AI-Ready Design System

Pronto (Havas), 2022–2026

Lead Product Designer; Figma Variables, Storybook, Figma Make

Cut QA time for visual inconsistencies by 40% and accelerated product development by 30% within six months by building Pronto’s first token-based design system.

Problem
The same action looked and behaved differently depending on which screen you were on, so users were relearning tasks as they moved through the platform.
Approach
I proposed a middle path: upgrade the highest-traffic screens inside the deadline, defer the complex workflows to a dated follow-up so they could not drift indefinitely, and codify the standard as we went.
Result
Token-based handoff replaced annotation-heavy specs, new features no longer needed debt remediation before shipping, and the refreshed priority modules landed on time.
Timeframe
6 months
Team
Product Manager, Project Manager, Business Analyst, Front-End Engineers, Back-End Engineers
The consolidated button set, slider controls and colour palette from the Pronto design system
Final design: the consolidated button set, slider controls and colour palette from the Pronto design system.

DiscoverFamiliar Actions, Relearned Every Screen

Familiar Actions, Relearned Every Screen

The same action looked and behaved differently depending on which screen you were on, so users were relearning tasks as they moved through the platform. It surfaced as bug reports and support tickets themed around visual mismatch and technical misalignment. Underneath was years of incremental development with no shared foundation between design and engineering.

Dozens of inconsistent action buttons collected from across the live product
Action buttons as they existed across the live product.

Cataloguing Discrepancies

I catalogued every duplicated pattern across the platform: colour values, spacing units, component variants. Alongside that I ran working sessions with the front-end engineers and Product on which components they avoided, where they had built workarounds instead of raising a ticket, and which areas felt fragile to touch. That is what shaped prioritisation.

Differently styled text inputs and form fields catalogued across the product
Inputs variations.
Differently structured action menus catalogued across the product
Action Menu variations.
Differently styled tags and status labels catalogued across the product
Tags variations.

The Middle Path, Not the Full Rebuild

A full rebuild would not fit the six-month runway, and the PM pushed back on a system-wide clean-up competing with roadmap delivery. I proposed a middle path: upgrade the highest-traffic screens inside the deadline, defer the complex workflows to a dated follow-up so they could not drift indefinitely, and codify the standard as we went.

Constraints

Timeframe
Six months, with a platform rename and rebrand running concurrently via an external agency.
Resources
One engineering squad, shared with other critical features across the platform.
Stakeholders
Management was committed to a full-scale redesign, setting the expectation for clients.

Course of Action

1Prioritise
Ranked screens by traffic and by the engineering cost to change them. Highest traffic + lowest risk.
2Consolidate
Reduced duplicate patterns before codifying anything, so the system captured a clean baseline.
3Constrain
Only necessary code changed, jointly scoped with engineering. CSS aligned to the rebrand.
4Codify
New standards captured in Figma and Storybook as the reference for everything that followed.

Twenty-Two Buttons Became Five

Building tokens on top of 22 divergent button variants would have frozen the inconsistency into the system, so consolidation came first. Five variants, primary through anchor, covered every state the product actually needed. Each retired variant was mapped to its replacement before anything was swapped, so engineering changed values rather than rewrote behaviour.

The five consolidated button variants, primary through anchor, in every state, over the faded collage of the buttons they replaced
Five levels of emphasis, primary through anchor.

Two Layers, One Source of Truth

Primitives hold raw values, semantics hold decisions. Background/Primary points at Color/Neutrals/0, so a rebrand changes one primitive and every surface follows. Engineering co-defined the naming convention and the governance model rather than receiving them, with Figma Variables as the design source and Storybook as the implementation spec.

Figma Variables panel with collections split into Primitives and Semantics
Variables categorised into Primitives and Semantics.
Semantic variables each pointing at a primitive colour value
Primitive values applied to semantic variables.
Selection colours panel listing the semantic tokens applied to a component
Semantic tokens used to define design decisions.

Calm at High Density

Density is the real constraint in this product, so the visual system has to do the work that decoration cannot. One type scale carries all hierarchy. Eight neutral steps build structure and three brand steps are reserved for action, which is what keeps a dense screen readable. Space, radius, stroke and control tokens set a repeating rhythm across every layout.

Text styles panel showing the single type scale
One type scale for hierarchy.
Brand and neutral colour swatches with their hex values
Eight neutral steps for structure, three brand steps for action.
A Pronto dashboard screen built with the space, radius, stroke and control tokens
Space, radius, stroke and control tokens set the visual rhythm.

Change Without Drift

The library was published with variant controls, so the fastest route to a new pattern was always an existing component. Anything genuinely new needed a full state set, documentation and a named owner before publishing. Contrast was encoded in the semantic pairs, and focus states and target sizes were defined in the components themselves.

Header component variants built from existing library components
Components built with all existing variants.
The Assets panel showing the published pattern library
Published to the design library.
Variant controls for a component in Figma's right-side panel
Variant controls in the right-side panel.

Handing Over the Keys

I ran training sessions for Product and Engineering on the new standards. Documentation and working examples meant the team could self-produce on-pattern concepts, and Figma Make prompt templates linked directly to the component library kept that output on-system. I presented the system to leadership so adoption was a decision rather than a request.

Storybook implementation of the token-based component library
Storybook implementation.

The System, Applied

The token set and component library above are the mechanism. Here's what they produce in a live product screen — Project Wizard's summary view, built entirely from published components with zero one-off styling.

Project Wizard summary view built entirely from published, on-system components

From Drift to Shared Truth

The system stopped being a source of inconsistency and became something Product and Engineering could work from without designer guidance. Token-based handoff replaced annotation-heavy specs, new features no longer needed debt remediation before shipping, and the refreshed priority modules landed on time.

  • QA time for visual inconsistencies fell by 40%.
  • Product development speed rose by 30%.

Measured as a before and after comparison across releases either side of adoption.

Learnings

Cataloguing every inconsistency before building anything took longer upfront, but it meant governance decisions were based on the full scope of the problem rather than the first few duplicates anyone noticed.

The system stuck because engineering helped define the token structure. Involving them at the architecture stage, not the implementation stage, is what made adoption real.

What I Would Do Differently

I proved the value after the fact. I would instrument component usage and QA volume from day one, so the case makes itself continuously instead of in a retrospective.

Future Planning

Extending token coverage to native components, and automated visual-regression checks against the token library so drift gets caught before it reaches QA.

Want the full walkthrough?

Next case study

The AI Timeline Wizard's Preview & Create step, with Refine Timeline and Add to Timeline actions

AI-Powered Experiences

Pronto (Havas), 2022–2026

Lead Product Designer; working prototypes and comparative testing

Lifted AI feature adoption from 4% to 32% by designing a Review–Refine human-in-the-loop framework for Pronto’s first two production AI features.

Read the case study: AI-Powered Experiences