The AI use cases that get funded first are often the ones that demo well, not the ones with the clearest business case. That ordering usually produces a pilot that is hard to defend six months later, because it was never tied to a decision anyone was actually struggling to make.
The stronger candidates are less exciting to present. They live in a recurring decision or workflow that already has a visible owner and a measurable amount of friction — a report that takes three people two days to reconcile every month, an approval queue that backs up every quarter-end. The technology matters less than whether the friction and the owner both already exist.
This is also why data trust matters more than data volume. An organisation with modest but well-owned data can usually get more real value from a narrow AI use case than one with abundant data nobody agrees is authoritative.