Banks Think They're Ready for Tomorrow's Fraud. Feedzai's Survey Suggests Otherwise

Feedzai's survey of 1,030 senior fraud and financial-crime leaders found a 13.3-point gap between financial institutions' self-rated readiness of 80.2/100 and their measured capabilities of 66.9. While 96% expect AI fraud-tool budgets to grow over the next two years, 73% report analysts spend too much time manually gathering data and only 41% have fully connected data across fraud, AML, compliance, and cybersecurity teams. Scams were the biggest source of fraud losses in 10 of 12 surveyed countries, and 87% would consider sharing fraud intelligence with other institutions, including competitors.

Analysis

The investable signal is not the readiness-score gap; it is that banks may have to spend on data plumbing and workflow integration before incremental AI delivers measurable fraud reduction. That shifts near-term value toward vendors able to connect payments, AML, identity and cyber data, and toward implementation partners—not automatically toward model vendors. Feedzai’s survey is vendor-sponsored and reports intentions and self-assessments, not verified budgets, loss reductions or purchasing decisions; treat the 96% budget-growth expectation as a demand indicator, not booked revenue. No listed-company exposure is established by the supplied data.

Second order: scams involving customers authorizing payments are harder to stop with transaction anomaly detection alone. Banks may face a costly trade-off among intervention, customer friction and reimbursement exposure, while better cross-institution signals could improve detection but remain constrained by privacy, governance and competitive concerns. Agentic systems could reduce analyst handling time, but errors and false positives create a human-review and model-control layer that may limit labor savings.

Near term (days), this is weak price discovery and likely no standalone trade. Over 1–3 months, watch vendor commentary and bank procurement evidence for spending converting from pilots to integrated deployments. Over 6–18 months, evidence of lower fraud losses or analyst workload—not AI adoption counts—would validate durable software demand. The contrarian risk is that fragmented legacy data and governance make integration slow, pushing expected AI monetization out even as budgets rise. The thesis weakens if banks report flat fraud-tech spending or no measurable improvement in losses, false-positive rates, or case-resolution time.

Key Decisions for Investors

• No directional position on this survey alone: the evidence is low-impact, vendor-originated, and lacks verified spend or outcome data.
• Watch listed fraud/identity and financial-crime software providers, including NICE and FICO, for disclosures tying deployments to integrated data, realized bookings, and measurable customer outcomes; do not infer exposure or upside from this report alone.
• For the next 1–3 months, monitor bank and vendor earnings for fraud-tech budget conversion, implementation timelines, and changes in scam-related loss or reimbursement commentary. Treat announced AI pilots without production metrics as non-confirmatory.
• Reassess the longer-term software-demand thesis if institutions show sustained declines in fraud losses and manual case time without materially higher false positives; fade it if integrations remain stalled or spending growth fails to appear in reported results.