Owen Lovelock
Business Development Director – Integres Software Solutions
The uncomfortable numbers behind AI’s investment problem
There’s a statistic that’s been sitting with me lately.
According to RAND Corporation, 80% of AI projects fail to deliver their intended business value. Not 20%. Not even 50%. Eight in ten.
And it gets more pointed. S&P Global found that 42% of companies abandoned most of their AI initiatives in 2025 – up from just 17% the year before. That’s not a blip. That’s a pattern.
The instinct, when you see numbers like that, is to look at the technology. Was the model good enough? Was the data clean? Did the team have the right skills? Those are reasonable questions. But I don’t think they get to the root of it.
Here’s the thing that keeps coming up in my conversations with technology and finance leaders: most AI projects don’t fail because the technology didn’t work. They fail because the organisation lost confidence in the investment. They lost confidence because they couldn’t see the economics clearly enough to know whether to keep going or if it was (showing signs of) delivering business value impact.
The visibility gap nobody’s talking about
Only 63% of organisations currently track their AI spend in any meaningful way. That means more than a third of businesses investing in AI can’t actually see what they’re spending. And of those that do track it, only 51% say they can confidently evaluate whether those investments are delivering returns.
Think about that for a moment. You wouldn’t run a marketing campaign without knowing your cost per acquisition. You wouldn’t expand a product line without understanding the margin. But somehow, with AI – where the cost profile is genuinely complex and often large – many organisations are making multi-million-pound investment decisions with a fraction of the financial visibility they’d expect in any other context.
What’s driving the cost you can’t see
The average monthly enterprise AI spend is now over £67,000 and rising by around 36% year on year. But the number most organisations see, the line items in the cloud bill or the SaaS licences, is only part of the picture.
The real total cost of an AI project spans compute for training and inference, data engineering pipelines, storage at scale, MLOps tooling, model monitoring and retraining cycles, and the engineering and data science time that rarely gets attributed back to specific projects. Spread across cloud accounts, Kubernetes clusters, on-premise infrastructure and third-party APIs, that cost is genuinely hard to aggregate – not because organisations are being careless, but because the tooling and processes to do it haven’t historically been in place.
The result is a cost profile that’s systematically underestimated at the start and poorly tracked throughout. And when budget cycles come around, projects that can’t articulate their financial performance clearly are vulnerable, whether they deserve to be or not.
The zombie project problem
There’s a failure mode that doesn’t make it into the statistics but is probably familiar to anyone who’s worked in a technology organisation: the project that’s been quietly consuming budget for eighteen months, not because anyone believes in it anymore, but because the cost of stopping it feels hard to justify when the true spend is unclear.
These zombie projects are expensive. Not just in direct cost, but in the engineering bandwidth and leadership attention they consume. And they’re largely a product of poor financial visibility. If you can see clearly what something is costing and what it’s returning, the decision to stop becomes straightforward. Without that visibility, inertia wins.
The case for looking clearly
None of this is an argument against investing in AI. Quite the opposite. The organisations that are getting the most from AI investment aren’t necessarily the ones with the biggest budgets; they’re the ones with the clearest view of what’s working and why.
That means building the financial discipline to see the real cost of AI projects: not just the headline numbers, but the full picture across infrastructure, people and platform. And it means creating the rhythm and culture to use that information to make faster decisions, deploy capital more precisely, and stop things that aren’t working before they become expensive lessons.
In the next part of this series, I’ll look at why AI cost visibility is specifically harder than traditional IT and what makes the economics of AI infrastructure so different.
In the meantime, if you would like to request more information about how Integres Software Solutions can help your organisation achieve greater value from your AI and Cloud investments through transparency, optimisation and automation, please submit the form below or email contact@techstories.ai. You can also register your interest in attending our June event exploring strategies for AI projects here.
