SOURCEFOLD

AI-assisted global content operations across markets

SOURCEFOLD global content operations workspace shown in light and dark modes with human review, source-change impact, and publishing-readiness panels.
SOURCEFOLD global content operations workspace shown in light and dark modes with human review, source-change impact, and publishing-readiness panels.
Type
Independent Product Design Concept
Role
Product Design · UX Strategy · Interaction Design
Timeline
2026
Focus
Enterprise UX · AI Workflows · Systems Design
Scope
Product Strategy · Information Architecture · Prototyping · Evaluation Planning

Case snapshot

Problem
Global teams lose clarity as one source becomes many market-specific states.
Ownership, review status, and version relationships become difficult to track.
Response
An AI-assisted content operations system centered on exceptions rather than duplicated workflows.
Key decision
Manage exceptions, not every variation.
Normal states stay quiet. Changes requiring human judgment rise to the surface.
Design outcome
Source relationships, ownership, review history, version context, and publishing state stay visible in one connected system.

Global content gets complicated quickly.

Complexity appears when ownership, local edits, review requirements, and publishing states begin to diverge across markets.

The challenge is maintaining control as those differences move through a distributed workflow.

Fragmented ownership

Teams need to know who owns the next decision when content moves across global and local workflows.

Version uncertainty

A source update can affect multiple markets differently, especially when local teams have already approved or edited content.

Automation without enough visibility

AI can accelerate adaptation, but teams still need to understand what changed, why it matters, and when human judgment is required.

SOURCEFOLD challenge diagramOne source campaign, Orbit Launch Campaign, branches into four markets — United States, Japan, France, and Brazil — each of which ends in a different operational state: Ready, Review Required, Source Updated, and Blocked.SOURCEOrbit Launch CampaignOne sourceMARKETUnited StatesMARKETJapanMARKETFranceMARKETBrazilReadyReview RequiredSource UpdatedBlocked

One source can create multiple operational states across markets.

The hardest moments happen between teams and tools.

I mapped the workflow from source creation through adaptation, local review, approval, and publishing to identify where visibility and ownership begin to break down.

The critical problems were not isolated screens. They were handoffs between people and system states.

SOURCEFOLD workflow diagramSwimlane diagram across Content Manager, AI/System, Market Reviewer, and Program Owner lanes: Create hands off to Adapt, which detects exceptions and hands off to Review for human judgment, whose decision updates readiness and assigns the next action back to Publish; Program Owner monitors throughout; and Track Changes after publishing can trigger a dashed Source Change return path back to Adapt.CONTENT MANAGERAI / SYSTEMMARKET REVIEWERPROGRAM OWNERSTART ADAPTHuman judgment requiredUPDATES READYNEXT ACTIONSOURCE CHANGECreateSource · Markets ·AdaptationAdaptAnalyze ·Adapt · DetectStatus · Reason · Owner ·Next ActionReviewAccept · Edit ·Revert · EscalateUpdate ReadinessDependencies ·History · OwnershipPublishExceptions →Preview → LiveTrack ChangesSource · Market ·Publication ChangesMonitor readiness · blockers · ownership · launch risk

AI generates, detects, and assigns. Humans retain judgment, escalation, approval, and publishing.

Manage exceptions, not every variation.

Global teams do not need to inspect every market equally.

They need to know what changed, why it matters, who owns the next action, and whether the issue requires human judgment.

This led to an exception-driven product model where normal states stay quiet and meaningful changes rise to the surface.

Surface attention

Prioritize unresolved exceptions instead of treating every market as equally urgent.

Explain impact

Show the downstream consequence of a change without exposing unnecessary system complexity.

Clarify ownership

Every exception has a status, reason, owner, and next action.

Preserve human decisions

AI can accelerate adaptation, but approved local work is never silently overwritten.

A content model designed for change.

Instead of treating each localization as an isolated file, I structured SOURCEFOLD around connected objects that preserve source relationships, local decisions, review history, and release state.

SOURCEFOLD object modelContent flows into Market and Variant, converging into a traceable Version; Version branches to a Review workflow above and a Publication workflow below, with a dashed change-impact signal running from a source update to Version, and a shared System Context layer of lineage, ownership, readiness, and decision history beneath the whole model.AI-ASSISTEDFLAGGEDRESOLVEDCHANGE IMPACTOBJECTContentMaster sourceOrbit Launch CampaignSource v3 → Source v4OBJECTMarketLocal operating contextJapan · JAMarket ≠ LanguageOBJECTVariantMarket-specific expressionJapan VariantOBJECTVersionTraceable content stateVariant v3 · Based on Source v4WORKFLOWReviewHuman decision workflowAI flags exceptionsWORKFLOWPublicationRelease stateApproved ≠ PublishedSYSTEM CONTEXTLineageWhere did this versioncome from?OwnershipWho owns the nextaction?ReadinessCan this market moveforward?Decision HistoryWhat was decidedpreviously?

