Most AI strategies have a people, a systems, or a culture problem. And sometimes all three.

Last week I said AI investments fail when routines don't change. This week: how to tell, before you start, whether your organisation is even built to make that change.

I've been working through a framework using four dimensions SPOC: (Not Single Point of Contact), but Systems, People, Organisation, Culture: and applying it specifically to AI. It's a simple four-part diagnostic, but when you run a leadership team through it, the gaps surface fast. Here's what it reveals in practice:

Systems: Do you have a formal process to intake, evaluate, fund and kill AI experiments? Not a spreadsheet. Not a Slack or Teams channel. A structured system with ring-fenced budget and milestones that are appropriate for AI exploration, and not the same financial gates you'd use for a product launch. Most organisations don't have this. Ideas enter informally and die the same way.

People: Do you have anyone who understands both the AI capability and the business problem it's solving? This is the AI Translator role and it's the most critical, most commonly missing hire in technology teams right now. Without it, the technical team builds things the business doesn't adopt, and the business team asks for things the technical team can't safely deliver.

Organisation: Is your AI exploration structurally protected from operational demands? Or does the same team run production systems and pilot new AI capabilities? If it's the latter, exploration will always lose to keeping the lights on. IBM learned this the hard way and their solution (ring-fenced resources, separate milestones, senior sponsorship) is directly applicable to how AI teams should be structured today.

Culture: Does your organisation treat a failed AI pilot as data, or as wasted budget? If the answer is the latter, the only AI projects that will ever proceed are the safe ones. Which means no real innovation.

The four most immediate places to look when an AI programme isn't landing, aren't the model, the data, or the technology. They're these four.

Which of the four is the hardest to fix in your organisation?