Three pillars of work, built around open questions.
Our agenda is organised around problems practitioners and policymakers face today. Each pillar has focus areas, open research questions, and outputs designed to be used, not just read.
Financial Crime & Fraud
Fraud has industrialised. Organised groups run scams like businesses, with recruitment, scripts, mule supply chains and cash-out routes. We study those operations end to end, and test which controls actually disrupt them.
Related insightsScam typologies
Investment, impersonation, job and romance scams, and how their scripts evolve.
Mule networks
Recruitment, account lifecycle and network signals for earlier detection.
Real-time payment fraud
Authorised and unauthorised fraud on instant rails, and effective friction.
AML & proceeds of crime
How fraud proceeds are layered and how monitoring can keep up.
Open questions
- Which interventions at the point of payment reduce scam losses, and at what cost to legitimate customers?
- Can privacy-preserving data collaboration detect mule networks that no single institution can see?
- How should reimbursement and liability models be designed so they reduce fraud rather than shift it?
Responsible AI
AI now makes or shapes decisions about who gets blocked, who gets credit and who gets investigated. We work on how those systems should be governed, explained and audited, and on how AI is being turned against financial institutions and their customers.
Related insightsModel risk & governance
Validation, monitoring and accountability for ML and generative AI in finance.
Explainability
Explanations that serve investigators, customers and supervisors.
Fairness & false positives
Who bears the cost when a fraud model is wrong, and how to measure it.
AI-enabled attacks
Deepfakes, synthetic identities and LLM-scaled social engineering.
Open questions
- What should a proportionate governance framework for generative AI in a financial institution contain?
- How do false positives in fraud models fall across different customer groups, and how can that be reduced?
- Which identity and verification controls remain robust when voice, face and documents can be synthesised?
Digital Trust
Trust is what turns access into adoption. We study the conditions that make people confident using digital financial services, what erodes that confidence, and how to rebuild it after harm, with particular attention to first-time and vulnerable users.
Related insightsDigital identity
Onboarding, authentication and account takeover in a mobile-first market.
Consumer protection
Disclosures, warnings and redress that work in practice.
Inclusion & vulnerability
How fraud risk differs for new, older and low-literacy users.
Victim recovery
Reporting journeys, recovery rates and secondary harm after fraud.
Open questions
- Which scam warnings do people actually notice and act on, in which language and format?
- How does an experience of fraud change a household's use of digital finance afterwards?
- What should a good fraud-reporting and recovery journey look like from the victim's side?
Methods that hold up to scrutiny.
We match the method to the question, and we say plainly what the evidence can and can't support.
Data collaborations
Analysis of transaction and case data under strict agreements, with aggregation and privacy-preserving techniques.
Typology research
Structured case reviews and practitioner interviews that map how schemes operate end to end.
Evaluation & red-teaming
Testing controls and AI models against realistic attacks, and measuring what changes outcomes.
Policy analysis
Comparative review of regulation and standards, translated into practical options for decision-makers.
Shape the agenda.
Our research priorities evolve with the threat landscape. Partners, practitioners and researchers can propose questions for future work.