Why 95% of enterprise AI pilots fail inside regulated institutions
MIT's 2025 study found that 95% of enterprise AI deployments produce no measurable business impact. Inside banks and fintechs, the cause is not model quality. It is that generic AI is built to always answer, and in regulated work, a confident guess is a fine.
8 min read
The question that kills AI projects in a bank is not technical
Regulated buyers ask where their data goes before anyone evaluates output quality. Most AI products send data to a provider's servers, and that is often a non-starter.
7 min read
What would AI have to be, before a compliance officer signs off?
A working list of what it takes for a risk or compliance owner to trust an AI system with regulated casework. Almost none of it is about how capable the model is.
8 min read
Why a static AI model quietly stops being correct
Regulatory rulebooks change constantly. A model is a snapshot frozen at training time, with no way of knowing when the rule underneath it has moved.
7 min read
The same rule-bound pattern is hiding in every back office
Onboarding, disputes, reconciliation, compliance alerts, tender responses: strip each down and the same structure appears every time.
7 min readMore pieces in progress. Check back for new writing on regulated back-office work.