A Measured Approach
to AI in UX Writing
At Evoke, we decided to start small with AI—focusing first on UX writing rather than trying to automate everything at once. This let us test the approach and build confidence before expanding further.
Our team of 25 designers was spending 8-12 hours per project writing UX copy—drafting text, revising it, and making sure it matched our brand guidelines. This careful process created good work, but it limited us to completing just 24 user journeys each year, which wasn't enough to meet demand.
As we grew, keeping everything consistent with our Helix design system required constant attention from our most experienced designers. And without a clear way to track what was working, we couldn't easily learn from our projects or improve our approach.
Where We Started (V1 Numbers)
Before AI, here's what our situation looked like:
Team: 25 designers
Cost per designer: £60,000 per year
Total yearly cost: £1.5 million
Projects completed: 24 user journeys per year (about 2 per month)
Time spent on each writing task: 8-12 hours
Quality checks: Done manually by senior designers, taking up 15% of their time
Building the Solution
In early 2024, we built a custom AI tool to help with UX writing. We had three clear goals:
Cut writing time by 40% without sacrificing quality
Complete twice as many user journeys—48 instead of 24—within a year
Save enough money to justify the investment within 18 months
The tool handles the repetitive parts of writing while designers stay in control of the creative decisions. This frees them up to think about bigger picture problems instead of spending hours on every piece of text.
What Changed After One Year (V1 Numbers)
We hit our targets and then some:
Writing speed: 40% faster (exactly what we aimed for)
Designer output: Each person could do 40% more work
Overall improvement: Between 10-36% more productive, depending on the project
Money saved: Between £150,000 and £690,000 per year
Revenue: Grew from £1.5 million to £2.19 million in the best scenarios
Delivery: Now completing 48 user journeys per year (double what we did before)
Speed to launch: 35% faster for new features
Version 2: Late 2025
After seeing the first version work well, we rebuilt the tool with better features:
If you would like to know more about the Version 1, please check this article
What I worked on:
Built the interface: Created the entire user interface over 6 weeks
Connected our design system: Made sure the tool understood all our design rules, styles, and colors
Added more AI options: Integrated three different AI models (OpenAI, Google's Gemini, and Claude) instead of just one
Created reusable parts: Built 24 interface components that work consistently across the tool
Made it fast: Got response times under 2 seconds for almost all requests
New features we added:
Three writing styles: Supportive, Directive, or Motivational—designers can pick what fits their project
Easier to learn: New designers can now start using it in 2 hours instead of 2 days
Better experience: The team rated it 4.6 out of 5 in satisfactionName here
Design System MCP
Teaching the AI to speak our design language
Using Cursor, I built a direct connection between our Helix Design System in Figma and the AI tool. This connection pulls all our design rules—colors, fonts, spacing, component names—and feeds them to the AI in a way it can understand and use. Now when the tool generates copy, it automatically follows our design standards without anyone needing to double-check it manually. The AI knows which colors we use, how our typography works, and what our components are called, so everything it creates fits seamlessly into our existing design work.
Design System MCP
Teaching the AI to speak our design language
Using Cursor, I built a direct connection between our Helix Design System in Figma and the AI tool. This connection pulls all our design rules—colors, fonts, spacing, component names—and feeds them to the AI in a way it can understand and use. Now when the tool generates copy, it automatically follows our design standards without anyone needing to double-check it manually. The AI knows which colors we use, how our typography works, and what our components are called, so everything it creates fits seamlessly into our existing design work.
What's Next: Q1 2026
We're planning to expand the tool into a complete design resource:
Our goals:
Make our entire Helix design system searchable through the tool
Let designers ask questions in plain English to find components, patterns, and guidelines
Cut the time spent looking for design information by 60%
Get 90% of designers using it regularly within 3 months
Support 12+ types of content, not just the 3 we have now
This will make the tool the go-to place for everything design-related, helping us work faster and more consistently.
