The usual automation business case counts labour saved against system cost. The authors argue four consequential categories sit outside that sum: tacit knowledge quietly disappearing, resilience declining, regulatory exposure growing, and the social fabric of an institution degrading.

What those share is that they are unpriced. They show up years later as an organisation that cannot cope with an unusual case because nobody remaining has handled one, and none of it appears in the spreadsheet that justified the decision.

PHP-AIO is a five-gate sequential protocol with a composite final check, assessing these risks per role and producing an auditable record. Working at role level is the right granularity, since automation rarely removes a whole job and often removes the parts through which people learned to do the rest. The framing is also unusually direct for this literature: the question is when not to automate, which presumes a real answer sometimes is not to.