Christian F. Rojas
ENES
All cases

Case 03

Quality governance for AI-assisted documentation

My team started using AI to draft, translate and review, and nobody logged the errors it introduced. I set up checkpoints, an error log and two roles so those errors change the prompts.

3
checkpoints, one per stage: draft, translation, pre-export review
5
workbook tabs: instructions, three stage checklists and an error log
2
new roles: TW Reviewer and TW Prompt Owner
System diagram: the quality loop for AI-assisted documentationAn AI draft, produced from prompts, passes 3 checkpoints, each with a checklist, before the approved output. The TW Reviewer applies the checks. The checkpoints record what they catch in an error log. The log feeds the TW Prompt Owner, highlighted, who adjusts the prompts for the next draft and closes the loop. The system is in production.AI draftfrom promptsTW Reviewer3 checkpointsa checklist eachapproved outputerror logwhat checks catchTW Prompt Ownercloses the loopappliesrecordsfeedsadjusts promptsin productionSystem diagram: the quality loop for AI-assisted documentationAn AI draft, produced from prompts, passes 3 checkpoints, each with a checklist, before the approved output. The TW Reviewer applies the checks. The checkpoints record what they catch in an error log. The log feeds the TW Prompt Owner, highlighted, who adjusts the prompts for the next draft and closes the loop. The system is in production.AI draftfrom prompts3 checkpointsapplied by TW Reviewererror logwhat checks catchapproved outputTW Prompt Ownercloses the loopfeedsadjusts promptsin production
Fig. 03 · System diagram

My decisions 04

  • Limit the scope to the three points where AI was already in use; no new processes.
  • Use a tabbed workbook for daily work and a shared sheet for the error log.
  • Zero ambiguity and no unnecessary complexity, because the team is new to this kind of workflow.
  • Separate the person who applies the checks from the person who owns the prompts.

Assisted execution 01

  • Claude built the workbook: formulas, validations and conditional formatting.

Context

I lead a team of two technical writers. We had started using AI at three points in our workflow: drafting from Product requirements, translating between Spanish and English, and reviewing before exporting to PDF.

At each point AI could introduce errors: lost styles, missing sections, untranslated text, inconsistencies. Review was informal, and nobody recorded those errors.

What I decided

I kept the scope to the three points that already existed. I did not want to add process to a team that was still learning to work with AI. For the same reason I asked for zero ambiguity and nothing more complex than needed.

A checklist alone is a filter. It stops an error once, and the same error can come back on the next run. So I split the work into two roles: one applies the checks, the other owns what the checks find.

How it works

  • A five-tab workbook. One tab of instructions: purpose, roles, steps, the standard sections that must be preserved and a six-step guide to reporting an error.
  • Three checklists, one per stage. Items by category, answered yes, no or not applicable, with notes. Each one has a block of “prompt signals” that shows when an error comes from the prompt and not from the source.
  • An error log. Validated fields, a summary by status and conditional formatting by status and impact.
  • TW Reviewer. Applies the checklists.
  • TW Prompt Owner. Watches the log and adjusts the prompts when an error recurs.

The loop is log → recurring pattern → prompt change, so the next run starts better.

What it produced

Before, what AI produced got an informal review. Now each stage has its checklist, and recurring errors lead to prompt changes. When HR asked me to describe my role, I noted that this error log is the only thing my area records formally today.

What it does not prove yet

The system is in use, but I have not measured its effect. I have no count of errors per stage or per month, no first-attempt pass rate and no count of recurring errors that led to a prompt change.

What I would do next

Pull those three figures from the error log every month. If the prompt changes work, the same errors should appear less often over time; if they do not, the log will show it.