A recent survey by Omdia reveals that nearly a third of companies report almost complete failure of their artificial intelligence proof-of-concept projects, while only nine percent manage to transition more than half of these projects into operational use. This contrasts with 46 percent of firms successfully moving over ten percent of their AI projects into production. The survey highlights a significant issue: the primary reason for the failure of these projects is not flaws in the technology itself but a lack of understanding regarding the complexities of AI deployment. Many organizations fail to clearly define business challenges that AI could address before initiating projects, leading to ineffective implementations. Additionally, only 32 percent of companies have identified specific human tasks that AI should supplement or replace, indicating a gap in strategic planning for AI integration.
A new report indicates that the use of artificial intelligence skills testing in the workplace has surged, recording a remarkable increase of 166 percent over the past year. This data, derived from 3.9 million skills tests conducted globally by TestGorilla, highlights a growing demand for proficiency in AI and coding, with other skills like coding debugging and computer literacy also seeing notable growth. Wouter Durville, CEO of TestGorilla, emphasizes that traditional hiring methods such as resumes and interviews are insufficient in revealing actual candidate capabilities. According to their 2025 State of Skills-Based Hiring report, 71 percent of employers believe skills testing is a more accurate predictor of job performance than resumes. As organizations strive to bridge the AI skills gap, the focus is shifting towards more objective assessments, with 93 percent of candidates reportedly not being questioned about their AI skills during interviews, according to Harvard Business Review.
Why do we care?
AI isn’t failing because the tech doesn’t work — it’s failing because companies don’t know what they’re trying to fix. Omdia says almost a third of AI proof-of-concepts are basically total wipeouts. And only nine percent of companies actually move half their AI projects into real-world operations. Not because the models are bad, but because nobody defined the business problem. Only a third of companies have even identified the human tasks AI should touch. That’s not innovation — that’s guesswork.
Meanwhile, hiring teams are scrambling. TestGorilla reports a 166 percent spike in AI skills testing, and employers say tests are way better than resumes at predicting performance. But here’s the kicker: 93 percent of candidates aren’t even asked about AI skills in interviews. So hiring is still operating like it’s 2018, while expectations are sitting in 2025.
The pattern is obvious: executives want AI outcomes without learning the tools. They’re not trained, they’re not involved in the workflows, and they’re making million-dollar decisions from a distance. You can’t outsource understanding — and that’s why POCs keep crashing.
This is exactly where IT service providers can shine — or stumble. Your clients want AI, but they don’t know what that means. They need someone to tell them what problems AI can actually solve, what processes need to change, and what skills their staff need before flipping any switches. If you jump straight into tools, you’re going to inherit the same failure rate Omdia is reporting.
The smart IT provider is the one who slows things down: define the problem, map the workflow, validate the human tasks, and only then talk about automation. AI advisory isn’t optional anymore — it’s the new line of business.