Most AI wins follow a failed AI project

You know how it goes: a team picks a problem and builds a solution, and the solution turns out to fit a different problem. A couple of recent findings do a good job of telling this story in the context of AI.

42% of companies abandoned most of their AI initiatives before production in 2025, up from 17% a year earlier. (S&P Global)

That looks like failure. But if you take the viewpoint of a founder or startup investor, it looks like a normal first round.

61% of successful AI deployments followed at least one failed AI project. (Stanford Digital Economy Lab, 2026)

The win often lives somewhere else. First attempts typically fail because the team treated AI as a technology project rather than a process and change-management programme.

In Stanford’s case studies, a customer support platform became the base for a sales copilot, and an internal invoice tool became a product sold to clients. MIT NANDA found that about half of AI budgets go to sales and marketing, yet some of the biggest savings came from the back office.

How to plan for the pivot

Keep the sponsor. In every Stanford case that could be checked, the same executive backed both the failed attempt and the success.

Ask who else could use it. Before shelving a project, list what can be reused: data, integrations and the team.

Move fast. Mid-market firms take about 90 days from pilot to rollout, against nine months or more for large enterprises (MIT NANDA). A faster loop means more tries at finding the fit.

A failed first AI project doesn’t prove that AI won’t work. It’s the market research for the project that will.