A pilot that works in six weeks can sit unlaunched for a year at a large American employer. The delay is almost never technical.
Vendor review was designed for stable software
Enterprise procurement assumes a product with a fixed version, a published security posture and a support contract. Reviewers ask what the software does, where data goes and who is accountable when it fails.
An AI vendor answers those questions differently. The behavior changes when the underlying model is updated, and the vendor may not control that model itself.
Reviewers respond by asking for guarantees the vendor cannot give. The file goes back and forth while the pilot team waits, often past the point where the sponsoring executive has moved on.
Data agreements have to be renegotiated
Existing contracts specify where customer data may be stored and processed. Sending that data to a model provider is frequently a new processing purpose that the original agreement did not contemplate.
Legal teams at hospitals, banks and universities treat that as a material change. It triggers review under sector rules, state privacy statutes and sometimes union or client contracts as well.
The renegotiation is slow because it involves parties outside the company. A subprocessor addendum can take longer to sign than the pilot took to build.
Insurance and audit ask separate questions
Cyber insurers and internal audit both want to know what happens when the system produces a wrong output that someone acts on. That question has no clean analogy in traditional software procurement.
Answering it usually requires the company to define a human review step and document it. Defining that step forces an operational decision that the pilot deliberately deferred.
Budget cycles add a second delay
Many large employers fund technology annually. A project that misses its window waits for the next cycle regardless of how ready it is.
Pilots often start with discretionary funds precisely to avoid this, which works until the pilot needs production hosting and a support contract. At that point it enters the normal budget process from the back of the line.
The result is a gap between the demonstration and the deployment that has more to do with the calendar than with the software.
The delay changes what gets built
Teams learn which projects clear review quickly and propose more of those. Internal tools touching no customer data move fastest, so internal tools are what large employers ship first.
Customer facing systems, which carry the larger business case, sit in review longest. That inverts the order a purely commercial analysis would predict.
Understanding the sequence matters for anyone reading adoption news, because a company shipping only internal AI tools may be constrained by its own paperwork rather than by ambition.