The hardest decision is what to ignore

Christina Schmitt put the position bluntly on 24 August: the hardest AI decision is no longer what to build, it is what to ignore. The problem organisations report is not a shortage of ideas but an accumulation of disconnected pilots, none with a route to operational scale.

The line worth quoting to a steering committee

Weak data foundations, Schmitt argues, cause organisations to scale inconsistency rather than intelligence. That is the mechanism behind a familiar pattern: a pilot that works on a curated dataset and degrades the moment it meets the production one, then gets rolled out anyway because the pilot succeeded.

Two standards, not one

The piece separates buyer-facing AI from internal productivity tooling and holds them to different governance bars. B2B buyers assess AI on whether it is useful, not on how it was built, and a system a customer touches carries a reputational exposure an internal tool does not. Applying one governance standard across both means either over-controlling the internal work or under-controlling the external.

What is actually being recommended

• Identify which opportunities genuinely move a buyer's decision.
• Determine which can realistically scale, before funding them.
• Prioritise on customer value and measurable outcomes.
• Build the discipline to stop initiatives that are not working.

No survey figures accompany the argument, and it is worth saying so. This is an analyst's framing rather than a data release, useful mainly for the fourth item on that list, which is the one most organisations have no process for at all.