The LLM Wiki: A Pattern for Smarter Organisational Knowledge Bases
Andrej Karpathy's 'LLM Wiki' pattern offers a compelling alternative to retrieval-augmented generation. Instead of re-deriving insights from …
Read insight
All insights
Why Most AI Projects Fail and How to Beat the Odds
·
3 min read
ai-project-failureai-strategytacit-knowledgeai-implementation
Around 80% of AI projects fail to deliver business value. We break down the three root causes, including the tacit knowledge problem most teams miss.
Why AI Projects Fail: Three Root Causes
The numbers are stark: research consistently shows that around 80% of AI projects fail to deliver meaningful business value. Not because the technology doesn’t work, but because organisations approach AI the wrong way.
80%
of AI projects fail to deliver meaningful business value.
Research consensus across recent industry studies.
After working across dozens of AI initiatives, we’ve identified three root causes that explain the vast majority of failures.
The three root causes
Where most AI initiatives quietly come undone.
- Starting with technology. Falling in love with capability before knowing whether the underlying problem matters.
- Ignoring the humans. Treating expert judgement and tacit knowledge as something to optimise around, rather than design with.
- Boiling the ocean. Attempting end-to-end transformation in one big-bang programme, instead of proving value incrementally.
Root Cause 1: Starting with Technology
The most common mistake is falling in love with a technology and then looking for a problem to solve. Teams hear about large language models, computer vision, or predictive analytics and immediately start building, without asking whether the problem they’re solving actually matters.
The fix: Start with a business problem that’s costing you real money or creating real friction. Quantify it. Only then ask whether AI is the right tool.
Root Cause 2: Ignoring the Humans
AI systems don’t operate in a vacuum. They need people to trust them, interpret their outputs, and act on their recommendations. Yet most projects treat the human element as an afterthought.
The problem runs deeper than user adoption. The assumption behind most AI projects is that the humans in the process are executing a workflow that can be mapped, modelled, and automated. But in most organisations, the most experienced people are doing something harder than that: they are applying knowledge that was never written down, drawing on pattern recognition built over years, and making judgements that no process document has ever fully captured. This is tacit knowledge, and when AI is designed without accounting for it, the system ends up optimising around the very expertise that made the process work in the first place.
The fix: Design for the human in the loop from day one. Understand their workflow, their expertise, and their concerns. Build systems that augment their judgement rather than replacing it.
Root Cause 3: Boiling the Ocean
Ambitious scope kills more projects than bad algorithms. Organisations try to build an end-to-end AI platform when they should be proving value with a focused use case.
The fix: Start small. Pick one well-defined problem, prove it works, measure the impact, then expand. The best AI programmes are built iteratively.
A Framework for Success
We use a simple three-step framework with our clients:
From discovery to scale
A measured path that pays off where 80% of programmes don't.
- Discover. Map expertise, identify friction, quantify the opportunity worth solving.
- Prove. Build a focused proof of value in 6 to 8 weeks, against a real workflow.
- Scale. Expand what earns trust and outcomes, sunset what does not.
It’s not glamorous, but it works. And in a field where 80% of projects fail, “it works” is a competitive advantage.
If you’re at the start of an AI programme and want to get the foundations right, a fixed-fee advisory session can help you plan what comes next.
Want to discuss how this framework could apply to your organisation? Get in touch.
FAQ
Frequently asked questions
Why do most AI projects fail to deliver business value?
Research consistently shows that around 80% of AI projects fail to deliver meaningful business value. The three most common causes are: starting with technology rather than a business problem, ignoring the humans whose expertise the system needs to work with, including the tacit knowledge that drives how decisions are actually made, and attempting too much scope at once. Addressing all three from the outset significantly improves the odds of success.
What is the most effective way to prevent a failing AI project?
The most effective approach is a three-step framework: discover (map expertise and quantify the problem), prove (build a focused proof of value in 6–8 weeks), then scale what works. Organisations that start with a well-defined problem, design around their people's expertise, and prove value incrementally consistently outperform those that attempt large-scale transformation from the start.
What role does tacit knowledge play in AI project failure?
Tacit knowledge (the expertise that experienced people carry in their heads but never write down) is one of the most overlooked factors in AI project failure. When AI is designed without accounting for this knowledge, it optimises around the very expertise that made the process work in the first place. Systems built with an understanding of tacit knowledge are more likely to be trusted and used by the people they are designed to support.
Keep reading
Related insights
AI for Knowledge Transfer: The Opportunity Most Organisations Miss
AI is widely used to process existing information. Its potential to capture expertise that was never documented in the first place is …
Read insightAI Doesn't Just Miss Tacit Knowledge. It Can Destroy It.
Most AI is designed to optimise tasks. But when it's built without reference to the expertise behind those tasks, it doesn't just fail to …
Read insightWhere this goes
Follow the thinking as it turns into products.
The ideas in this writing are becoming products. The waitlist is the earliest way in. For a deeper conversation about your situation, we hold a small number of fixed-fee advisory sessions each month.
New insights are shared on LinkedIn as they publish.