Wound down 2023 · $4bn peak valuation
Olive AI
A healthcare automation company valued at $4bn that wound down in 2023 because the technology did not deliver the savings it had promised customers.
What happened
Olive AI sold automation to hospitals and health systems, promising to take cost out of administrative processes. It raised substantially, including a $400m round that valued it at around $4bn.
Roughly two years later the company sold off its remaining businesses and wound down operations. Reporting on the collapse pointed at a straightforward cause: the technology did not deliver the savings that had been promised.
This is the least dramatic entry on this page and possibly the most relevant to anyone assessing AI companies now. There was no fraud finding and no market cycle to blame. The product was sold on a quantified outcome, and it did not produce it at the scale claimed.
Healthcare made the gap visible faster than most sectors would have. Hospital finance teams measure whether a promised saving materialised, and they renew or do not renew accordingly.
Visible at the time
What an investor could have seen
These are things that were observable before the collapse, not hindsight dressed as foresight. Some failures genuinely could not be seen coming; where that is the case, this section says so rather than inventing a warning.
- A product sold on a quantified saving, with no independent verification that the saving materialised at existing customers.
- Rapid revenue growth driven by new logos rather than by expansion at existing accounts.
- Customer references arranged by the company rather than reached independently.
- Net revenue retention not disclosed alongside headline growth.
For your own diligence
What to do differently
- When a product is sold on a measurable outcome, verify the outcome with a customer who has had it long enough to measure. This is the single check that separates an AI company that works from one that demonstrates well.
- Ask for net revenue retention, not just growth. A company adding customers faster than it loses them can look excellent for two years while every cohort quietly fails to renew.
- In AI specifically, ask how the company knows its output is good — whether there is an evaluation suite, and whether performance is tracked over time. Impressions do not survive contact with production.
- Sectors with rigorous buyers surface these problems faster. That is uncomfortable for the company and useful for an investor, because the feedback is real.
Checked, not remembered
Sources
Legal outcomes on these pages move — convictions are appealed, sentences reduced, pardons and commutations granted. Every status here is stated with the date it was accurate to, and every figure is traceable to the reporting linked below.
The mechanics