Banks are racing to adopt AI-driven surveillance, but new research shows almost none of them have the underlying data to make it work.
According to Wordwatch, 1LoD’s 2026 Surveillance Benchmarking Survey and Report finds that 89% of banks want AI-enhanced trade surveillance, yet only 11% currently have it. Some 78% want generative AI assistants for their analysts, but just 7% have deployed one. No bank reported having AI fully embedded in either trade or holistic surveillance, with e-comms adoption at 8% and voice at 9%.
Budget and regulatory caution are the usual scapegoats, but the survey undermines both. Around 78% of banks say they have sufficient budget to run an effective surveillance function, and 67% say regulatory sanction risk no longer limits their ability to shift toward risk-based approaches, up from 59% two years ago.
The real constraint sits upstream, in the data itself. Asked what most hinders their surveillance systems, banks split fairly evenly across fragmented data across silos (24%), a lack of standardised formats or identifiers (24%), and inconsistent or poor-quality data (23%). Together, 71% of responses point to the state of the underlying records rather than the analytics layered on top. Legacy or outdated surveillance systems compound the problem, rated a high or medium challenge by four in five banks.
Governance, by contrast, has largely been resolved, with just 7% citing unclear regulatory expectations as a high challenge. The industry has settled who owns surveillance. It has not settled how to feed it with usable data.
False positives remain the clearest symptom. Some 93% of banks rate them a meaningful drag, with 52% calling them a high challenge. The survey attributes this directly to fragmented capture and ageing platforms upstream of the alert stage, meaning better detection engines simply reprocess the same noisy data faster. Holistic surveillance illustrates the point starkly: 48% of banks have linked none of their controls, none have linked all of them, and no bank combines trade and communications data before the alert stage.
Validation is the next hurdle. Around 37% of banks cite difficulty validating NLP and LLM-based systems as a top model risk challenge. Financial Conduct Authority head of secondary market oversight Richard Littlechild said technology is “an enhancer, not a replacement,” placing strong data governance among the fundamentals regulators expect to already be in place.
The report argues that before AI can cut review burdens, records must be complete, held in original format, traceable through a clear chain of custody, and reconciled against source. That work typically falls outside the surveillance budget entirely, usually owned jointly with IT, leaving many AI projects stuck at proof of concept.
Read the full Wordwatch post here.
Copyright © 2026 FinTech Global
