AI Expert-Call Surveillance for MNPI Detection
~75%
2–4 months
Not disclosed
Encore needed to detect nuanced Material Non-Public Information (MNPI) disclosures in expert-network call transcripts, but existing approaches lacked the interpretability required for compliance audits and the reliability demanded under regulatory scrutiny.
PressW refined an LLM architecture combining structured prompt engineering for compliance-risk detection, chain-of-thought reasoning for transparent decisions, and rigorous testing frameworks benchmarked against compliance standards.
The build paired structured prompts for risk identification with step-by-step chain-of-thought reasoning, strategic model selection to balance performance and efficiency, and robust testing protocols benchmarked to compliance standards.
Compliance firms, expert-network platforms, and regulated financial-services organizations that need reliable, auditable MNPI detection across large volumes of call transcripts.
- Expert-network call transcripts (source corpus)
- Encore's existing surveillance platform (deployment target)
- Compliance-standard benchmark set used for testing
- Audit trail capturing the reasoning behind each flag
- Call transcript ingest into structured risk-identification prompts
- Chain-of-thought reasoning output producing an interpretable flag with rationale attached
- Model selection layer balancing performance against processing efficiency
- Benchmark testing harness scoring detection against compliance standards
Higher MNPI Detection Accuracy
Audit-Ready, Explainable Decisions
Supported Acquisition by ACA Group
The design priority was interpretability rather than raw detection. Chain-of-thought reasoning renders each flag in language a compliance reviewer can read at a glance and defend under audit. Built correctly, that explainability layer costs nothing in accuracy or latency.
Bryson Greenwood
Encore Compliance uses an LLM system with structured prompts that identify compliance risk in call transcripts, paired with chain-of-thought reasoning so each flag carries its rationale. That interpretability is what makes the flags usable in a compliance audit, and detection accuracy improved materially after the refinement.
Structured prompt engineering for compliance-risk detection, chain-of-thought reasoning for transparent decisions, strategic model selection balancing performance against efficiency, and testing protocols benchmarked to compliance standards. The specific models are not disclosed.
Detection accuracy for potential MNPI disclosures improved materially, though no percentage was disclosed. Chain-of-thought reasoning made every flag interpretable for compliance audits, and the platform contributed to Encore Compliance being acquired by ACA Group.
PressW reports meaningful results within weeks. The engagement refined an existing architecture through structured prompt design, chain-of-thought reasoning, model selection and benchmark testing.
Compliance firms, expert-network platforms, and regulated financial-services organizations that need reliable, auditable MNPI detection across large volumes of call transcripts.
Have a similar challenge?
