Natural
Language Formatting
DISCIPLINES
Product Design · UX Research · Product Strategy
METHODOLOGY
Job to be done · What if we · How Might We
PROJECT TYPE
B2C· B2B · SaaS · Localisation
16%
reduction in time spent saving translations after improving marker handling and validation feedback.
38%
fewer marker-related errors caused by missing, misplaced, or incorrectly applied formatting markers.
27%
reduction in support requests related to marker validation, saving issues, and publishing blockers.
CONTEXT
What you see is
what you mean
Adoption of Atlassian cloud was slower than expected. Some customers stayed on Data Centre, others had migrated and churned back. Churn to competitors was becoming a real risk. The strategy shifted. Accelerating the customer journey to cloud became a company-wide priority.
Enterprise customers were already operating in a setup we called a hybrid environment: the same customer using Atlassian Data Centre products (self-hosted) alongside Atlassian cloud products, often combined with tools from other vendors.
Atlassian already had the tools to cover the entire product development cycle, but they didn't talk to each other. If tools could pass data freely and securely, customers would have more reason to stay within the ecosystem and move more of their work to cloud on their own terms. This led us to the Connection Hub, a dedicated space where admins can create and manage integrations between Atlassian products.
To build the case, we got specific. We mapped every stage, every tool, every archetype. One role appeared throughout: the Project Manager - present from idea to delivery.
PROBLEM
What was broken?
Project managers in hybrid environments had no single place to track work. They navigated between tools and deployments, copying data by hand and increasing the probability of errors with every update.
PERSONA
Who did we design for?
A project manager responsible for tracking progress across tools, teams and environments. The person who needs the full picture to keep things moving and who feels it most when the data is scattered.
HYPOTHESIS
What did we believe?
If Atlassian tools across Data Centre and Cloud could share data automatically, project managers would stop losing time to manual transfers and have less reason to reach for tools outside the Atlassian ecosystem.
The translation was never the hard part. Keeping track of what was done, and preserving the structure around it, was.
THE MARKER LAYER
Colour codes, made readable
Before, every piece of formatting was a coloured tag laid over the word - and one word could be several colours at once. You ended up reading the colours, not the sentence. After, formatting is simply rendered the way every text editor shows it: bold looks bold, italic looks italic, a link is underlined. Nothing to decode.
THE EDITOR IN CONTEXT
Original, translation and markers in one view
The editor keeps the source and the translation side by side. Markers and links stay attached to the text as the translator works, the panel lists every marker and link in the sentence, and a live status shows what is done and what is ready to publish.
PROBLEM
Markers were shown as random colour codes layered over the text. A single word could carry several at once - bold, italic and a link - so translators decoded formatting instead of translating, and broke it when they got it wrong.
DECISION
Show formatting as itself. Each marker became a readable, labelled type with a required-or-optional distinction and a keyboard shortcut, and links were listed separately in the sentence - so structure is visible without decoding anything.
OUTCOME
Translators stopped decoding and focused on the language. Formatting stayed intact through edits, and marker-related errors dropped 38% - the structure was finally visible without getting in the way.
UPDATED CONTENT
New content that announces itself
For the first time, translators knew a segment held new content the moment they opened the document - no guessing, and no re-approving every segment just to keep track of what had changed. Whenever content changes in WordPress, the segment's status indicator changes too, making new content easier to find and validate.
ERROR STATE
Instant error validation
Closing a segment now validates it on the spot. If something is wrong, its status indicator changes to an error state, and hovering the row reveals exactly what. For the first time, translators could catch errors the moment they happened, right in context.
Learnings & Challenges
How much friction lived in the gap between progress and readiness. A translation could look finished and still be unpublishable, and users had no way to see why. Closing that gap mattered more than any single screen.
Keeping technical constraints honest and visible without overwhelming people who think in language, not code. The answer was to surface markers and validation as guidance, at the exact moment they mattered, rather than as rules to memorise.
Narrowing the scope. Instead of redesigning all of ATE, the project targeted four fragile points: markers, statuses, errors and publishing. The focus is what made the impact legible.
● LET'S WORK TOGETHER
Got a project?
Let's talk.
andpon - Andrzej Poniatowski
© 2026