INETCO, a Vancouver-based payment fraud prevention software company, has launched BullzAI Investigate, an agentic AI module designed to accelerate fraud case management for banks and payment service providers. The product sits alongside the existing BullzAI monitoring platform and is available now.
The module uses a proprietary small language model deployed entirely on-premise, meaning transaction data does not leave the customer’s infrastructure. Specialised AI agents collate transaction datasets, identify behavioural patterns and automatically triage incoming alerts, surfacing the highest-risk cases for analyst review. Each recommendation comes with an explainable risk score that INETCO says gives investigators auditable reasoning for every decision. The system refines its own output through a human-in-the-loop supervised learning cycle, incorporating analyst feedback after each closed case.
INETCO says early results show investigation times dropping from a range of 10 to 30 minutes to approximately 20 seconds, a reduction it puts at 97 to 99 per cent. The company also cites a recommendation precision rate of approximately 95 per cent, though it has not independently published the methodology behind that figure.
Ugan Naidoo, chief technology officer at INETCO, said: “Agentic AI automates the heavy lifting by collating transaction data, triaging alerts and delivering explainable risk scores that support faster, more transparent decisions. Rather than replacing analysts, INETCO BullzAI Investigate serves as an intelligent partner that works continuously behind the scenes, allowing fraud teams to investigate more effectively while human oversight remains firmly in control.”
A Chartis Research report featuring a BullzAI deployment at an unnamed South African bank is cited by INETCO as third-party evidence of the platform’s performance, though full details of the Chartis methodology and the bank’s scale of deployment were not included in the release.
Market context
Fraud operations tooling is a rapidly expanding segment of the broader financial crime compliance market. Alert fatigue has become a recognised structural problem: most anti-fraud platforms generate volumes of low-risk notifications that absorb analyst time that would be better directed at complex, high-value cases. Several established vendors, including NICE Actimize, SAS and Featurespace, already offer AI-assisted case management layers, and a growing cohort of specialist challengers are competing on explainability and on-premise data residency as differentiators, particularly for institutions in jurisdictions with strict data localisation rules.
The on-premise deployment model is a deliberate positioning choice. Cloud-based fraud tooling faces resistance from some tier-one banks and regulated institutions in markets where transaction data cannot be processed outside a defined geographic boundary. Running the language model on the customer’s own infrastructure addresses that concern directly, though it also places the maintenance and update burden on the institution rather than the vendor, which has its own operational cost implications.
Regulatory read-across
The emphasis on explainability and human oversight reflects the direction of travel in AI governance for financial services. The EU AI Act classifies certain high-risk AI applications in financial services under requirements that include transparency, auditability and human control of consequential decisions. In the UK, the FCA and PRA have both signalled expectations around model risk management for AI systems used in credit, fraud and operational decisions. INETCO’s stated design principle, keeping human analysts in the decisional loop with clear reasoning trails, is commercially sensible in that regulatory environment, though institutions will still need to conduct their own model risk assessments before deploying the tool in regulated workflows.
The company monitors more than 100 billion transactions annually across its customer base, according to its own figures.