Insights
Compliance, financial crime, and what AI really changes.
Regulatory developments translated into operational consequences, and honest assessments of where machine intelligence belongs in a control environment.
Choosing a monitoring system: what to test before you buy
Vendor demos are built to succeed. What to test instead: your own data, who can change a rule, the investigator's screen, and the exit terms.
What a tuning exercise actually involves
Scope, evidence, timeline and the file an examiner asks for first. What to expect before commissioning threshold optimisation work.
How to reduce false positives without quietly reducing coverage
Alert volume is a symptom, and thresholds are the last lever, not the first. The order of operations that keeps a reduction defensible.
What AI actually fixes in transaction monitoring, and what it doesn't
Machine learning is very good at ranking alerts and very bad at inventing typologies nobody has described. A practitioner's map of where the lift is real, where it is marketing, and what has to be in place before either.
The EU AML package: what actually lands on operations teams
Beyond the headline of a single rulebook and a new authority: the specific changes to beneficial ownership, cash limits and supervisory expectations that will reshape day-to-day control design.
Measuring an analyst, not their output
Alert throughput is a terrible quality metric. A weighted QA framework for scoring decision quality (narrative, escalation judgement, evidence handling) without turning review into a productivity race.
Keeping dispute rates low: a control view of chargebacks
Chargeback rate is a lagging indicator of decisions made ninety days earlier. Separating true fraud from service failure from cardholder misuse, and what actually moves the number.
Screening fuzziness is a policy decision, so treat it like one
Match thresholds are usually inherited from a vendor default and never revisited. What a defensible calibration exercise looks like, and how to document the risk appetite behind it.
The model documentation a supervisor will actually ask for
Performance metrics are the easy part. The harder questions are about lineage, drift, challenger models and who signed off: a checklist for teams deploying their first model in a control.
The Compliance Signal
Get new analysis by email. Short, occasional, no vendor content.
Occasional. Unsubscribe any time.