The product model connects content lineage, human review, and release state without collapsing them into a single translation status.

A market is not simply a language.

It also carries its own review requirements, publishing conditions, accessibility needs, and local context.

This distinction allows SOURCEFOLD to treat each market as a traceable operational state instead of a static localization file.

Designing the rules before refining the interface.

The most important design work happened around system behavior: when automation should proceed, when it should stop, and how local decisions should survive future source changes.

SOURCEFOLD key decisions diagramThree stacked decision modules — AI Communication, Source Synchronization, and Publishing Model — each contrasting a rejected direction against SOURCEFOLD's chosen direction, with a one-line takeaway per decision.DECISION 01AI CommunicationHow should AI outputbe communicated?REJECTEDNumeric Confidence Score74% · 82% · 91%Looks precise, not actionableVSCHOSENActionable Review StateReady · Review Required ·Escalated · BlockedTells users what to do nextUsers need next-action clarity more than decorative AI precision.DECISION 02Source SynchronizationWhat happens when sourcechanges after approval?REJECTEDAuto Sync / OverwriteCan erase approved local workVSCHOSENHuman-Edit PreservationUpdate affected field ·Keep local version · Review manuallyPreserves approved local decisionsSource updates should trigger review, not silently replace approved local work.DECISION 03Publishing ModelHow should globalpublishing be controlled?REJECTEDAll-or-Nothing ReleaseOne blocked market delays allVSCHOSENMarket-Level ReleasePublish ready markets whileholding unresolved markets.Follows market readiness, not uniformityPublishing decisions should follow market readiness, not global uniformity.

Product Experience

I focused the product around one campaign — Orbit Launch — across four markets with different operational states.

The product experience focuses on the moments where users need to understand, decide, and resolve — not every screen in the platform.

  1. United StatesReady
  2. JapanReview Required
  3. FranceSource Updated
  4. BrazilBlocked

Know what needs attention.

The Global Workspace is organized around exceptions rather than generic dashboard metrics.

Each market exposes the information needed to act:

  • Status
  • Reason
  • Owner
  • Version relationship
  • Publishing state
  • Next action

This allows the content manager to understand campaign readiness without opening every market individually.

SOURCEFOLD Global Workspace showing Orbit Launch market exceptions, owners, versions, publishing states, and next actions
France market variant detail showing Approved and Source Updated states with an explicit next action

Design decision — 01

Normal states recede.Exceptions surface.

Give reviewers evidence, not a confidence score.

When SOURCEFOLD detects a market-specific issue that requires judgment, it pauses automation and brings the reviewer into a focused decision workspace.

The review experience separates three layers:

Source

What the original content says.

Adapted Variant

What AI proposed for the market.

Review Context

Why the system believes human judgment is required.

The reviewer can edit, approve, or escalate while the system preserves who changed what and why.

SOURCEFOLD AI Review workspace with an AI-suggested Japan adaptation awaiting human review
AI-suggested review state
AI Suggested
Human-edited review state
Human Edited
Human-approved review state
Human Approved

Design decision — 02

AI proposes.Humans decide.The system preserves the decision.

Show downstream impact before changing local work.

When the source offer changes from 20% to 25%, SOURCEFOLD does not mark every market as outdated.

Instead, the system evaluates which variants are actually affected.

  1. United StatesSafe to Update
  2. JapanNo Action Needed
  3. FranceReview Required
  4. BrazilSafe to Update

France requires attention because its approved local variant still references the previous offer.

SOURCEFOLD Source Change Impact overview showing field-level downstream effects across four markets
Expanded France source-change detail showing the promotional offer change from 20 percent to 25 percent
France impact expanded
France market variant showing the approved local edit preserved after the source update
Approved local variant preserved

Design decision — 03

Affected does not automatically mean actionable.

Preserve local decisions without losing source alignment.

When approved local content conflicts with a new source version, SOURCEFOLD makes the trade-off explicit instead of forcing synchronization.

Update affected field

Apply the source change while preserving unrelated local edits.

Keep local version

Record an intentional market exception.

Review manually

Create a new human-edited version before approval.

SOURCEFOLD Version Resolution workspace comparing the current France variant with the updated source
Current local version and updated source comparison
Current local version
Update affected field resolution selected
Updated source selected
Update affected field decision state
Update affected field
Keep Local Version decision state
Keep local version
Manual Review decision state
Review manually

Design decision — 04

Local divergence can be intentional.

The system should record it, not erase it.

Publish independently without losing global visibility.

Approval and publishing are intentionally separate.

Before release, users can see which version of each market will go live, which markets are excluded, and why.

This allows approved markets to move forward even if another market remains blocked.

