As 2025 draws to a close, compliance teams across financial services wrestle with a persistent challenge: drowning in false positives from their communications surveillance systems. In MirrorWeb's final webinar of the year, VP of Product Jamie Hoyle unveiled a solution that reimagines how compliance monitoring should work in the age of AI.
The session focused on Sentinel AI. MirrorWeb's newly enhanced surveillance platform eliminates false positives before they ever reach compliance teams. Rather than helping teams review noise faster, it eliminates that noise entirely.
Below is a recap of the key insights and capabilities introduced during the session.
Jamie opened with a reality check about the current state of communications compliance. Traditional lexicon-based surveillance systems flag every variation of risk-related terms, generating 85-90% false positive rates across the industry. The result is compliance teams spending hours clicking "dismiss" on newsletters, IT memos, and harmless conversations - including legitimate uses of flagged words like "guarantee."
There's a vast difference between guaranteeing that the Steelers will win on Sunday versus guaranteeing a client's investment returns. Legacy keyword systems can't tell the difference, creating two critical problems: wasted time on obvious noise, and alert fatigue that increases the risk of missing genuine violations when they appear.
When firms face overwhelming alert volumes, most respond by sampling - reviewing just 10-20% of flagged messages. Using a Minesweeper analogy, Jamie illustrated the danger: firms might find one or two risks in their sample, but they're not examining the other 80-90% of flagged content.
This approach creates blind spots where firms don't know what risks they're missing. The unexamined messages may contain the most critical violations, but they will never be surfaced.
You can check out our interactive Complysweeper game for yourself here. 💣
MirrorWeb's solution implements three complementary features designed to eliminate noise before it reaches compliance teams:
AI-Powered Suggestions automatically identify marketing addresses and repetitive content patterns that generate repeated false flags. Compliance teams can exclude these sources with a single click, ensuring they never appear in review queues again.
AI Post-Review represents the core innovation. After Sentinel's scenario-based system performs initial filtering, an AI layer analyzes every remaining flagged message for false positive patterns. Design partners testing the system saw dramatic results: across 1.5 million messages monthly, alert volumes dropped from approximately 55,000 flags (at a 5% sample rate) to just 1,000 genuinely relevant risks.
That's a 98% reduction in false positives - and critically, the remaining alerts were substantive risks requiring attention rather than dismissible noise.
Continuous Learning through the suggestions feature allows the system to adapt to each organization's specific patterns. When the AI detects repeated dismissals of certain phrases or senders, it proactively suggests exclusions, giving compliance teams ongoing control to fine-tune detection without manual configuration.
Because Sentinel AI reviews 100% of communications rather than samples, compliance teams gain confidence they're not missing critical risks. The system identifies all genuine violations hiding in the data, pulling them into a focused review queue with clear explanations of why each message requires attention - whether for suitability concerns, investment solicitation, or high-pressure sales tactics.
The platform's native data architecture provides additional advantages. Unlike legacy systems that flatten all communications into email format, MirrorWeb captures each platform natively, providing full context (timestamped edits, replies, emojis, deletions, etc) for evaluation.
Q: When does Sentinel AI roll out, and how is it priced? Sentinel AI is available immediately as a separately priced add-on to MirrorWeb's archiving subscription, with the MirrorWeb team handling implementation and pricing discussions.
Q: Can Sentinel be customized for specific asset manager types like VCs? The base system works across RIAs, VCs, hedge funds, and private equity firms. For larger organizations, MirrorWeb offers custom policy development and scenario tuning to address firm-specific risk profiles and surveillance requirements.
Q: Can it handle unusual false positive patterns, like employees whose names trigger flags? Absolutely. Sentinel's native data platform architecture doesn't run policies against sender/recipient fields, eliminating these name-based false positives that plague email-converted systems.
Q: How long does it take for newsletter elimination to work effectively? The false positive detection provides immediate improvement from day one. The newsletter exclusion suggestions then allow ongoing fine-tuning as the system learns organizational patterns, creating compounding improvements over time.
Q: What about false negatives - violations that don't use obvious keywords? Sentinel's scenarios are designed to catch workaround language and non-obvious phraseology. While current functionality focuses on comprehensive detection through sophisticated rule sets, the roadmap includes expanded capabilities for identifying suspicious communication patterns and back-testing to surface potential gaps.
Q: Is this just "different noise" or genuine noise reduction? Real data from six-month deployments shows 93% alert reduction with standard Sentinel, increasing to 98% with Sentinel AI. More importantly, remaining alerts represent genuine risks or borderline cases requiring judgment - not obvious dismissals.
Q: How does this differ from other vendors adding AI features? The philosophical difference is fundamental. Some legacy vendors have introduced AI assistants that help compliance officers review false positives more quickly - essentially clicking a "summarize" button that takes 20 seconds to confirm something is noise, for every alert.
Sentinel AI rejects that premise entirely. The goal isn't reviewing 10,000 alerts 25% faster - it's seeing only the 150 genuine risks that actually matter. Rather than optimizing the wrong process, Sentinel AI eliminates the need for that process altogether.
Q: Is Sentinel pre-trained or using assisted learning? The system combines pre-trained scenarios with individualized suggestion learning. Critically, MirrorWeb doesn't use customer data to retrain models across the customer base, maintaining isolation and transparency. The suggestions feature learns patterns specific to each organization without cross-customer data sharing.
Jamie emphasized MirrorWeb's commitment to "glass box" rather than "black box" AI. Organizations need to understand why messages are flagged and how decisions are made - not just trust mysterious algorithms. This transparency becomes essential during regulatory examinations when firms must explain their surveillance methodologies.
Our message is clear: In an era of genuine AI capabilities, compliance tools should eliminate problems rather than making them slightly more manageable.
For compliance teams manually reviewing thousands of false positives or sampling small percentages and hoping for the best, Sentinel AI provides comprehensive coverage with focused attention on genuine risks.
The shift from checkbox compliance exercises reviewing 10% samples to confident, comprehensive surveillance of 100% of communications represents the transformation modern compliance programs require in an increasingly scrutinized regulatory environment.
Watch the complete webinar below to explore how eliminating false positives can give your team time back to focus on what truly matters: protecting your firm and building defensible compliance programs for an increasingly scrutinized regulatory environment.