Six to eighteen months from pilot to production. What makes the transition take this long.

Research finds that 56% of enterprises report it takes six to eighteen months to move a generative AI project from intake to production, and the average enterprise scraps 46% of pilots before reaching production at all. The technology is rarely the bottleneck. Pilots that take six to eighteen months to reach production typically had working technology within weeks. The delay accumulates in the decisions that were deferred to the post-pilot phase: governance classification, data handling rules, workflow integration design and ownership assignment. These decisions are not difficult. They are difficult to make after the pilot has run and the organisation has moved on to other priorities.

Three operating model decisions, consistently deferred, generate most of the production lag. Governance classification determines what risk category this deployment sits in and what review it requires. It gets deferred until the pilot is ready to deploy. At that point it encounters a process that has no established timeline and requires inputs from multiple stakeholders who were not involved in the pilot. Workflow integration designs how the AI output fits into the surrounding process. It gets deferred until after the pilot proves the AI works, at which point the existing workflow resists the change. Ownership assignment names who is accountable for the deployed workflow in production. It gets deferred until deployment is imminent, at which point finding a committed owner takes weeks. Each of these decisions takes days when made at the outset. Together, when deferred, they take months.

For each AI initiative in the pipeline, front-load three decisions before the pilot begins: which governance classification applies and what the review process entails; how the surrounding workflow will be redesigned to accommodate the AI output; and who the named production owner will be, confirmed in writing. These decisions are not contingent on the pilot results. They are operating model decisions that must be in place for the pilot to have a production path. The six-to-eighteen-month lag is largely the cost of the deferral. Organisations that make these decisions before the pilot begins consistently find the transition to production takes weeks rather than months, because the transition was already designed.