Ask a treasury analyst at a large multinational what their morning looks like, and a familiar pattern emerges: log into several banking portals, export balances, paste them into a spreadsheet, reconcile against the general ledger, and produce a cash position.

Walk into almost any large enterprise today and you'll find no shortage of AI ambition. Innovation teams have backlogs of use cases. Business units are running pilots. Vendors are pitching agentic workflows for every function from finance to field service. If the constraint on enterprise AI were a lack of ideas, most organizations would already be transformed.

Every Procure-to-Pay (P2P) business case looks the same on the way in: reduced processing costs, faster cycle times, better spend visibility, stronger compliance. The numbers are compelling, the software demo is clean, and the go-live date gets circled on the calendar as the moment value starts flowing.

Ask almost any knowledge worker who has adopted an AI copilot over the past year whether it's made them faster, and the answer is usually an unqualified yes. Emails draft themselves in seconds. Code gets scaffolded before the developer finishes thinking through the architecture. Reports that took an afternoon now take twenty minutes.

For the better part of two decades, procurement transformation has followed a predictable pattern: map the approval workflow, add routing logic, build escalation rules, and layer in more visibility at each stage of the chain. The implicit assumption behind almost every P2P and workflow investment has been that the way to fix procurement is to fix what happens after a request comes in.

Every boardroom today has an AI conversation. Some are evaluating copilots. Others are experimenting with agents. Many have already invested in models, platforms and pilots. The excitement is real—and so is the investment.

Ask a CFO what fragmented operations cost the business, and most will point to the visible line items: the extra headcount in shared services, the software licenses for systems that overlap, the consulting spend on the last integration project that didn't quite finish.‍Those are real costs. They're also not the biggest ones. The largest cost of fragmentation almost never appears on a budget line, because it shows up as delay, as rework, and as decisions made on incomplete information — and none of those get their own account code.‍‍

Enterprise software vendors have spent the last two years racing to put "intelligent" in front of every product name. Intelligent invoicing. Intelligent search. Intelligent forms. Each application, on its own, is genuinely smarter than the version that shipped five years ago.And yet the enterprises buying these applications aren't seeing proportionate results. The reason is simple: intelligence at the application level doesn't compound. Intelligence at the operations level does.

Every enterprise has an AI pilot running somewhere. A chatbot on the intranet. A summarization tool bolted onto email. A proof-of-concept that impressed the innovation committee and then quietly stalled.‍Few of these pilots become production systems. Fewer still show up on a P&L. The reason has nothing to do with model quality. It has everything to do with connection.‍‍