Search

AI Review Agent

The AI Review Agent reviews files submitted through File Request actions inside active flows. It evaluates submissions against action metadata, configured criteria, and attached knowledge to determine whether requirements are met. Based on defined rules and confidence thresholds, it can reopen the request with feedback, advance or complete the action, or escalate to a human reviewer. This helps keep file review consistent and ensures submissions are complete and correct before the workflow progresses. This reduces avoidable review cycles and keeps document-driven workflows moving with consistent, governed checks.

 

Key Benefits

  • Saves internal review time by automatically advancing clean submissions and reopening requests with clear feedback when fixes are needed.

  • Keeps oversight in place by escalating uncertain cases to a human reviewer based on confidence thresholds.

  • Supports higher document volume without requiring the same increase in manual review capacity.

  • Standardizes file review using defined criteria and attached knowledge, rather than ad hoc checks.

  • Reduces rework by catching incomplete or incorrect submissions before they move forward.

 
 

Admin Configuration 



 

  1. Navigate to the Manage tab and click on AI Agents.
  2. Toggle on AI Review Agent.


     
  3. Go to Connections and select an AI Service:
    1. Moxo AI (default), or
    2. Connect an external LLM via API key (OpenAI, Google Gemini, Anthropic Claude).

       


       
  4. Click Save to apply the configuration.

 

Flow Template Configuration

The Review Agent is currently exclusive to File Requests, and can be configured directly within these actions by Flow Owners. 

To set up the Review Agent inside a flow:

  1. Open the desired Flow Template and click Edit.
  2. Select the desired File Request action.
  3. In Additional Options, toggle on Require File Review, and assign the AI Review Agent as a reviewer.


     
  4. Click the 'Configure' button to set tailored prompts and escalation thresholds.
    1. Add Additional Prompts if required.
    2. Under Escalation, define when a step should be routed to human review and select the reassignment role.


       
  5. Click Save to apply the configuration to the action.

The AI confidence threshold determines how certain the AI Review Agent must be before approving a submitted file. If the AI’s confidence score is above the threshold, the step proceeds automatically; if it is below, the action is escalated to a human reviewer. This ensures automation is balanced with the right level of human oversight.  

 

Client Experience

Here’s what assignees will experience when the AI Review Agent is active in a File Request step:


 

  • After users upload and submit their files, the Review Agent immediately begins evaluation and provides results in real-time.
  • If the files do not meet the requirements, the agent reopens the action and provides specific feedback on what to correct.

  • If submissions meet requirements, the agent completes the action or advances it to the next reviewer based on configuration.

  • If confidence falls below the defined threshold, the agent escalates the review to the assigned human reviewer.

    The agent provides review feedback is to the assignee immediately upon evaluation.  

Best practices

  • Keep review prompts concise and focused on objective file criteria.

  • Set clear escalation thresholds so borderline submissions route to human judgment.

    • Recommended confidence thresholds

      • 60% (lower threshold): Best for low-risk, high-volume steps with simple checks, where speed matters.

      • 70–80% (balanced): A strong default for moderately detailed requirements, such as onboarding or standard client submissions.

      • 90% (high threshold): Best for sensitive or regulated steps (compliance, KYC, financial documents, legal agreements) where accuracy is critical.

  • Review usage insights regularly to refine prompts and reduce unnecessary escalations.

  • Write clear File Request descriptions so the agent evaluates submissions against the right context and requirements.