Introduction to Case Review Agents

Overview

Case review agents are AI-powered assistants built directly into Persona‘s Cases product that help automate manual review for identity decisions. They leverage the power of large language models (LLMs) to work alongside reviewers by analyzing the information already in Persona cases, surfacing a structured recommendation, and documenting their reasoning.

With the help of case review agents, manual review teams will be able to spend less time on routine decisions and more time on the cases that actually need human judgment. This is especially helpful for teams that are struggling with high volumes of manual reviews, inconsistent decisions between reviewers, and the challenge of scaling their operations without sacrificing quality.

Why manual review is hard to scale

As verification volumes grow, manual review teams run into the same problems:

• It's slow. Cases sit in queues waiting for a reviewer to have bandwidth. A case submitted at midnight waits until morning.
• It’s inconsistent. Two reviewers can look at the same case and reach different conclusions due to differences in judgment and interpretation.
• It doesn't scale cheaply. More volume means more headcount or more BPO spend, and neither solves the consistency problem.
• It creates audit risk. When decisions live in reviewers' heads rather than documented systems, it's hard to explain them to auditors or compliance teams.

Case review agents are built to break this tradeoff: faster review without sacrificing quality or control.

How case review agents work

Once a case is ready for review, the agent can begin its analysis immediately. The process follows a clear, auditable path:

1. A case is created: After a user completes an Inquiry or another event occurs that requires review, a new case is created in Persona.
2. The agent analyzes the information: The moment the case is ready, the agent begins its work. It has full access to the case data that sits within Persona’s platform, including:
   • Verification data (e.g., Government ID, Selfie, Database checks)
   • Reports (e.g., watchlist, adverse media)
   • Behavioral signals captured during the verification flow
   • Prior case history
   • Information from your own systems (via integrations)
3. A recommendation is made: Based on the rules you've configured, the agent makes a clear recommendation (i.e., approve, decline, or escalate for human review) and provides a structured summary explaining its reasoning. Every signal it considered and every conclusion it reached is documented.
4. Your team takes action: Depending on your setup, the recommendation can be used to automatically approve or decline the case, or it can be sent to a human reviewer for a final decision on complex cases. Case review agents can also leave comments or add tags to the case. Every action is logged for a complete audit trail.

Example: A KYB (Know Your Business) review

Here’s how a case review agent might handle a typical business verification case:

1. A business submits its information: A new business signs up and provides its formation documents, tax ID, and UBO information. A case is automatically created.
2. Persona's verification checks run automatically: The business's tax ID is checked against government databases, beneficial owners are screened against sanctions and watchlist reports, and formation documents are scanned for key details.
3. The case review agent gets to work: Once checks are complete and the case reaches a reviewable state, the case review agent is triggered and begins analyzing the submitted data.
4. The case review agent surfaces a recommendation: The case review agent finds the business is in good standing and beneficial owners are clear. It recommends "Approve" and generates a structured case summary with supporting findings and reasoning.
5. The case is automatically resolved: Because the recommendation met the pre-configured threshold for automatic approval, the agent resolves the case before it ever reaches a human reviewer. The business can be notified moments after submitting their information.

Additional features

• Integrations: Connect case review agents to tools like Slack so teams can receive case updates, review recommendations, and take action directly within existing tools. The agents can also pull in external context, like CRM deal stage data from Salesforce, so decisions reflect the full customer picture, not just identity data.
• Custom Risk Scoring: Case review agents can assign a risk score to each case based on your specific criteria, helping reviewers prioritize the highest-risk cases first.
• Automated Case Summaries: The agents generate structured summaries explaining their reasoning. This cuts down on manual documentation and ensures every decision is clearly explained.
• Built-in PII Controls: Sensitive customer data is protected at the platform level, with configurable access controls that ensure agents and reviewers only see the information they're authorized to use.
• Full case context and memory: Case review agents retain context across a case, so reviewers don't need to re-explain background information each time they interact with it.

Getting started: How we work with you

Setting up a case review agent is a collaborative process designed to ensure the agent makes decisions exactly the way your team would.

1. Guided Setup: We start with your Standard Operating Procedures (SOPs). Our identity experts work with you to understand your policies, risk thresholds, and how your team handles nuanced edge cases.
2. Rigorous Backtesting: Before a case review agent makes a single live decision, we test it against your historical case data. This process validates that the agent’s decisions align with the real-world decisions your team has made in the past. You see performance dashboards before launch, not after.
3. Iterative Deployment: After deployment, we continue to work with you, using a feedback loop to refine agent behavior as new fraud patterns emerge and your business needs evolve.

Operational guardrails

Persona provides tools to help companies address identity challenges, but it is up to each business to determine its risk tolerance and goals. Case Review Agents are designed with the following guardrails:

• Human-in-the-Loop by Design: You always have control. You can configure agents to handle routine decisions autonomously while routing high-stakes or complex edge cases to your team for a final human review.
• Complete Audit Trail: Every action an agent takes and every piece of information it analyzes is meticulously logged. This provides a clear, defensible record for compliance reviews and internal audits.
• Privacy and Data Controls: Agents operate within Persona's secure environment. They respect the same PII masking and data access controls that you set for your human team members.

Frequently Asked Questions (FAQ)

Do I need to be a technical user to set up a case review agent?

No. Our team of identity experts works with you in a hands-on process to configure and deploy agents based on your existing policies and historical data.

Can a case review agent learn and adapt over time?

Yes, the backtesting and feedback process is iterative. As your case data grows and you provide feedback, our team can work with you to refine agent behavior to adapt to new patterns and edge cases.

Can case review agents make final decisions automatically?

Yes, but only for case types where you've explicitly configured auto-resolution. By default, agents surface recommendations for human review. You control the level of automation.

How is this different from Workflows?

Workflows are great for automating "if-then" logic (e.g., if a Verification passes, send a webhook). Case review agents are designed for more complex decision-making that requires analyzing and synthesizing information from multiple sources to arrive at a judgment, similar to how a human reviewer operates.

Is my data used to train models for other customers?

No. Your data stays isolated to your organization and is never shared with or used to improve models for other Persona customers or LLM providers. When backtesting and calibration are performed, they use your historical case data solely to configure and tune the agent for your specific use case, not to train Persona's underlying AI models. All third-party AI providers are contractually bound via Data Processing Agreements (DPAs) to prohibit training on customer data.