By Nic Bowman. As the CEO of a group of amazingly talented senior technical software developers with over a decade in the trenches of enterprise systems, I've seen my share of technology rollouts that promised the earth and delivered an atlas. From the early days of cloud migrations, to low code custom builds, to the more recent scramble for artificial intelligence (AI) and AI agents, one pattern keeps coming back: the gap between ambition and execution.
Today, that gap is wider still. Recent reports point to a 95 per cent failure rate for generative AI pilots in enterprises, with 42 per cent of organisations scrapping most of their initiatives this year alone. Behind those figures sit billions of dollars of investment that have evaporated into thin air.
Skip ahead: Get a custom-built AI solution that delivers on promises
The Numbers Don't Lie
When it comes to hype vs. numbers, the numbers win.
- 95%. The MIT Media Lab's The GenAI Divide: State of AI in Business 2025 report, released in July, crystallised the current state of malaise. Despite an estimated $30-40 billion poured into generative AI by enterprises, 95 per cent of these pilots yielded no measurable business return. Often the models work fine and the integration is what fails.
- 85%. Gartner places the failure or underperformance rate more conservatively at nearly 85 per cent of AI projects failing to meet expectations.
- 60%. Gartner predicts that by 2027, 60 per cent of organisations will fail to realise the anticipated value of their AI use cases, largely because of incohesive data governance frameworks.
The RAND Corporation landmark report found that AI projects fail more often through misalignment in process, organisational structure, and expectations than through poor algorithms.
There's hope. Explore Trusted Ways to Use AI in Business
The Anatomy of AI Failure
AI projects tend to buckle under their own weight for the same handful of reasons.
Serious Data Quality Issues
The most cited culprit is data quality, or the lack of it. AI runs on data. Yet, as Gartner's survey revealed, 63 per cent of organisations either lack or are uncertain about having AI-ready data practices. Traditional data management is enough for reporting dashboards or basic analytics, but AI needs datasets that are clean, contextual, unbiased, and continuously refreshed.
Integration Challenges
Integration comes a close second. MIT's research traces much of that astonishingly high failure rate to "flawed enterprise integration," where off-the-shelf tools like ChatGPT are bolted onto legacy systems with no thought for workflows.
Forbes' deep-dive into the MIT findings points to cultural friction here: IT teams fret over performance risks, while HR grapples with adoption barriers. Meanwhile shadow AI spreads as individual users bring in unsanctioned tools, and any central strategy erodes.
What about Governance Gaps?
Governance gaps make all of this worse. AI is probabilistic, prone to hallucinations or biases that demand oversight.
The RAND report identifies insufficient governance as a primary failure mode, with projects collapsing under compliance risks or ethical lapses. The problem is most acute in regulated sectors like finance or healthcare, where the EU AI Act's "high-risk" classifications have caught many off guard with audits that small teams simply can't handle.
Skills shortages round out the quartet. PMI's 2024 blog on AI missteps estimates 70-80 per cent failure rates, partly because of a shortage of expertise, both in prompting models and in bridging AI with domain knowledge.
The MIT report notes that only 33% of internal AI builds succeed, versus 67% when external partners are used.
View AI deployment success story: AI and Digital Innovation: TTT Financial Group
Unrealistic Expectations
Finally, there are unrealistic expectations. The myth that AI slashes costs overnight keeps projects underfunded and starved. Chasing trends, whether the latest LLM or agentic systems, without objectives tied to them invites disillusionment.
The list isn't exhaustive, but the pattern is clear: AI failures are organisational rather than technological. When the groundwork is laid, the technology performs as promised.
Learn from our experts: Explaining Microsoft Copilot as a Strategic AI Advantage
The Fundamentals That Matter for Successful AI Projects
So what now? We've seen the stats and the failures. Should we just stop trying? No - we learn. At riivo, we've internalised this lesson through years of building digital transformation solutions with AI-augmented platforms.
In our view, AI is an amplifier for what already works. To use it well, organisations need to prioritise three things: a comprehensive knowledge base, sound processes, and rigorous governance.
Knowledge is King
Start with the knowledge base: a living repository of domain-specific insights, structured to feed AI models accurately. In practice, this means curating data for relevance rather than volume, so it is tagged, versioned, and enriched with metadata.
That is the opposite of the scattershot data practices dooming most pilots. A well-maintained knowledge base reduces bias and drift and keeps outputs in line with organisational reality.
Streamline Workflow Processes
Processes come next. They're the unglamorous glue that turns AI from an experiment into part of operations. We favour modular, iterative workflows over big-bang implementations, drawing on agile principles honed in software dev. We are also experts at process automation, which starts with streamlining the processes you already have.
Get Good Governance
Without governance, even the best knowledge and processes unravel. Our framework mandates ethical reviews, audit trails, and rollback mechanisms for every AI component. We align with standards like ISO 42001 for AI management, so compliance is built in from the start.
Engage with External Experts
Only a small portion of internal AI projects succeed, and the success rate doubles when external partners are brought in, so engaging outside experts is a critical step in deploying AI successfully. Look for a digital transformation company with a solid foundation in software development and business best practice.
Chart a Pragmatic Path Forward
As we close out 2025, AI feels like it's at a crossroads. The failure stats (95 per cent, 80 per cent, 42 per cent) are a call to recalibrate. At riivo, we've seen firsthand how a solid knowledge base shows up blind spots, how refined processes smooth adoption, and how governance instils confidence.
For leaders eyeing AI, my advice is simple: audit your foundations before you build. Partner with those who do this on a daily basis - successfully - and remember: technology serves strategy, not the other way round.