AI workflow feedback loop
Every editor review turns into fixes, so each article comes out better than the last.
The problem
Editors fixed AI drafts by hand, and their feedback stayed in the doc. The same problems came back run after run, because the fixes never reached the workflow or the brand kits.
What I did
- Designed the data model: three linked databases for raw feedback, sorted items, and fixes.
- Built automatic capture that saves each article's feedback, with the workflow version stamped on it.
- Built the AI step that sorts each change, checks it against the brand kit and the workflow, and groups related items into root-cause fixes.
- Sent each fix to its owner: brand kit fixes to the content team, workflow fixes to the AI team.
The result
- First run: 6 articles → 37 feedback items → 18 fixes, 11 of them high confidence
- All 16 fixes from the lead's review were shipped
- The workflows now improve with every review instead of repeating the same mistakes
How it works
