AI PoCs to production: a balanced perspective

The efficiency and value of AI Proof of Concept (PoCs) projects and pilots are the subject of heated debate, fuelled by reports claiming that only a handful make it to production. This has been interpreted as proof that generative AI is failing to scale, that stalled PoCs mean AI cannot deliver against investment drivers, and worse still, that we are heading for an AI bubble. However, Omdia's 2025 AI Market Maturity Survey* reveals a more nuanced, balanced prognosis for AI POCs, and what this means for AI. 

Omdia's survey reveals enterprises are actively engaged with AI PoCs. Most are drawn to conservative to moderate experimentation, with 40% running 6-20 PoCs, while 18% operate 21-50 PoCs, showing broader AI commitment across multiple use cases. Only 4% of enterprises run over 100 PoCs, with the most intense activity at this level dominated by the largest companies in the survey. This is not surprising, as PoCs at this scale are resource-intensive and expensive, and even more so for pilots. Twenty-one percent of enterprises in the survey run fewer than 5 PoCs, which could reflect early-stage adoption or lack of resources. It is notable that 43% of firms in the survey with revenues below $100 million are running fewer than 5 PoCs. 

Forty-six percent of enterprises say that over 10% of PoC projects move forward to production, with most of these (37%) falling in the 11-40% range, while 10% of organizations report success rates above 40%. Under a third of respondents say fewer than 5% of PoCs make it to production, while 21% put the figure at 5-10%. This points to a mixed, nuanced picture for PoC progress, a bifurcation rather than universal failure where many enterprises are successfully transitioning from AI PoCs to production while others are still clearly struggling. 

The whole point of a PoC is to qualify AI initiatives and services at an early stage and to use the old proverb, separate the wheat from the chaff. Frankly, it would be worrying if all initial PoCs - typically limited in scope and experimental - passed all success metrics without refinement. Conversely, even if only a small percentage advance to production, properly validated AI initiatives can deliver significant, far-reaching impact. PoCs serve as decision-making tools and should not be viewed purely as a numbers game.  Moreover, Omdia’s AI Market Maturity Survey shows that AI solutions that have been deployed are either meeting, or exceeding expectations against a wide range of investment drivers. For example, 30% of enterprises report that AI deployments aimed at increasing productivity have exceeded expectations, while 49% say they are meeting expectations.

Figure 1: Enterprise POCs moving to production

The main reason PoCs fail is not usually because of inherent defects in the AI technology being tested, but because enterprises and vendors do not understand the complexity of what AI deployment involves. To help address this, Omdia has developed a comprehensive framework to guide enterprises and vendor partners through successful AI deployment, covering the entire journey**. What enterprises fail to recognize is that most of the critical effort in AI PoCs happens before they begin and continues after completion. The AI journey does not start with the PoC, but rather with the articulation of a value statement that identifies the business challenge or opportunity that AI can, in theory, address. This must be mapped onto a defined use case and the core AI capabilities needed to support it. Enterprises are now in a position to select an AI technology solution that is most likely to deliver optimal results. The role of a PoC is not to select an AI solution from scratch, but rather to provide a structured validation of a solution that has already been shortlisted.

Even when the PoC has been completed with satisfactory results, there is still a long way to go on the path to deployment and scaling. For many AI use cases and solutions, a more comprehensive pilot in live conditions will be required, and the pilot is more likely to be successful if the PoC has been well executed. The transition from AI pilot to scaling is a strategic transformation, not a replication exercise and even a successful pilot does not guarantee success at scale. AI scaling requires architectural evolution, robust governance, and staged expansion to avoid performance and adoption failures. Once scaling has been achieved, AI systems need ongoing support and maintenance to ensure they remain dependable and that performance is optimized in line with evolving business needs. At the same time, as AI deployments mature new opportunities can emerge, such as adjacent use cases or new technology developments that can enhance existing implementations. In reality, the path to AI deployment is not linear, but rather one marked by feedback loops, iterative refinement and continuous improvement. 

*AI Market Maturity – 2025 Data. Conducted in 3Q 2025 among 448 enterprises across major global regions, industry verticals, and companies by size (revenues, employee headcount)

** AI Proof of Concept to Production: Essential Foundations and Critical First Steps
     AI Proof of Concept to Production: Best-in-Class PoCs, Pilots, and Scalable Success