AI product retrospectives need clear measures for closures, pivots and abandoned pilots

A recent roundup brings older cases back into view, but S&P’s research distinguishes company-level abandonment from the share of pilot projects stopped.

An October 24 retrospective revisits AI products that closed, changed direction or disappointed users. Such lists can be useful, but the cases and the statistics used to explain them need different definitions.

A discontinued application, technology absorbed into another product and an experimental corporate pilot that never reaches production are separate outcomes. Combining them into one failure rate obscures what actually happened.

The denominator changes the claim

S&P Global’s 2025 research says 42% of companies abandoned most of their AI initiatives before production. It separately reports that the average organisation stopped 46% of its proof-of-concept projects before production.

The first figure concerns companies; the second concerns projects. Neither is a measured share of all AI startups that have closed.

The underlying Use Cases 2025 survey involved 1,006 respondents and was conducted in late 2024. It is background research, not a new 2026 census of product failures.

Technology can continue after a business changes

HP’s February 18, 2025 announcement illustrates another distinction. It described a definitive $116 million agreement for Humane technology, talent and intellectual property, with closing then expected at month-end. An asset agreement is a different event from demonstrating that the original consumer product achieved lasting success.

That does not make the history unimportant. It makes the dates and the unit of analysis important: product, project, company and acquired capability do not mean the same thing.

A useful retrospective can identify patterns in demand, delivery and economics without turning every change into proof of a single industry-wide conclusion. The latest discussion is best read as a prompt to inspect those specific outcomes, rather than as evidence that every AI experiment shares the same fate.

Sources

- [AI product retrospective](https://techcrunch.com/2026/10/24/the-ai-graveyard-a-running-list-of-projects-and-startups-that-didnt-make-it/) — TechCrunch - [Generative AI growth and mixed enterprise outcomes](https://www.spglobal.com/market-intelligence/en/news-insights/research/2025/10/generative-ai-shows-rapid-growth-but-yields-mixed-results) — S&P Global /451Research - [HP’s February2025 Humane asset agreement](https://www.hp.com/us-en/newsroom/press-releases/2025/hp-accelerates-ai-software-investments-to-transform-the-future-of-work.html) — HP