How Do You De-Risk Cloud and AI Programmes When Nobody Can Prove They’re Working?
Budget season has a habit of doing this to people. The slides are polished, the programme names sound impressive, and then someone on the exec team drops the question that lands like a brick through a greenhouse:
“We’ve spent a fortune on AI and cloud. What exactly have we got?”
You can almost feel the room split. Half the table thinks the hype has run away with the chequebook. The other half thinks this is a generational shift. Both sides have a point. That’s the uncomfortable bit.
And the honest answer, for most organisations, is: we’re not sure yet.
An RGP survey of 200 finance chiefs found that only 14% have seen a clear, measurable impact from their AI investments. Fourteen percent. Meanwhile, Gartner projects worldwide AI spending at £1.9 trillion in 2026, a 44% year-on-year increase (and yes, that’s trillion with a T, the kind of number that stops looking real after a while) – with generative AI model spending alone growing at over 80%. One figure from Deloitte’s survey of 1,854 executives stuck with me: satisfactory ROI takes two to four years. Only 6% saw payback in under a year. The usual expectation for tech investments? Seven to twelve months.
That’s not a rounding error. That’s a structural mismatch.
We’ve done this dance before
The tech industry has a long and colourful history of spending first and measuring later. AI isn’t the first time. It’s just the most expensive.
The dot-com era gave us companies valued on “eyeballs” – pets.com burned through £240 million in under two years. The ERP wave promised to fix the business; McKinsey found three-quarters of projects failed on time or budget, two-thirds with a negative ROI. Cloud migration? Same script. 75% over budget. The Cloud Security Alliance reported 90% of CIOs experienced failed or disrupted ERP-to-cloud migrations.
Then came “big data will make decisions for us.” Blockchain replacing databases (still waiting on that one). The metaverse replacing the office. Each one taught the same lesson: the technology was real, but the gap between installing it and getting value from it was larger than anyone budgeted for.
AI fits the pattern, with a twist. The tech is genuinely useful early. That’s what makes this cycle faster. It’s also what makes the “Now What?” question more urgent. Easy wins get extrapolated into wild forecasts. And that’s how the bubble inflates.
Which brings us to the numbers – and they’re not comforting.
Gartner predicted at least 30% of generative AI projects would be abandoned after proof of concept by end of 2025. That was conservative. Their January 2026 analysis put it at 50%. Meanwhile, 42% of companies walked away from most AI initiatives in 2025, more than double the 17% a year before. MIT and RAND place the broader failure rate between 70% and 85%.
Think about that for a second. Not a minority of projects. The majority.
Gartner has already flagged that over 40% of agentic AI projects will be cancelled by 2027. Their February 2025 data found 63% of organisations didn’t have the right data management practices for AI. Without decent foundations, you’re building on sand.
Yet investment keeps climbing. Deloitte asked nearly 2,000 executives and found nine in ten plan to increase AI spending again this year, despite all of the above. There’s a word for behaviour like that when the evidence doesn’t back it up. In polite company, we call it “strategic conviction.”
The cloud hangover nobody talks about
The AI ROI problem didn’t start with AI. It started with cloud.
Most organisations never properly sorted their cloud economics. They migrated workloads without clear cost models, without governance. Gartner forecasts public cloud spending will grow over 21% in 2026, pushing it well beyond the £542 billion spent in 2025. That’s a lot of money flowing through systems many businesses still don’t fully understand.
RGP’s research found 86% of CFOs say legacy tools present a significant barrier. That cloud migration you rushed through five years ago? It’s now the bottleneck. The technical debt? Compounding.
CloudZero’s 2025 research finds only 51% of organisations can confidently evaluate AI ROI. Without visibility, you’re flying blind. And you can’t de-risk what you can’t see.
A CFO I spoke to recently put it better than any report: “We’re not anti-AI. We’re anti-fog. Show me where the money went and what came back.” That’s the whole problem in two sentences.
So how do you actually de-risk this?
Right. Enough doom. Some organisations are getting it right.
Larridin’s State of Enterprise AI 2025 report, surveying over 1,000 leaders, found that 89% of enterprises have adopted AI tools, but only 23% can measure their return. The ones who can? 5.2x higher confidence in investments. 3.8x higher continued investment rates.
That gap is where the opportunity sits. Here’s what separates the two camps:
1) Fix the foundations before adding more load
Don’t bolt AI onto broken cloud infrastructure. Audit the existing estate. The World Economic Forum noted that hidden AI costs – data preparation, governance, change management, can rival the technology’s sticker price. Get the house in order before inviting more guests.
2) Measure across three dimensions, from day one
Larridin found that organisations measuring AI usage, proficiency, and business outcomes dramatically outperform those that don’t. Establish baselines before you deploy. If you can’t measure it, you can’t manage it. And you can’t defend it to the board.
3) Learn from the shadow AI your people are already using
This one might surprise you. The most effective AI adoption in many organisations isn’t happening through official channels. It’s happening on personal subscriptions, quietly, at desks. MIT finds workers from over 90% of surveyed companies are using AI regularly, even though only 40% have an official licence. And that informal usage often delivers better ROI than the formal stuff. So instead of fighting it, learn from it. Then formalise the bits that work.
4) Buy before you build
There’s a stat from Menlo Ventures that catches my eye: 76% of AI use cases are now purchased rather than built internally, up from 53% in 2024. And according to MIT, external partnerships achieve 66% deployment success versus 33% for internal builds. The instinct to build it yourself feels like control. But the data says it fails twice as often.
5) Invest in people like you mean it
According to RGP, 68% of CFOs rank AI skills among the biggest challenges. Larridin’s data tells the same story, formal training programmes drive 2.7x higher proficiency. The technology is only as good as the people using it, and a webinar and a wiki don’t count as investment.
6) Kill what isn’t working
This one sounds trivial. It isn’t. The bubble atmosphere grows when nothing ever dies and every initiative lives on as a “learning.” Learning is good. Zombie portfolios are not. Honestly, the bravest thing I’ve seen a CIO do was stand up in a quarterly review and kill three AI pilots in one go. It felt brutal. Six months later, the two remaining projects are actually delivering.
A quick word on timing, because it matters more than people think.
There’s an urgency here. Larridin’s survey finds 85% of senior leaders believe they have less than eighteen months to build competitive advantage before falling behind permanently. Gartner predicts 15% of day-to-day work decisions will be made autonomously through agentic AI by 2028, up from zero in 2024.
The organisations that figure out measurement and governance now will shape what comes next. The rest will join a long list that confused spending with strategy.
The uncomfortable truth (and the hopeful one)
The uncomfortable truth: AI productivity gains are real, controlled studies show 25-55% improvements, research finds frequent users saving over nine hours a week. But most organisations aren’t failing because AI doesn’t work. They’re failing because the foundations underneath it don’t.
The hopeful truth: de-risking isn’t mysterious. It’s audits, data quality, governance, skills, and measurement. It’s the plumbing, not the penthouse. But the plumbing is what stops the whole thing from flooding.
And when the foundations are right? The AI ambition stops crashing. It lands. And it delivers.
That’s the kind of value that survives the hype cycle.
𝗟𝗘𝗧’𝗦 𝗖𝗢𝗡𝗡𝗘𝗖𝗧
📩 Got a programme that’s struggling or building one from scratch? Send me a message – no obligation, no sales deck.
