiment in the Compass Sentiment Report
   **55 → 71%**overall sentiment in the Compass Sentiment Report · Points against the maximum across all questions, scale 1 to 5: 55.1% from 35 responses, 70.5% from 33. The questionnaire grew from 7 to 17 questions.

- 11 → 82%
   positive responses about Compass
   **11 → 82%**positive responses about Compass · Share of answers rating Compass 4 or 5, from 35 and 33 responses.

- 28 → 69%
   Engineering: positive responses on the code library
   **28 → 69%**Engineering: positive responses on the code library · Engineering only, on the code library: average 3.1 → 4.0, from 18 and 13 responses.

We planned in trimesters with OKRs. The first objective in Q4 2022 was deliberately narrow: the UI kit ready for designers by year end, for engineering from Q1 2023. In T2 2023 there were three objectives. In Jira, each key result got its own version and each theme its own epic, from foundation through components and modules to strategy and operations. At the end of every trimester you could see which work fed which objective, and how much ran without one.

1. 15 components and modules documented to our MVP definition
   Reached in T2 2023.
   O1 · Usability and maintainability

2. Brand and marketing guidelines fully in Zeroheight · open
   Started: brand and marketing took on ownership, but it was not complete within the trimester.
   O1 · Usability and maintainability

3. Change communication strategy · cut
   Cut on purpose: there were not enough people for a communication strategy of its own.
   O1 · Usability and maintainability

4. At least 2 qualitative and quantitative impact metrics
   Reached in T2 2023.
   O2 · Make product impact measurable

5. Remaining share of the old kit in code, for deprecation · cut
   Cut on purpose: where the old kit still sat in the code could not be found reliably.
   O2 · Make product impact measurable

6. Responsive criteria defined
   Reached in T2 2023.
   O3 · High-quality design output

7. All 57 components meet the criteria
   Reached in T2 2023.
   O3 · High-quality design output

I dropped two key results on purpose: where the old kit still lived in the code couldn't be pinned down reliably, and we lacked the people for a dedicated communication strategy. OKRs aren't a scorecard for me; they're a way to prioritise, and you're allowed to change them mid-flight.

### Outlook: From foundation to partner in projects

Up to T2 2023 we built the foundation. After that, Compass was to move into a project-based way of working: share goals and the roadmap openly, and sit at the table from the start of bigger projects.

The design system as a stakeholder in future projects

- 3.6
   average agreement, scale 1 to 5

- 67%
   positive answers

What was on the table for the next version · Requests from the survey, T2 2023

1. A visible roadmap
   A roadmap for the next trimester, so teams can plan.

2. Patterns and flows
   Guidelines, patterns and global flows, not just components.

3. Code and docs
   Code documentation and a Storybook that matches Figma and the docs.

4. Mobile and multilingual
   The mobile and localisation initiative, restarted in T2 2023.

The vision from the start was not off the table. It now came after the building blocks teams had needed right away, and would have been the next step.

What remained

When the systems team was disbanded, Compass stayed and is still in use. To me that is the real test of a design system: whether it keeps running without its team.

See also:[Camelot Design System](https://max.does.systems/en/work/camelot/)[Design System UnityUI · AI-ready](https://max.does.systems/en/work/unityui/)

10 · Looking back

## What I learned and would do differently today

### Caveats

The initiative was mine; many people built Compass: first a guild, then the systems team with design, QA and engineering. The figures come from an internal survey with 35 and 33 responses; values per discipline rest on small groups. As of October 2026.

### Learnings

1. Usable first, complete later
   The building blocks came before the docs. Teams got faster, and documentation use at 2.2 was the price.

2. Big rollouts need project leadership
   A rollout across the whole platform is a project of its own. It ended the trimester as a learning in our report.

3. Speed needs a shared label
   Without consistent ticket labelling, velocity cannot be measured honestly. Until then, the sentiment report carried it.

4. Act proactively
   Don't wait for teams to ask for the system. Go to them and show what it can do.

### What I would do differently today

1. Involve engineers earlier
   More engineering in the concept phase would have made technical integration easier from the start.

2. Start with pilot teams
   Rolling out to a few teams first would have brought earlier feedback and more room to adjust.

3. Plan measurement from day one
   I'd introduce a shared label for design-system work before the first component, not propose it after the fact.

4. Check every state early
   Test every state for contrast and recognisability from day one, not in review. The disabled toggle showed why.

> A design system is a product in its own right. It needs clear ownership and ongoing care. Either you build it to last, or it becomes overhead.
> 
> Maximilian Dahl

fin (for now)

**Case notes · How to read this case.**

Evidence · My own case study, the Q3 2022 baseline survey and two internal trimester reports (Compass Report and Sentiment Report T2 2023).

Measurement · Internal survey, scale 1–5, 35 and 33 responses. Per-discipline figures rest on small groups (design 5 and 7, engineering 18 and 13). The impact is a team effort; causality is limited.

Charts · Rebuilt from the survey values, no screenshots.

Last updated · October 2026

Rebuild · The embedded components are rebuilds from max.lab: measured from the rendered original, written from scratch. The original code stays with Container xChange.
