Why "AI strategy" decks rarely survive contact with IT

A polished AI strategy deck, built by a consultancy that never talked to the engineering team, is one of the most reliable ways to burn six weeks and produce nothing that ships. The strategy is usually sound in the abstract and disconnected from what the existing systems can actually support.

The gap shows up in predictable places: a recommendation assuming API access to a legacy system that has none, a timeline that doesn't account for a security review process the strategy team never knew existed, or a data flow that looks simple on a slide and requires six weeks of integration work in reality.

The fix is sequencing, not more strategy work: involve whoever owns the actual systems in the room during strategy formation, not after it's finalized. A strategy shaped with technical constraints in mind from the start is slower to write and dramatically faster to actually implement.

The consultancies whose recommendations actually ship are, almost without exception, the ones who treated engineering as a stakeholder in the strategy, not a downstream implementer of a plan made without them.