How do you measure the ROI of AI in marketing?

The usual trap is measuring output (pieces published, drafts generated) instead of outcomes (traffic, conversions, cost per outcome against the baseline). The second trap is the denominator: industry surveys have shown for years that most companies cannot attribute bottom-line impact to their AI spend, precisely because they don't log supervision cost and never defined the baseline before starting.

This lab's practice: every system is documented with its full cost and dated output before claiming anything, as in the y8y case, which publishes costs and states explicitly which metrics cannot yet be claimed. The whole framework, in AI systems.