The pressure to "do AI" has never been higher. Boards are asking. Investors are asking. Competitors are announcing. And inside almost every organisation we work with, someone is being told to come back next week with a plan to deploy AI somewhere, ideally yesterday.
The result is predictable. Tools get bought before the problem is defined. Pilots launch on top of data that nobody trusts. Use cases get chosen because the vendor demoed well, not because the workflow is the right one to change. Eighteen months later, the organisation has a portfolio of demos that never made it past POC — and a fading appetite for the next round.
The hype trap
Hype is not a marketing problem. It is an operating problem. It pulls organisations into three traps:
- Tool-first thinking. Treating AI as a product to be procured rather than a capability to be built.
- Pilot theatre. Optimising for "we shipped something" rather than "we changed how the work runs."
- Foundation deferral. Putting data quality, integration and governance into a "phase two" that never arrives.
None of these are unique to AI. We have seen the same patterns in every wave of enterprise technology since CRM. What is different now is the cost of being wrong: AI compounds faster, the competitive gap opens wider, and the workforce implications are deeper.
You cannot scale on top of foundations you have not built. Every shortcut taken at the start has to be repaid later, with interest.
What the foundations actually are
"Foundations" is one of those words that gets used a lot and defined rarely. In practice, the foundations that determine whether AI scales come down to four things:
1. Data that is fit for AI
Not "perfect" — fit for purpose. Catalogued, classified, accessible to the right systems, and with a clear lineage. Most enterprises do not have this for more than a fraction of their estate. Until that gap closes, every AI use case will be slowed by data wrangling that nobody scoped.
2. An architecture that can integrate
AI is most valuable where it sits inside existing workflows, not next to them. That requires APIs, event streams, identity, and a vendor strategy that does not lock you into a single model provider. The organisations that get this right are buying optionality, not just capability.
3. A workforce that is being brought along
Foundations are not just technical. They are human. If the people whose work is changing do not understand why, what is in it for them, and how their role evolves, they will not adopt the tools. And without adoption, the value never lands.
4. Governance that goes faster, not slower
Good governance is not a brake. It is the difference between an AI initiative that can be defended in front of a regulator or a board, and one that can't. Build it in early, make it boring, and it becomes a competitive advantage.
The shape of foundation-first AI
The organisations that are getting lasting impact from AI tend to share a few characteristics. They have moved past tool-shopping. They have a small number of high-value use cases that are deeply integrated, not a long tail of disconnected demos. Their leadership talks about workflows and outcomes, not models and prompts. And they treat foundations as a permanent investment, not a one-off project.
That posture is harder to start, but it compounds faster. By the time the hype-driven organisations are abandoning their tenth POC, the foundation-first ones are deploying their second and third use cases on top of plumbing that already works.
— the takeaway.
Slow down to scale faster
The counter-intuitive truth about AI is that the fastest path to scaled impact almost always starts with a slower beginning. Define the strategy. Invest in the foundations. Pick the use cases that are worth doing for their own sake, not because they demo well. And then move — fast, and on rails that hold.
The hype will pass. The organisations that built foundations during it will still be there.