:quality(80))
Beyond the Efficiency Bubble: structured for what’s next
Part 3 of Beyond the Efficiency Bubble
There’s a slide that shows up in every enterprise AI roadmap. It says something like “scale across the organisation” and sits three quarters of the way through the deck, after the pilot results and before the budget ask. It looks like a natural next step, the kind of thing you nod at and move past. It is actually a cliff.
Because scaling AI isn’t a matter of doing more of what the pilot did. It’s a matter of whether the thing underneath the pilot can carry the weight of what comes next. And for most organisations, if they’re honest about it, the answer is that it cannot.
Parts 1 and 2 of this series named the problem. The plateau is real. The efficiency dividend is real, but only if someone decides to claim it. What nobody has said plainly enough is this: the decision to claim it is an architectural one, and most enterprises are still making it with the wrong architecture. Some of them have already tried to fix it. 68% of organisations migrated to a new CMS in the past three years, according to the State of CMS 2025 report. 57% are planning to migrate again. The platform changed. The structural problems didn’t. That scar tissue is real, and it’s sitting in every room where the next migration is being discussed.
:quality(80))
The invisible 90%
When enterprise leaders hear “AI-ready architecture,” most of them think about integrating a model. Connecting an API. Maybe fine-tuning something on their own data. That’s roughly 10% of the problem. 93% of marketers and developers say their current CMS is failing their business, according to the State of CMS 2025 report surveying 1,300 professionals. The failure isn’t the model. It’s everything around it.
The other 90% is what the model needs to be useful at scale. Structured content that AI can actually read, not the blobs of text and formatting that legacy CMS platforms store as undifferentiated HTML. Governed integrations connecting CMS, CRM, PIM, and commerce through clean APIs with clear ownership, not the fragile patchwork most enterprises built between 2015 and 2022 and have been quietly afraid to touch ever since. And embedded oversight: approval workflows, audit trails, content lineage, all built into the architecture rather than bolted on after the first compliance audit surfaces what’s been missing.
40% of CIOs in 2025 prioritised technical debt remediation specifically to enable scalable AI integration. The people closest to the infrastructure already understand that AI readiness is a foundation question, not a feature question.
And here’s the part that catches even the organisations that think they’ve solved it: having a modern platform isn’t the same as using it well. A growing number of enterprises have already moved to composable or headless architecture, made the investment, survived the migration. But the editorial teams still work the old way. AI-powered content features sit unused. Integrations with personalisation engines and analytics platforms are half-built. The platform is there. The value isn’t. Architecture is necessary, but it’s only the beginning. Without governance, adoption, and a content model designed to take advantage of what the platform can actually do, the investment underdelivers on the same promises the old one did.
:quality(80))
Architecture as the decision that matters
Here’s how you can tell whether an organisation has built for AI or bolted it on: ask what happens when something changes. In a bolted-on architecture, every change is a project. A new market requires a new site, a new language requires a new content team, and a regulatory requirement means a governance layer retrofitted by people who weren’t involved in the original build. Each addition increases complexity, and complexity increases the gap
between what AI promised and what it delivered.
In a composable architecture, change is configuration, not construction. Organisations investing in composable foundations are seeing 80% faster feature deployment compared to those on monolithic stacks, and that number isn’t about technology preferences. It’s about the accumulated cost of friction.
This matters even more now that the EU AI Act enters full enforcement for high-risk systems in August 2026. The requirements are specific: data inventory, content lineage, human oversight, risk classification, and audit-grade documentation. Non-compliance carries penalties of up to €35 million or 7% of global revenue. These aren’t requirements you meet with a policy document. They’re requirements you meet with architecture, and the window
for building that architecture before enforcement begins is measured in months.
The closed loop
Part 1 of this series started with a room going quiet. Someone asked how to scale, and nobody had an answer.
I’ve sat in that room. More than once. And what I’ve learned is that the silence isn’t confusion. The people in it usually know exactly what the problem is. They know the platform can’t carry what’s being asked of it. They know the workarounds are getting more expensive. They know the next AI initiative will hit the same wall as the last one. The silence is the gap between knowing and deciding.
And the decision is a binary one. You either build the foundation that makes AI a native capability of your organisation, or you keep patching the one that fights it. There isn’t a middle path that works, because the middle path is what most enterprises have been on for the last two years, and it’s the reason 42% of them abandoned their initiatives entirely.
The models will keep getting better, cheaper, and more capable. That trajectory is predictable. What isn’t commoditising is the ability to use them well: the architecture that turns a model’s output into something your organisation can trust, scale, and govern. That’s the actual differentiator, and it’s the part that takes the longest to build, which is precisely why it matters most.
The organisations that act on this in 2026 will set the terms for the next decade of enterprise competition. The ones that don’t will spend that decade paying the compounding cost of a decision they didn’t make when it mattered.
The foundation is the strategy. Build it now.
This is Part 3 of Beyond the Efficiency Bubble, a series on why enterprise AI stalls and how to fix the foundation.