Case Study Financial Services

Reducing dwell time from 197 days to 4 hours

A major financial institution was ingesting less than 8% of its security telemetry. After deploying LLM reasoning at full coverage, mean attacker dwell time dropped from 197 days to under 4 hours.

197 days → 4 hours

Mean attacker dwell time

8% → 100%

Security telemetry coverage

94%

Reduction in false positive escalations

1000×

Inference cost reduction vs prior LLM pilot

$4.2M

Estimated avoided breach cost in Year 1

3 weeks

Time from deployment to full production

The Organisation

Global Fortune 500 Financial Institution · Financial Services

The Challenge

The institution's SIEM was configured to ingest roughly 8% of total security telemetry — a deliberate cost-control decision made when storage and inference costs made full ingestion financially unjustifiable. Their SOC team was working from an incomplete picture, relying on heuristic rules and threshold-based alerts that sophisticated threat actors had learned to evade. Over an 18-month period, three separate incidents were later traced back to signals that existed in the unindexed telemetry — signals that would have surfaced the intrusion weeks or months earlier.

The Approach

The institution connected their full telemetry pipeline — authentication logs, network flows, endpoint events, and cloud access logs — to LLM reasoning at 100% coverage. Rather than replacing their existing SIEM, the enriched intelligence was routed back into their existing workflow, with LLM-scored events, confidence levels, and cross-event pattern summaries delivered alongside standard alerts.

"We had the data. We just couldn't afford to read it. Now we can — and the first thing we found was an active lateral movement pattern that had been sitting undetected in our authentication logs for 11 days."

Chief Information Security Officer

Key Finding

The most significant finding in the first 30 days was a coordinated credential stuffing campaign spread across 847 IP addresses, operating at volumes designed to stay below individual threshold rules. The pattern was invisible at 8% coverage. At 100%, it resolved immediately. The campaign was neutralized before any accounts were compromised.

Results at a Glance
Mean attacker dwell time 197 days → 4 hours
Security telemetry coverage 8% → 100%
Reduction in false positive escalations 94%
Inference cost reduction vs prior LLM pilot 1000×
Estimated avoided breach cost in Year 1 $4.2M
Time from deployment to full production 3 weeks
Get in Touch

Talk to us about your data.

Tell us about your event stream and we'll show you what full LLM reasoning coverage looks like for your environment.

Or book a call directly →