Last updated October 8, 2026
In 2025, 42% of companies abandoned the majority of their AI initiatives, up from 17% in 2024, according to S&P Global Market Intelligence. Data privacy, security risk, and cost topped the list of obstacles. Four controls set before the build address those causes: one measurable outcome, one clean dataset, guardrails before go-live, and a budget ceiling.
S&P Global's Voice of the Enterprise survey of 1,006 IT and line-of-business professionals in North America and Europe found the share of companies abandoning the majority of their AI initiatives before production more than doubled in a single year. The average organization scrapped 46% of its AI projects between proof of concept and broad adoption.
Gartner's numbers point the same direction. In July 2024, Gartner predicted at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025. In January 2026, Gartner reported the actual figure reached at least 50%, citing poor data quality, inadequate risk controls, escalating costs, and unclear business value. Each abandoned pilot also makes the next AI budget request harder to defend in front of the CFO and the board.
Put the S&P and Gartner findings side by side and four causes repeat.
S&P also found that organizations with lower failure rates weigh compliance, risk, and data availability when they select projects. Our earlier post on starting AI projects from a documented operational problem explains how to choose the use case, and the checklist below picks up from there.
Run all four checks before a pilot receives funding, because a project missing any one of them matches the pattern behind the numbers above.
Write down the single number the project will move, its current baseline, and the date you will measure it. A target such as cutting invoice exception handling time by 50% before the end of Q2 gives finance a figure to verify at review time. Assign an executive owner for the metric who sits outside the IT team running the build.
Identify the one dataset the use case depends on and fix ownership, permissions, duplicates, and retention for that dataset alone, leaving enterprise-wide cleanup for later phases. For a Microsoft 365 Copilot pilot, the dataset might be one SharePoint site with a clear owner and correct sensitivity labels. Our guide to assessing AI data readiness before you build walks through the quality, governance, and infrastructure checks in detail.
Data privacy and security risk were the two leading challenges in the S&P survey, each cited by 38% of respondents. Guardrails reduce both risks when they exist before users touch the system. In a Microsoft environment, the baseline includes four controls.
Microsoft Purview's data security and compliance protections for generative AI, including Data Security Posture Management for AI, give visibility into how Copilot and other AI apps interact with organizational data, with feature availability depending on licensing tier. For the policy itself, see what every LLM policy should cover first.
S&P found cost was the worst-performing AI project KPI in its survey, and Gartner lists escalating cost among the top abandonment causes. Set a monthly ceiling before the pilot starts, and divide spend by the metric from the first check so cost per outcome stays visible.
In Azure, Cost Management budgets can trigger action groups that run automation when a threshold is reached, such as stopping or scaling down resources. For Microsoft 365 Copilot, the ceiling is the seat count, so license the pilot group first and expand only after the outcome is measured.
CloudServus runs these four checks through its AI Readiness Assessment, which reviews technical readiness, strategic alignment, and organizational capacity in your Microsoft environment before any build begins. As a Top 1% Microsoft Solutions Partner globally and an Azure Expert MSP, CloudServus brings Microsoft-certified engineers who configure the Purview, Entra ID, and Azure Cost Management controls themselves. The assessment ends with a pilot plan carrying an owner, a dataset, guardrails, and a spending ceiling, and much of the work can qualify for Microsoft funding.
S&P Global Market Intelligence reported that 42% of companies abandoned the majority of their AI initiatives in 2025, up from 17% in 2024. Gartner reported that at least 50% of generative AI projects were abandoned after proof of concept by the end of 2025.
Gartner cites poor data quality, inadequate risk controls, escalating costs, and unclear business value.
Copilot works within each user's existing Microsoft 365 permissions, so any overshared content becomes easier to find once Copilot is enabled. Sensitivity labels, DLP policies, and access reviews reduce that exposure before rollout.
A complete ceiling covers license or seat costs, consumption charges such as tokens and compute, and implementation hours, paired with an automated alert or action at a defined threshold.