FIS Global — Dynamic Trade Compliance Monitoring via Cortex and Large Language Models
FIS's rules-based alert system for monitoring trading activity against FINRA regulations was generating a large volume of low-priority, redundant alerts. Analysts faced significant manual review burden with no intelligent triage layer. Hakkoda evolved the system into a dynamic, risk-based classification engine using Snowflake Cortex and large language model (LLM) capabilities. The resulting system scores each event with a quantifiable risk score using a multivariable approach that weighs violation threshold, sentiment, comments, confidence, and risk category, then classifies alerts into two distinct categories based on complexity.
- Rules-based alert system generating high volumes of low-priority, redundant alerts against FINRA regulations
- Every alert treated as equal severity with no scoring or routing logic to differentiate genuine risk
- Cumbersome manual alert review process consuming significant analyst time per compliance event
- No dynamic assessment of risk genuineness, leaving high-severity signals buried in alert noise
- Dynamic risk scoring weighs violation threshold, sentiment, comments, confidence, and risk category per event
- Alerts classified into two distinct categories by complexity, routing each to the appropriate review queue
- Improved accuracy of trade activity monitoring with redundant alerts eliminated from analyst queues
- Multivariable LLM-based approach isolates legitimate threats and surfaces high-severity signals automatically
| Value Driver | Operational Impact |
|---|---|
| Alert noise reduction | 60–85% reduction in low-priority alerts reaching human analysts. High-severity signals surface automatically. |
| Analyst review time | Review time per alert reduced from approximately 12 minutes to under 2 minutes per compliance event. |
| Monitoring accuracy | Improved accuracy of trade activity monitoring with redundant alerts eliminated from analyst queues. |
| Risk assessment quality | Multivariable dynamic risk scoring introduced where none previously existed, isolating legitimate compliance threats. |
| Compliance exposure | Reduced manual review burden diminishes risk of missed high-severity signals at institutional alert volumes. |
Capital Group — LLM-Based Sentiment Analysis for Quantitative Trading Strategy
Capital Group's quantitative trading teams and executive leadership depended on manually aggregated financial news to inform strategy and decision-making. There was no automated mechanism to apply sentiment analysis at scale across the volume of news the firm needed to track. Hakkoda built an LLM-powered application that automated web scraping across financial news and media outlets, produced real-time sentiment scores for quant strategy inputs, and replaced manual executive briefing cycles with AI-generated summaries across a wide variety of macro news sources.
- Massive volume of financial news requiring manual aggregation across dozens of sources to present to executive leadership
- No automated sentiment scoring mechanism available for quantitative trading strategy decisions
- Manual research processes and data collection required for tracking macro news events across the market
- Executive decision-making slowed by bandwidth-constrained, analyst-dependent briefing preparation cycles
- Automated web scraping continuously aggregates financial news across a wide range of outlets and media sources
- LLM application produces sentiment analyses usable by quant teams for trading strategy decisions in real time
- AI-generated executive summaries replace manual briefing preparation for leadership decision cycles
- Macro research tracking automated at scale with consistent coverage across a wide variety of news outlets
| Value Driver | Operational Impact |
|---|---|
| Analyst time savings | Manual news aggregation eliminated. Analyst capacity redirected from data collection to strategy and analysis. |
| Signal-to-decision speed | Quant teams receive automated sentiment scores in real time rather than waiting on manual research cycles. |
| Coverage breadth | Scalable automated scraping vs. selective, bandwidth-constrained manual tracking across the analyst team. |
| Leadership decision quality | Consistent, data-driven executive summaries replace manual briefing preparation for leadership decision cycles. |
MAMBA Research Live — Cortex AI for Earnings, News, and Institutional Analyst Intelligence
MAMBA is Hakkoda's capital markets accelerator, built specifically for asset management, trading, and institutional research workflows in financial services. Research Live, its NLP-powered research module, uses Snowflake Cortex AI to automate data gathering and allow analysts to interact with earnings releases, SEC filings, breaking news, and economic announcements through natural language queries (NLQ). The result is faster synthesis, real-time anomaly detection layered onto financial text streams, and consistent research quality independent of individual analyst bandwidth or coverage capacity.
- Institutional analysts manually pulling and synthesizing earnings releases, SEC filings, and economic announcements
- Significant lag between market-moving events and analyst insight delivery to portfolio managers
- Research coverage quality dependent on individual analyst bandwidth, creating uneven depth across the bench
- No structured mechanism to apply anomaly detection across unstructured financial text at the speed markets require
- Cortex AI automates data gathering, surfacing real-time financial information via natural language queries (NLQ)
- Analysts query earnings and economic announcements conversationally through Cortex Assistant for rapid synthesis
- Real-time forecasting and anomaly detection layered onto news and earnings text streams
- Consistent research quality across analyst bench regardless of individual capacity or coverage load
| Value Driver | Operational Impact |
|---|---|
| Research throughput | Scaled without headcount increase. Automated gathering replaces manual data collection cycles at the analyst layer. |
| Speed to insight | Breaking market events synthesized in minutes rather than hours of manual analyst work per event. |
| Coverage consistency | Uniform research quality across analyst bench, independent of bandwidth or individual expertise gaps. |
| Risk signal quality | Anomaly detection adds a structured risk signal layer across unstructured financial text, not previously available at speed. |