SOURCEFOLD Publishing Readiness showing three of four markets ready and Brazil blocked from this release
Confirmation before publishing three eligible markets
Publish 3 markets
Publishing result with three markets live and Brazil pending
3 live · 1 pending

One system, different responsibilities.

SOURCEFOLD uses a shared content model, but the interface exposes different levels of complexity depending on the user's responsibility.

  1. Content ManagerGlobal readiness · ownership
  2. Market ReviewerAssigned decisions · supporting evidence
  3. Program OwnerLaunch risk · unresolved dependencies

One shared model, different levels of operational detail.

SOURCEFOLD Global Workspace for a content manager
Global WorkspaceCampaign-level readiness and ownership for the content manager.
SOURCEFOLD My Reviews queue for a market reviewer
My ReviewsAssigned decisions and supporting context for the market reviewer.

One content model, different interaction contexts.

The approved market variant remains consistent across devices while interaction controls adapt to each environment.

Desktop exposes more persistent context.

Mobile consolidates controls into touch-first patterns.

TV prioritizes focus states, larger targets, and reduced navigation depth.

SOURCEFOLD market preview in Desktop mode
Desktop
SOURCEFOLD market preview in Mobile mode
Mobile
SOURCEFOLD market preview in TV mode
TV

One hierarchy, two visual environments.

The interface adapts to light and dark environments while preserving the same hierarchy, operational states, and interaction patterns.

SOURCEFOLD Global Workspace in light mode
Light mode
SOURCEFOLD Global Workspace in dark mode
Dark mode

Designing the states between the screens.

SOURCEFOLD is defined not only by its primary screens, but by the operational states that connect them.

Workflow

Ready
Review Required
Source Updated
Blocked

Review / provenance

AI Suggested
Human Edited
Human Approved
Escalated

Version

Current
Affected
Needs Re-approval
Approved Exception

Publishing

Not Ready
Ready
Scheduled
Live

These states drive tables, review panels, version logic, publishing behavior, and notifications across the product.

Evaluating the system model, not visual preference.

The prototype is designed to test whether users can understand and act on SOURCEFOLD's operational states:

  • what needs attention
  • why AI paused for review
  • what a source update affects
  • whether approved local work is protected
  • which markets will actually publish

Iteration 01 — Attention hierarchy

Whether users notice what needs attention before anything else.

SOURCEFOLD campaign workspace showing the attention queue and market readiness
Exception overviewTest whether the highest-risk markets and their next actions surface before routine status.
SOURCEFOLD publishing readiness screen showing three ready markets and one held market
Release readinessTest whether ready, held, and excluded markets are distinguishable before publishing.

Iteration 02 — AI review comprehension

Whether users understand why AI paused for review and what evidence supports it.

AI-suggested adaptation awaiting human review
AI SuggestedTest whether the provisional AI state and the reason for review are recognizable.
Human-edited adaptation in the review workspace
Human EditedTest whether human ownership and edited provenance remain clear after intervention.
Human-approved adaptation in the review workspace
Human ApprovedTest whether approval reads as a separate, explicit decision rather than an automatic state change.

Iteration 03 — Source change / version resolution

Whether users understand what a source change affects and how to resolve it.

Source-change impact overview across four markets
Impact overviewTest whether users can identify which markets and fields a source update affects.
Expanded France impact state showing the affected promotional offer
Affected marketTest whether France's approved local value is visibly protected while the source changes.
Version resolution screen with Keep local version selected
Resolution decisionTest whether keeping the local version communicates an intentional exception and its next state.

From global overview to local decision.

The final prototype follows Orbit Launch through the moments where operational complexity becomes visible:

  1. Global Workspace
  2. Japan Review
  3. Human Edit & Approval
  4. France Source Change
  5. Version Resolution
  6. Publishing Readiness
  7. Cross-device Preview
  8. Partial Market Release
Explore interactive prototype ↗

Opens the complete SOURCEFOLD prototype in a new tab.

Designing for global scale means designing for change.

Design outcome

SOURCEFOLD turns a fragmented localization workflow into an exception-driven operating model. Instead of asking teams to inspect every market, the system surfaces where judgment is required while preserving source relationships, local ownership, version history, and publishing state.

What the concept demonstrates

  • Exception-driven enterprise workflows
  • AI-assisted review with human control
  • Content lineage and version awareness
  • Explicit ownership
  • Intentional local exceptions
  • Market-level publishing
  • Scalable operational states

What I learned

Automation needs boundaries.

The value of AI is not maximizing automatic changes, but reducing repetitive work while making judgment points explicit.

Versioning is a user experience problem.

Users should understand what changed and what it affects without needing to think like engineers.

Exceptions are part of the system, not failures of it.

A scalable global product needs to support intentional divergence, partial readiness, and unresolved work without losing clarity.

What I would explore next

  • Permissions and governance
  • Localization memory based on previous human decisions
  • Team-defined automation policies
  • Broader content types and channels