AI in fraud and financial crime is no longer experimental
Part 4 of 6 in The payment investigations series
It is operational.
📊 What the survey shows
- 51% of organisations already have AI live
- 47% are implementing it
- Yet only 19% operate with full AI autonomy
- Most decisions still rely on human oversight
🚀 Ambition is there
Nearly all organisations pursue advanced AI initiatives:
- Generative AI 76%
- Real-time anomaly detection 57%
- Agentic AI 51%
Yet automation remains limited. Heavy manual work dominates, increasing cost and slowing response as volumes and fraud complexity grow.
🧱 The main blockers are structural
- Data privacy and ethical use
- Compliance with emerging AI regulation
- Limited in-house data science capability
- Reliance on outsourcing and cloud platforms to fill gaps
- Data governance issues and silos blocking execution
As a result, priorities remain defensive, prioritising secure payment environments over optimisation and customer experience.
🧭 The picture is consistent
AI adoption is widespread, but full automation is rare. The constraints are trust, data governance, regulation, and operational realities, not a lack of advanced AI techniques.
🌍 At Alpina Analytics, this is the problem space we are focused on: evidence-backed investigation with TxAgent, on data prepared by TxFlow.
Source: AI in action: the reality of AI adoption in fraud and financial crime, ACI Worldwide with Finextra, 2026.