Your report example surfaces a distinction that many automation plans miss: evidence production and decision ownership are different jobs. I have seen a review process become faster after AI summarized exceptions, yet less trustworthy because nobody could reconstruct which source, rule, or override produced the final sign-off. The workflow looked simpler at the interface while its accountability became more diffuse. Would you make evidence lineage a design requirement before deciding which steps to remove, or only for workflows above a particular risk threshold?
I’d make evidence lineage a design requirement before deciding what to remove, but not insist on the same level of rigour everywhere. Even low-risk workflows benefit from knowing which source, rule or AI-generated interpretation produced an outcome. As the consequence of error increases, that lineage needs to become progressively more explicit, durable and reviewable.
The danger is exactly what you describe: a workflow can become simpler at the interface while becoming less accountable underneath. If simplification removes the ability to reconstruct how a decision was reached, it has removed part of the control structure rather than merely removing waste.
So I would separate the two questions: first, what evidence and decision ownership must remain visible; second, which steps can safely be automated or removed. Risk determines the strength of the evidence trail, but it should not determine whether one exists at all.
Your report example surfaces a distinction that many automation plans miss: evidence production and decision ownership are different jobs. I have seen a review process become faster after AI summarized exceptions, yet less trustworthy because nobody could reconstruct which source, rule, or override produced the final sign-off. The workflow looked simpler at the interface while its accountability became more diffuse. Would you make evidence lineage a design requirement before deciding which steps to remove, or only for workflows above a particular risk threshold?
I’d make evidence lineage a design requirement before deciding what to remove, but not insist on the same level of rigour everywhere. Even low-risk workflows benefit from knowing which source, rule or AI-generated interpretation produced an outcome. As the consequence of error increases, that lineage needs to become progressively more explicit, durable and reviewable.
The danger is exactly what you describe: a workflow can become simpler at the interface while becoming less accountable underneath. If simplification removes the ability to reconstruct how a decision was reached, it has removed part of the control structure rather than merely removing waste.
So I would separate the two questions: first, what evidence and decision ownership must remain visible; second, which steps can safely be automated or removed. Risk determines the strength of the evidence trail, but it should not determine whether one exists at all.