Productivity that moves downstream

Forrester has published an argument that deserves attention beyond marketing departments: AI can raise an individual employee's output while lowering the organisation's, and the two are routinely measured as if only the first existed.

The framing comes from a warning by Shopify's chief executive against what he called AI slop grenades — generating something quickly and shifting the work of reviewing, interpreting or fixing it onto a colleague. The arithmetic Forrester applies to it is blunt. If AI saves one employee twenty minutes and creates thirty minutes of work downstream for others, that is not a productivity gain.

The measurement problem is the substance. First-phase adoption emphasised individual productivity, which is the level at which the saving is visible and the cost is not. The cost lands on a different team, in a different week, attributed to a different process.

On where organisations actually are, the report cited in the post finds 66% of marketing leaders say their organisations are still encouraging experimentation while they work out the right mix of AI capabilities. Forrester does not treat that as failure — experimentation is necessary — but notes that experimentation without norms pushes the cost of learning onto someone else.

The recommendation is governance of an unglamorous kind: establish when AI should be used, what quality threshold its output must meet, when human review is required, and who remains accountable at each stage. Tolerating imperfect learning is not the same as tolerating unfinished work being transferred to a colleague.