Notifications
Clear all
Topic starter
08/06/2026 4:42 pm
TL;DR: A S&P Global survey of more than 1,000 enterprises found 42% abandoned most AI initiatives in 2025, while the average organisation scrapped 46% of AI proofs of concept before production, pointing to cost, privacy, and security failures, according to WorkOS and S&P Global. The real constraint is not model quality alone, but whether governance, data readiness, and human operating models can survive production pressure.
NHIMG editorial — based on content published by WorkOS: Why most enterprise AI projects fail and the patterns that actually work
By the numbers:
- 42% of companies abandoned most of their AI initiatives in 2025, a dramatic spike from just 17% in 2024.
- The average organisation scrapped 46% of AI proof-of-concepts before they reached production.
- Over 80% of AI projects fail, which is twice the failure rate of non-AI technology projects.
Questions worth separating out
Q: How should organisations move AI pilots into production without creating governance gaps?
A: They should require production-readiness gates that combine identity controls, data governance, monitoring, and user adoption before any pilot scales.
Q: Why do enterprise AI programmes fail even when the model performs well?
A: Because model accuracy does not solve access, workflow, data, and accountability problems.
Q: How do security teams know whether AI governance is actually working?
A: They should look for clear lineage on training and retrieval data, named owners for the stack, documented human override paths, and measurable operational SLAs.
Practitioner guidance
- Define production readiness gates for AI programmes Require secure authentication, compliance sign-off, monitoring, and user training before a pilot can move beyond sandbox use.
- Treat data lineage as a control requirement Document which datasets feed the model, who approves access, how retention is enforced, and where governance metadata lives.
- Map human approval points in every AI workflow Specify which actions remain human, where exceptions escalate, and how overrides are recorded.
What's in the full article
WorkOS's full article covers the operational detail this post intentionally leaves for the source:
- The specific enterprise examples behind the four patterns, including how each organisation moved from pilot to operational use.
- The practical workflow issues that slowed adoption, such as user trust, rollout coordination, and production hardening.
- The savings and productivity outcomes cited for individual organisations, which are useful if you need implementation-level context.
- The article's own view of how to recognise whether an AI programme is likely to ship or stall.
👉 Read WorkOS's analysis of why enterprise AI initiatives fail →
Enterprise AI projects: where governance and deployment break down?
Explore further
08/06/2026 6:28 pm
Enterprise AI failure is fundamentally an identity and operating-model problem, not a model-selection problem. The article shows that teams lose when production demands secure authentication, governed data, and coordinated ownership at the same time. That is an identity discipline issue because the system cannot be trusted if the humans, services, and workflows around it are not explicitly governed. Practitioners should read this as a warning that deployment discipline, not model sophistication, decides whether AI becomes operational.
A few things that frame the scale:
- The average estimated time to remediate a leaked secret is 27 days, despite 75% of organisations expressing strong confidence in their secrets management capabilities, according to The State of Secrets in AppSec.
- Organisations maintain an average of 6 distinct secrets manager instances, creating fragmentation that undermines centralised control.
A question worth separating out:
Q: What is the difference between a successful AI pilot and a production-ready AI service?
A: A pilot proves the model can work in isolation, while a production-ready service proves the whole operating model can sustain it. That includes secure access, compliance workflow, support ownership, observability, and a user journey that does not collapse under real-world pressure.
👉 Read our full editorial: Enterprise AI fails when governance, data, and adoption drift
