Complead added a shadow mode in which a new AI agent runs on live cases, records the decision it would have made and takes no action. A comparison screen covering Name Screening, Ongoing Monitoring, Advanced Adverse Media and Transaction and Fraud Monitoring shows the agreement rate with a likely range, agree and disagree counts, missed true positives and breakdowns by rule, exportable to Excel. Three go live checks (100 analyst decided cases, a 95 percent agreement target, zero missed true positives) inform a manual, recorded go live step.
Our readBanks can validate an agent against their own analysts' decisions before granting it autonomy and keep that record for model risk review and audit. Missed true positives are surfaced as their own metric, which is the failure mode examiners care about.