Deloitte's pilot-to-production figure counts executives above a 40% threshold

The 25% is a share of respondents, and the bar each one had to clear was 40% of their own pilots ^[1]. Three quarters of the executives surveyed sat below that line ^[2]. The threshold treats portfolios of very different sizes alike: one pilot live out of two is 50% and clears it, while seven live out of twenty is 35% and does not ^[3]. Boards that read that number as an industry-wide conversion rate for pilots are reading the wrong number.

The diagnosis attached to it comes from one practitioner's caseload. Taras Tymoshchuk, founder and CEO of Geniusee ^[9], wrote in a Forbes Tech Council column that when he reviews an AI initiative he asks for the current baseline, the target improvement, expected volume, an acceptable failure rate and a cost ceiling ^[4]. He puts an accountable business owner next to those five numbers. "A pilot without that ownership can remain active long after its business case has disappeared," he wrote ^[5].

The advice has a beneficiary: a firm that builds production systems has an interest in pilots being gated hard and then funded. The item on the checklist with independent survey support is the cost ceiling. IBM and Oxford Economics surveyed 2,000 technology executives in early 2026 and found 84% had not fully operationalized AI financial management, and 85% lacked complete visibility into real-time AI spending ^[6][7]. Inside that 85%, the ceiling is untestable, because the company does not know what the pilot is running at. Tymoshchuk lists model calls, retrieval, infrastructure, monitoring, human review, incident handling and support as the costs a small pilot hides ^[16]. "The relevant metric is reliable business value per dollar spent," he wrote ^[12].

The McKinsey finding cited alongside it is a correlation. Its 2025 State of AI survey found AI high performers were nearly three times as likely as other organizations to have fundamentally redesigned workflows. They were three times as likely to report strong senior-leadership ownership and commitment ^[8]. The survey puts the two together and leaves the order open. Redesigned workflows may earn senior attention as easily as senior attention produces redesigned workflows.

One project in the column shows what a gated proof of concept produces. The goal was multi-angle images with controllable 3D camera parameters. Three approaches were tested. Two ran into constraints around third-party APIs and Gaussian splatting, and a hybrid Gemini and Nano Banana workflow proved strongest on output quality, camera control and commercial viability ^[10]. What the exercise delivered was a narrowed architecture choice, and the client approved production as the next phase ^[11].

For anyone approving a pilot this quarter, the question is whether those five numbers go into the approval memo or get discovered during scaling. Approve a pilot with a recorded baseline, an accepted failure rate and a monthly cost ceiling, and it can be stopped next quarter by comparison against those figures. Without them, the sponsor defines success after the results are in. The Deloitte report, as quoted, does not say which of the three quarters below the 40% line skipped that step.