I was asked to give a talk in early 2026 on AI transformation and its commercial viability. It was a brilliant conversation. The room was engaged, optimistic, curious. And rightly so. AI is already changing how we work, how productive we can be and how quickly ideas turn into output. I use it every day. Most of us do. Let’s be honest — very few leaders are drafting long documents from scratch anymore.
With thoughtful prompting and human judgement, AI delivers the bulk of what we need at speed. It is genuinely game-changing.
But that wasn’t the conclusion of the talk.
What I didn’t say was: we’re ready. Let’s roll AI out end-to-end across the business and expect transformation to follow.
Because we’re not.
AI capability is racing ahead of organizational readiness
The uncomfortable truth is that AI is still unreliable for end-to-end business transformation. Hallucinations. Fabricated references. Confident nonsense. Anyone who has used large language models seriously has seen it. This isn’t a criticism — it’s the nature of probabilistic systems trained on imperfect data.
LLMs feel intelligent, but they only know what they know. Additional training helps, but it is still bound by the quality, structure and governance of the data you give them. We are back to an old rule, amplified at scale: garbage in, garbage out.
That gap between AI ambition and data reality is where many organizations are struggling. As Chantal Hannell, IT Director at Weightmans, a UK law firm, put it, “Data is where AI ambition most visibly breaks down. Without strong data governance, organizations struggle to adopt and extract value from new technologies.”
Today, this is manageable. We are deploying AI tactically — to specific problems, in contained parts of the organization. We are seeing early results. Productivity gains. Faster analysis. Better decision support.
But here’s the rub.
Every tactical AI implementation is being layered onto estates that are already weighed down by years of technical and data debt.
Technical debt compounds like financial debt
Most organizations have been through multiple waves of digital transformation. Customer-facing platforms. SaaS adoption. Integration layers stitched together under pressure. Then COVID added another layer — technology implemented at speed simply to keep the business alive.
Much of that estate has never been consolidated. Shadow IT still exists. Business teams still buy tools and ask later, “Can you integrate this?” CIOs know this pattern all too well. It’s hard to say no, even when we should — a tension many CIOs openly describe as they try to balance pressure to adopt emerging technology with the drag of legacy estates.
The result is an estate that works, but only just. Expensive to run. Hard to change. Fragile under pressure.
Technical debt behaves exactly like financial debt. There is the principal — the work required to fix foundations. And there is the interest — the ongoing cost paid through duplicated spend, cloud sprawl, slow delivery, security exposure and exhausted teams.
As Hannell observed, while duplicated spend can often be measured, “the hidden impact is on agility — our ability to adopt and attain value from emerging products and technologies.” That loss of agility is far harder to quantify, but far more damaging over time.
Independent commentary has made this point repeatedly: deferring remediation doesn’t avoid cost, it multiplies it over time — the so-called fix-it-later mindset simply hides the bill.
AI doesn’t remove this debt. It accelerates it.
Data is the real constraint
And then there is the data.
Data quality, ownership, governance, lineage — this is where most AI ambition quietly stalls. This is not solely the CIO’s fault, but it is absolutely our responsibility.
Most data belongs to the business. And business teams are under the same pressures as IT: fewer people, tighter budgets, more demand. Critical data is prioritised. Everything else degrades until an urgent request appears.
Hannell also pointed to the long-neglected fundamentals: “Information architecture, taxonomy and sensible naming conventions are not new ideas — but they are critical if organizations want to trust and use their data effectively.”
AI amplifies this reality. Poor data does not just produce poor outputs — it produces confident, scalable wrongness.
This is why many AI initiatives plateau. Not because the models are weak, but because the foundations are.
Industry voices are increasingly clear on this point: as AI becomes more autonomous, unresolved technical and data debt magnify risk rather than value, particularly as organizations push towards more agentic AI.
Why CFOs matter more than ever
Clearing debt requires investment. That’s where the tension sits.
We are operating in a tight economic climate. Growth is slow. Margins are under pressure. Capital is constrained. CFOs quite rightly hold the purse strings for additional investment, and large, multi-year transformation programmes are a hard sell right now.
That reality is stark for many organizations. Nicolas Raynaud, CFO at the Science Museum Group, described how technology investment decisions are shaped by uncertainty around returns: “We have not taken on debt for technology investments where the ROI evidence was not obvious.” Even where efficiencies are expected, “we are not ready to assume cashable savings that could service debt repayments.”
This article is not an argument for big-bang transformation. In many organizations, that would be irresponsible.
It is an argument for targeted, tactical investment — now — to prepare for what is coming.
Raynaud highlighted how decisions to defer technology investment often resurface as unavoidable cost. Choices made during COVID to delay laptop replacement and network upgrades are “now coming back to haunt us… and eat up most of the funds available for technology investment.”
Just as with financial debt, you don’t wait for perfect conditions to stop interest compounding. You make deliberate choices to reduce exposure, create headroom and protect future optionality.
That means simplifying core platforms. Reducing integration sprawl. Investing in data quality and governance. Making fewer things do more.
These are not vanity projects. They are balance-sheet protection.
Leadership means preparing before the wave hits
This isn’t about knowing what to do. Every CIO already knows their organization is carrying too much technical and data debt.
The real problem is that most organizations never finish transformations. They box them up, move on and then put the next initiative on top. Over time, complexity hardens into layers that no one feels empowered to undo.
AI risks becoming the next layer.
But it doesn’t have to.
AI can be more than another system to integrate. Used deliberately, it can help expose where complexity is killing flow, where data can’t be trusted and where teams are spending time working around broken foundations. It can shine a light into places leaders have historically struggled to see.
More importantly, AI permits leaders to act.
Boards are excited. Users are engaged. Momentum exists. That momentum can either be wasted or used to finally tackle the unglamorous work that has been avoided for years.
Yes to AI. And because of that, yes to simplifying platforms, fixing data and paying down debt deliberately.
AI is not the solution to technical debt. But it might be the reason we finally do something about it.
The AI tsunami is building.
Get ready.
This article is published as part of the Foundry Expert Contributor Network.
Want to join?