Two default paths exist for enterprises starting an AI initiative. Hire a consulting firm, get deep expertise, pay for hours that don't compound — the same discovery questions get asked and answered from scratch on every engagement, because that's how the billing model works. Or buy a SaaS tool, get a reusable product, and hope the vendor understands your industry's compliance requirements well enough that you don't have to explain them from zero.
Neither path is wrong, exactly. They're just optimized for different things. Consulting optimizes for depth per engagement. Software optimizes for reuse across customers. The gap between them is where a lot of AI initiatives quietly die — not from bad ideas, but from paying consulting prices for work that should have been a repeatable process, or getting generic software recommendations that ignore what actually matters in a regulated industry.
The interesting question isn't which model wins. It's whether the knowledge captured in a consulting engagement — the questions that matter, the patterns that recur across banking or healthcare or manufacturing — can be encoded once and reused, without losing the judgment that made the consulting valuable in the first place. That's a harder problem than picking a side, but it's the one worth solving.