Case Study · Workflow Scoping

Scoping an AI-supported content workflow for a product documentation team

Environment Oxygen XML Editor, DITA
Role Pilot design, prompt development, UAT
Focus Workflow fit, deployment scoping

A structured pilot of AI-assisted authoring in a DITA environment. The capability was real, but the pilot's primary output was a clearer picture of what it would take to stand up the workflow responsibly inside an enterprise documentation system.

The problem

Technical writers were under mounting pressure from frequent release cycles while working in structured DITA environments with strict content and validation requirements. AI could generate useful text, and with the right prompting it could generate valid DITA. But using that output in real authoring work was still awkward: writers had to move between tools, manage copy-and-paste workflows, validate generated markup, and decide where AI assistance fit into existing review and publishing processes.

AI-assisted authoring was technically possible, and tools like Positron were designed for structured content environments. The harder question was implementation fit: what it would take to stand up the workflow responsibly inside an established Oxygen/DITA environment, given existing licenses, proxy constraints, validation requirements, prompt maintenance needs, and review workflows.

What was built

The pilot tested AI-assisted authoring against real documentation workflows, not isolated prompt output. The work included custom prompts for common documentation tasks, criteria for reviewing output quality and usability, UAT coordination with writers, feedback collection, result analysis, and reporting on capability, workflow fit, and operational readiness.

Prompt development

  • Custom prompts for summarization, short descriptions, and DITA transformation
  • Designed for repeatability across documentation tasks
  • Tested against real content with defined quality criteria

UAT and feedback

  • User acceptance testing coordinated with pilot participants
  • Feedback collected systematically across task types
  • Results analyzed for usability patterns and workflow friction

Operational assessment

  • Output quality evaluated against authoring and validation requirements
  • Licensing, maintenance, and governance requirements mapped
  • Deployment constraints surfaced and reported to stakeholders

What the pilot found

Formatting and editorial prompts — grammar correction and short description drafting — performed most reliably. UAT participants saw clear value in the direction of AI-assisted authoring, but the pilot also revealed a gap between basic prompt assistance and the level of workflow support that would create meaningful time savings.

The feedback clarified why problem-space knowledge mattered. Basic editorial prompts could support training, consistency, and less experienced writers, while seasoned writers needed workflow-aware capabilities that reduced real authoring friction. At least one manager participant recognized the training value clearly.

For experienced writers, the most valuable future direction was not an advanced editing layer. It was workflow-aware authoring support: creating new topics, inserting valid structured content, updating maps, and supporting authoring steps inside the content model. The same AI capability had different value depending on who was using it and what work they were trying to complete.

Generative tasks such as outlining and question generation scored lower and required more review. The capability was real within a defined scope, but the pilot showed that usefulness depended on task complexity, writer experience, and how deeply the tool could participate in the authoring workflow.

Key finding Confirming AI capability is not the same as confirming operational readiness for deployment.

Four constraints emerged that were independent of capability and shaped the deployment decision:

License dependency

Positron required Oxygen Enterprise licenses, creating a cost barrier before any writer could participate in the workflow.

Model instability

Underlying model updates broke existing prompts without warning, introducing silent regressions with no clear ownership path.

Proxy incompatibility

The internal AI proxy broke Positron's built-in functionality, resulting in missing features and connection errors across participant environments.

Prompt maintenance burden

Every prompt required ongoing design, testing, and maintenance — with no governance model in place for managing a prompt library at scale.

The decision

The pilot confirmed that useful AI-assisted authoring capability existed within a defined scope. The bigger finding was what sustainable deployment would require: workflow integration, prompt governance, tooling maintenance, licensing clarity, and organizational support for an ongoing evaluation cadence.

Given competing priorities and the pace of change in AI tooling, the decision was made to defer full adoption. The alternative was a lighter-weight approach — a custom AI interface built directly into the CMS, with lower overhead, clearer ownership, and a narrower surface area to maintain.

That decision came directly from the pilot's findings: not that the capability was insufficient, but that the smallest reliable system for this context was a different one. The value of the pilot was distinguishing tool capability from deployment readiness before committing to a heavier implementation path.

Related methods and patterns

Method used Smallest Reliable System Scoping A scoping framework for defining the minimum viable AI-supported workflow — clarifying what to automate, what to maintain manually, and where the automation boundary sits.
Pattern observed Workflow Bloat AI-supported workflows that expand to include more review steps, approvals, or manual correction than the original process required — common when the automation boundary is defined too broadly at the outset.