Analyst Augmentation at a Stock Exchange
Surveillance that read every filing by hand now reads everything — and escalates only what matters.
The scenario
A mid-sized European stock exchange's surveillance and research teams were drowning in volume: hundreds of filings, announcements and news items daily, with analyst capacity to review only a sample. The head of market surveillance framed the problem plainly: the risk was not that analysts worked too slowly, but that the unread 70% of material was where the next incident hid.
The exchange deployed a RAG layer over its full corpus of filings and listing rules, so analysts query the entire record in plain language and get answers with citations. Alongside it, an agent-drafted alert system reads every new filing and announcement, scores it against surveillance patterns, and drafts a short brief for anything unusual — anomaly, rationale, source links. Human analysts review every alert before any action, and only analysts can open an investigation.
The design deliberately kept accountability visible: regulators were briefed early, every agent decision is logged with its evidence, and the escalation rules were agreed with compliance before go-live rather than negotiated after an incident.
A year in, coverage is effectively total rather than sampled, time from filing to first analyst look has collapsed, and analysts report their job shifted from reading to judging — the part regulators actually want humans doing.
How AI enters the workflow
Ingestion
AIEvery filing, announcement and relevant news item is ingested and indexed within minutes of publication.
Pattern screening
AIAgents score new material against surveillance patterns: unusual disclosures, trading-adjacent language, rule triggers.
Alert drafting
AIFor flagged items, an agent drafts a brief: what is unusual, why it may matter, with source citations.
Analyst triage
HumanA surveillance analyst reviews each alert, dismisses or deepens it, and may query the full corpus via RAG.
Deep investigation
Human + AIFor real cases, the agent assembles timelines and related filings; the analyst directs and interprets.
Escalation & action
HumanOnly a human opens an investigation or contacts a listed company; the decision and rationale are logged.
Model review
Human + AICompliance and surveillance leads review alert quality monthly and tune thresholds and rules.
Ask about this workflow
ADA, the taskforce deputy, explains exactly how human and AI share the work — ask anything.
The human + agent taskforce
Team leader — approves every deliverable
Team leader — approves every deliverable
SAGE
Knowledge Architect
PRISM
Readiness & Assessment Agent
SENTINEL
Governance Agent
Kuni
AI Learning Companion
Agents propose and execute. The human leader always approves the final result.
Outcomes
| Metric | Before | After |
|---|---|---|
| Filing coverage | ~30% sampled | 100% screened |
| Filing to first analyst review | 2 days | 45 minutes |
| Analyst time on judgment vs reading | 35% | 75% |
Key takeaways
- Total coverage beats sampled coverage — AI changes what 'thorough' means in regulated oversight.
- Citations in every alert made regulator conversations easy instead of defensive.
- Escalation rules agreed with compliance before launch prevented the hardest governance fights.
- Analysts moved from reading to judging; retention improved as a side effect.
Related concepts
Go deeper
Ready to bring AI into your organization?
Talk to us about a guided adoption path for your team — from first use case to production.
Ask about this concept