01Independent market analysis
Some companies bet the balance sheet on automation that isn’t ready. Others sit still until the market quietly reprices everything they do. Both go broke — just from opposite ends of the same scale. We track both patterns, measure how overheated the market is, and publish the moves that survive contact with reality.
No dunking, no doom. We analyze patterns, not people — and every case closes with the move that would’ve worked instead.
Departures
Famous last words — either way.
We do everything with AI now. — We’ve always managed without it. — The model can check its own work. — We’ll wait and see how it shakes out. — We replaced the whole team. — Our moat is twenty years deep.
02The instrument
A composite of six indicators — from earnings-call rhetoric to automation reversals — normalized and weighted into one reading. Both ends of the scale are failure zones; the middle is the operating band.
03The two failure patterns
Failure pattern · Overheated
Headcount cut before the workflow is proven. Forty pilots and no owner. An AI label on the pitch deck doing the work the product should do. Overheating is not enthusiasm — it is spending conviction faster than evidence accumulates.
Early symptoms
Failure pattern · Frozen
Refusal rarely looks like refusal — it looks like prudence. But while the hype cycle burns other people’s money, the frozen firm loses bids to competitors who quietly automated the boring parts. Standing still is also a bet.
Early symptoms
04Case anatomy
On the record
The analyses are anonymized composites. These aren’t: real companies, documented events, sourced figures. Klarna walked back its AI-first support; Builder.ai’s label outran the build to insolvency; Air Canada was held liable for its chatbot’s own advice.
Open the case files →Standing analysis
Six indicators, revised every quarter, tracked as one series. Not to predict the market — to notice when conviction and evidence drift apart.
Open the market monitor →Instruments
Three calculators that turn this site’s claims into something you can test: the reversal bill, the waiting curve, and the five-test scorer.
What to do instead
Five tests that separate automation that compounds from automation that corrodes — plus a cooling-down sequence for the overheated and a starting sequence for the frozen.
05Coverage
The named case files are pinned where they happened — across Europe, Asia and the Americas — with arcs carrying each into the review desk. The fainter dots are the anonymized composites.
On the record, mapped
Bright pins are named case files at their real locations; faint dots are anonymized composites. Patterns, not people.
Bankrupt by AI is run by Balane GmbH. We help leadership teams separate good automation from bad, set kill criteria, and build an AI capability plan that doesn’t bet the company.
Request a strategy briefing