Last reviewed:
What is an AI POC? Definition and right instinct vs wrong instinct
An AI POC (proof of concept) is a short pilot project aimed at testing the technical feasibility and business value of an AI application before any industrial investment. According to estimates cited by RAND (2024), more than 80% of AI projects fail, twice the rate of IT projects without AI.
The failure rate of AI POCs in enterprise is consistently documented. RAND Corporation (2024, 65 interviews with data scientists and engineers): by available estimates, more than 80% of AI projects fail, twice the rate of IT projects without AI. MIT Project NANDA (July 2025): 95% of generative AI pilots have no measurable impact on the P&L. S&P Global Market Intelligence (survey of 1,000+ enterprises, 2025): 42% of companies abandoned most of their AI initiatives, up from 17% a year earlier, and the average organisation scraps 46% of its POCs before production. Five structural causes dominate failure. Business data not ready or not governed. Use case ill-defined or drifting (use-case drift). Lack of pre-AI measurement. No POC-to-production transition planned from the start. Gap between technical demo and real-world conditions (integration, loads, security).
Concrete example
Illustrative case. A 1,100-employee private bank launches in 2025 a POC for an AI assistant for wealth advisers. Three differences from a classic POC. The perimeter is narrow: 12 volunteer advisers, 3 months of testing, on a precise customer sub-segment. Measurement is prepared upfront: 8 business indicators baselined over the previous 6 months. The transition to production is planned from the start: if indicators improve by more than 15%, deployment; otherwise, abandonment. Result at 3 months: 11% average improvement, below the threshold. Decision to abandon taken in committee, with capitalisation of learnings for the next POC.
FAQ
What is an AI POC?
A POC (proof of concept) is a short pilot project aimed at testing the technical feasibility and business value of an AI application before any industrial investment. Its purpose is to decide, quickly and cheaply, whether to continue or to stop.
What is the real failure rate of AI projects?
It is high and consistently documented. RAND (2024) estimates that more than 80% of AI projects fail, twice the rate of IT projects without AI. MIT (Project NANDA, 2025) found that 95% of generative AI pilots have no measurable impact on the P&L, and S&P Global that 42% of companies abandoned most of their initiatives.
Why do so many POCs fail?
Five structural causes dominate: business data not ready or not governed, an ill-defined or drifting use case, lack of pre-AI measurement, no POC-to-production transition planned from the start, and a gap between the technical demo and real-world conditions (integration, loads, security).
How should a POC be framed to avoid failure?
Treat it as a business experiment, not a technical showcase. A narrow perimeter (10 to 20 users maximum), 3 months of testing, a solid pre-AI baseline, continuation or abandonment criteria written before the start, and a production transition pre-planned if those criteria are met.
Should an abandonment be seen as a failure?
No, it is a legitimate outcome. A POC abandoned in 3 months, at limited cost, is an operational success; a POC continued for 12 months at a loss, for lack of a stopping criterion, is a strategic failure. The average organisation in fact scraps 46% of its POCs before production.
See also
Further reading
The Root Causes of Failure for Artificial Intelligence Projects, RAND Corporation, 2024
Sources
- The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed, Ryseff, De Bruhl & Newberry, RAND Corporation, August 2024 (65 interviews with data scientists and engineers). https://www.rand.org/pubs/research_reports/RRA2680-1.html
- The GenAI Divide: State of AI in Business 2025, MIT Project NANDA, July 2025 (report). https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf
- AI experiences rapid adoption, but with mixed outcomes, Highlights from VotE: AI & Machine Learning, S&P Global Market Intelligence (451 Research), 2025 (survey of 1,000+ enterprises). https://www.spglobal.com/market-intelligence/en/news-insights/research/ai-experiences-rapid-adoption-but-with-mixed-outcomes-highlights-from-vote-ai-machine-learning