[01] Agent Architecture
AI agents that reason like investigators
Hesper AI deploys autonomous investigation agents on every claim. Each agent plans its own investigation, gathers evidence from multiple sources, cross-references findings, and delivers an audit-ready report - compressing 14 days of manual work into under 60 minutes.
[02] How It Works
Multi-step reasoning, not static rules
Traditional claims automation follows pre-built decision trees. If a claim matches a rule, it gets flagged. If it does not, it passes through. This misses complex fraud that does not fit neat categories and generates noise on legitimate claims that happen to trip a threshold.
Hesper AI agents work differently. Each agent receives a claim and reasons about it - deciding what to investigate, what data to pull, and how to interpret the results. When the agent finds a discrepancy, it does not just flag it. It investigates further, gathering corroborating evidence, checking alternative explanations, and building a case narrative that explains exactly what it found and why it matters.
01
Reason
The agent analyzes the claim, identifies risk signals, and builds a structured investigation plan tailored to this specific case.
02
Gather
It pulls data from claims systems, external databases, public records, and OSINT sources in parallel - not sequentially.
03
Resolve
Findings are cross-referenced, contradictions are highlighted, and everything is compiled into an evidence-backed report ready for review.
[03] Agent Capabilities
What the agents can do
Four capability layers work together to handle the full investigation lifecycle. Each layer is designed to mirror what a senior SIU investigator does - but at machine speed and across every claim, not just the ones that make it to the top of the queue.
[01] Reasoning & Planning
Breaking down complex investigations
Each claim gets its own AI agent that decomposes an investigation into discrete steps. The agent identifies what needs to be verified, prioritizes high-risk signals, and builds an investigation plan - the same way a senior SIU analyst would approach a new case, but in seconds instead of hours.
[02] Data Integration
Pulling from multiple sources simultaneously
Agents connect to claims management systems, third-party databases, public records, and OSINT sources in parallel. Instead of an investigator manually querying NICB, ISO ClaimSearch, DMV records, and social media one at a time, the agent pulls everything at once and cross-references the results.
[03] Analysis
Pattern detection and anomaly identification
The agent runs document forensics, behavioral analysis, and network mapping in parallel. It detects tampered documents, identifies connections between entities, flags billing anomalies, and cross-validates every data point against authoritative sources - surfacing the inconsistencies that matter.
[04] Output
Reports, evidence, and recommendations
The agent compiles everything into a structured, evidence-backed case file. Denial justifications cite specific findings. Evidence packages include source attribution. Every recommendation is explainable and audit-ready - so your team can act immediately instead of spending days writing up the case.
[04] Architecture
How it all connects
Hesper AI sits between your claims systems and your investigation team. Claims flow in through API or platform upload. Agents do the work. Reports flow back to your team ready for review and action.
Input
Claims Systems
FNOL, documents, photos, policy data, prior claims
Processing
Hesper AI Agents
Reasoning, data integration, forensics, analysis, report generation
Output
Your SIU Team
Case narratives, evidence packages, denial justifications, SAR filings
External data sources
[05] Security & Compliance
Built for regulated industries
Insurance claims data is sensitive. We designed Hesper AI with security and compliance as foundational requirements, not afterthoughts bolted on before an audit.
SOC 2 (in progress)
Type II audit underway. We are building to SOC 2 controls from day one - not retrofitting compliance after the fact.
Zero data retention
Claims data is processed in memory and discarded immediately. We never store your documents, never train on your data, and never retain PII.
GDPR-ready
Data processing agreements, right-to-deletion support, and privacy-by-design architecture. Ready for EU and global privacy requirements.
Encryption everywhere
TLS 1.3 in transit, AES-256 at rest for any transient storage. All API communication is encrypted end-to-end between your systems and ours.
[06] FAQ
Frequently asked questions
How do Hesper AI's agents differ from rule-based automation?
Rule-based systems follow static decision trees - if X then Y. Hesper AI's agents reason dynamically. They evaluate each claim individually, decide what to investigate next based on what they find, and adapt their approach as new evidence surfaces. A rule-based system might flag a claim because the amount exceeds a threshold. An agent investigates why the amount is unusual, checks the claimant's history, verifies the vendor, and builds a case narrative with cited evidence.
What data sources do the agents connect to?
Agents integrate with your existing claims management system and simultaneously query external sources including NICB, ISO ClaimSearch, DMV records, NMVTIS, business registration databases, professional license registries, and public records. They also run OSINT and social media scans. All of this happens in parallel - the agent does not wait for one query to finish before starting the next.
How long does an investigation take?
A full investigation - from claim intake to investigation-ready report - typically completes in under 60 minutes. This includes document forensics, database verification, OSINT research, network analysis, and report generation. The equivalent manual process takes an SIU investigator 10 to 14 business days.
Does Hesper AI replace human investigators?
No. Hesper AI handles the repetitive, time-consuming parts of investigation - data gathering, cross-referencing, document analysis, and report drafting. Your SIU team reviews the agent's findings, makes final decisions, and handles cases that require human judgment like in-person interviews or complex litigation. The goal is to give every investigator the capacity to handle 10x more cases.
See the agents in action
Book a 30-minute demo and see how an AI investigation agent works on a real claim.
Request Demo