FCA Tests AI Models to Strengthen Market Surveillance
Explore how regulators and industry collaborated in NayaOne’s sandbox to validate AI models that make market monitoring faster and more reliable.
Outcomes
5x
Reduction in False Positives
2x
Detection Precision Improvement
7
Participating Vendor Teams
$100k
Average PoC Cost Saving Per Vendor
Technology Vendors Suited to Evaluation
Business Problem
Traditional surveillance systems struggle to detect complex and evolving forms of market abuse.
Most rely on static, rules-based alerts that flood compliance teams with false positives and miss subtle manipulation across markets.
With increased trading velocity and AI-driven strategies, regulators and firms need a faster, more intelligent way to identify abuse patterns, validate models, and collaborate securely on new detection methods.
Challenges
- High false-positive rates from legacy, rule-based systems
- Limited cross-market visibility into correlated trading activity
- Data silos across venues, products, and instruments
- Complex manipulation patterns beyond parameter-based thresholds
- Lack of safe, real-data alternatives for model experimentation and benchmarking
From Idea to Evidence with NayaOne
The FCA ran a three-month Market Abuse Surveillance TechSprint (May – July 2024) in the FCA Digital Sandbox powered by NayaOne. Nine global teams collaborated using ~1TB of pseudonymised trade, order book, news, and pricing data to design AI/ML-based surveillance models.
Within NayaOne’s platform, participants built and tested anomaly-detection models, used LLMs to reduce false positives, and validated multi-source analytics – all with full auditability for supervisors.
Key testing activities included:
- Developing isolation forest and Bayesian network models for anomaly detection
- Using econophysics models to analyse limit order book dynamics
- Applying large language models to contextualise alerts and filter noise
- Correlating structured (trade data) and unstructured (news, comms) signals
- Capturing audit trails and artefacts for regulatory visibility and review
The sprint validated that AI and ML can materially improve surveillance effectiveness and efficiency. It showed that using synthetic and pseudonymised datasets within a secure sandbox enables firms to build better models faster – without compromising compliance, privacy, or supervision. The initiative also provided the FCA with a repeatable framework for AI adoption in market integrity monitoring.
Impact Metrics
PoC Timeline Reduction
12 weeks with NayaOne vs 12 – 18 months traditionally
Time Saved in Vendor Evaluation
1+ year
Decision Quality
Higher confidence in AI model explainability, validation accuracy, and regulatory readiness.
KPIs
- Signal quality improvement: Reduction in false positives across tested models
- Detection accuracy: Precision of anomaly identification vs baseline rules engines
- Data coverage: Volume and diversity of datasets processed during modelling
- Model development time: Duration from concept to validated output
- Cross-firm participation: Number of teams or institutions collaborating securely
- Audit and traceability: Completeness of model explainability and experiment record
Explore How AI and Synthetic Data Are Transforming Market Surveillance
See how regulators and financial institutions use NayaOne’s sandbox to test, benchmark, and scale AI-based surveillance models – securely, collaboratively, and with full regulatory oversight.
