A practical framework for deciding which content goes to AI-assisted workflows and which requires expert-led delivery, based on consequence rather than preference.
- Sort by consequence, not by content type
- Three tiers of consequence
- The mistake that costs the most
Sort by consequence, not by content type
Most procurement policies sort by document type — manuals here, marketing there. That produces the wrong answer often enough to be expensive, because two documents of the same type can carry completely different risk.
A more reliable test: if this document is wrong, who finds out, and what does it cost? An internal FAQ that is wrong causes a support ticket. An IFU that is wrong can cause a recall.
Three tiers of consequence
Low — errors cause mild inefficiency. Examples may include internal reference or repetitive text, after checking for safety, legal, personal-data and publication implications. AI-assisted basic QC is an option for genuinely low-consequence use.
Medium — errors cause commercial damage. Product websites, brochures, user manuals, training material. AI draft plus expert refinement.
High — errors cause regulatory, legal or safety failure. Regulatory submissions, patents, contracts, safety instructions, semiconductor datasheets. Expert-led with full QA audit, with no exceptions made for schedule pressure.
The mistake that costs the most
It is rarely over-specifying. It is routing a high-consequence document through a low-consequence workflow because it looked routine — a datasheet treated as "just a spec", a label update treated as "just a small change".
Compare review cost with plausible rework, reprint, delay and safety exposure for your own project. Set the tier by consequence at intake, and make the classification someone's explicit job.