21/08/2026
Matt Dyer shares a candid chat about his recent chats with business leaders.
Every business I've spoken to this year is being sold AI.
Almost none of them are being asked the question that actually matters: where does this fit?
The quick wins are real — I'm not disputing that. Ask a tool to summarise your leases or triage your maintenance requests and it will do it, this quarter, for a modest fee. That's genuinely useful.
But nobody is asking what happens when you've bought four of them.
We have done this before. A decade of SaaS gave us a system for everything, and then years of work trying to make those systems talk to each other. Most organisations are still in the middle of that project.
The tools were individually excellent and collectively a mess.
AI is heading the same way, faster.
Three questions I'd want answered before signing anything:
Who owns the IP? In most cases it sits with the developer, not you. And there's a harder follow-up: even if you did own it, where would you put it? Do you have anywhere for it to live?
What can it actually see? "We've already got AI, it's in our finance system" is the most common thing I hear. But an AI that can only see finance data is only ever going to be useful in a finance sense. The value sits in the questions that cross domains — property, operations, commercial, maintenance — and no embedded tool can answer those.
What will it cost in five years? Not what it costs today. My working assumption is more per unit, for more units. Current pricing reflects a market buying share, not a market earning margin.
The answer isn't a two-year data warehouse project. That's the false choice most pitches quietly depend on — that doing this properly means doing it slowly.
A virtual foundation goes in alongside your first AI use case or “Job to be done", not two years ahead of it.
Centralise the foundation. Distribute the AI use cases.
We spent a decade removing silos from our data. It would be a remarkable thing to spend the next decade rebuilding them in AI.
The tools are finally mature enough for a real AI strategy. Which is exactly why getting it wrong now is so expensive. Don't be sucked in by the AI 'sugar hit'.
