The number — Buying AI succeeds about two times in three; building it in-house, one in three

MIT's 2025 study of AI in business measured a stark gap across 300 real deployments. Buying a solution from a specialized vendor and forming a partnership reaches deployment roughly two times out of three; building the tool in-house succeeds only about one time out of three. The instinct to “just build it ourselves” therefore doubles the risk of failure, and it stays invisible until several months of spending have gone by. Before launching an internal AI project, honestly cost out the full price of building it yourself (data, integration, maintenance, retaining the skills) and compare it against the same need met by a proven vendor.

Source: MIT NANDA, “The GenAI Divide: State of AI in Business 2025,” July 2025 (sample: 52 executive interviews, 153 leaders surveyed, 300 deployments analyzed).