There’s a version of the frontline AI story that goes like this: buy the AI, train it on some documents, give workers access, and it starts answering questions. Simple, fast, transformative.
That version skips three legs of the stool. And without all three, the AI doesn’t work, not in any lasting way.
The three legs are context, delivery, and instrumentation.
Context. The AI answers from what it knows. What it knows comes from the knowledge base you provide. If that knowledge base is a pile of outdated PDFs, department-specific silos, and five different versions of the same policy that nobody reconciled, the AI answers confidently from chaos. Garbage in, garbage out isn’t a new principle. It just has higher stakes when the person asking is a nurse checking a medication procedure or a ramp agent asking whether it’s safe to push back an aircraft. The precondition for useful AI is a real communication system of record. Accurate, current, permissioned, organized. The AI rides on top of that. Skip the record and you’ve given the AI a foundation that isn’t there.
Delivery. Even a perfectly configured AI does nothing if workers don’t engage with it. And frontline workers aren’t desk workers, they don’t have inboxes, they’re not sitting at a portal waiting for a notification, and they’re not downloading a new app because IT sent an email about it. Getting AI to the frontline requires a channel the workforce already uses, with an identity system that knows who each worker is without an email address, and permissions that reflect their actual role right now, not who they were at onboarding. Distribution is the hard part of frontline AI. The model is the cheap part.
Instrumentation. Is it working? Without visibility into what questions are being asked, what’s being answered from the knowledge base versus falling back, which parts of the workforce are engaging and which aren’t, you’re flying blind. The intelligence that makes AI useful also generates the data that tells you whether it’s useful. Both halves matter.
Organizations that skip the system-of-record step are locking themselves out of the AI era, not because AI won’t work for them in theory, but because the AI has nothing real to work with. The bolt-on version produces a product demo, not a production system.
Three legs. You need all of them.