Enhance approval decisions with AI-powered analysis
Use agentic approval to add AI-powered analysis to approval tasks. It analyzes each request against your policies and request data, recommends an approval option, and gives its reasoning to help approvers decide.
Prerequisites
Before configuring agentic approval, ensure you have:
- AgenticApproval license module — Contact our Sales team to enable this feature
- Configured approval task — A working approval task with defined options
- AI connection — An active AI connection that you configured in your system settings
- Policy documentation — Written approval policies and guidelines
How agentic approval works
If an approval sends a form back for rework, the AI Approval task evaluates the latest resubmitted form values instead of the original submission.
Agentic approval has two components:
Policy analysis
AI analyzes the request against your uploaded policy documents:
- Checks request data against defined policy rules.
- Identifies missing required information.
- Evaluates criteria based on policy guidelines.
- Considers request context and supporting documents.
Agent recommendation
The Agent provides a recommendation with:
- Recommended approval option — The option the AI recommends based on policy analysis.
- Detailed reasoning — Explanation of the factors the AI considered.
- Concerns — Any red flags or items requiring attention.
Configuring agentic approval
To set up agentic approval in your Approval task, follow these steps.
Enable AI assisted approval
After configuring your Approval task with options (refer to Approval task configuration), select Configure Task and navigate to the Agent Configuration tab.
To enable AI assisted approval:
- Toggle Enable AI assisted Approval to activate the feature.
- Select your AI Connection from the dropdown (for example, Anthropic | Sonnet 4).
To manage AI connections, go to AI sources under Settings in your system configuration.
Add policy documents
You can provide policy guidance to the AI by uploading files or by entering free text.
Upload policy files
Upload documents that define your approval criteria:
- In the Documents section, drag and drop files or click Drop files here or click to upload.
- Supported formats: PDF, DOCX, DOC, TXT, MD.
- You can upload multiple files for comprehensive coverage.
Upload instructions, rules, guidelines, and policies that the AI model should consider during its approval analysis.
Add free text policies
Alternatively, type or paste policies directly into the Add free text section:
- Enter your policy text in the text area.
- Use clear formatting with headings and bullet points.
- Include specific thresholds and requirements.
Example policy content:
### Residential loans- **Under $500,000**: Approve if rate >= 4.5% and terms <= 30 years- **$500,000 - $1,000,000**: Requires review for rate verification
### Commercial loans- **Under $1,000,000**: Approve if rate >= 5.5% and terms <= 20 years- Debt service coverage ratio: Minimum 1.3xSelect documents for analysis
After uploading documents:
- In the Uploaded documents section, review all available documents.
- Select which documents the AI should consider when analyzing approvals.
- Use Select all to include all uploaded documents.
Configure approval options
Name your approval options clearly and align them with your policies:
- Use descriptive option names (for example, “Approve”, “Review Required”, “Reject”).
- Ensure options correspond to policy outcomes.
- The AI recommends one of these configured options.
Using agentic approval as an end user
When an approver opens a task with agentic approval enabled, they see the AI recommendation card and the question and answer chat.
AI recommendation card
The recommendation card displays:
- Recommended decision — The suggested approval option.
- Policy used — Which policy file(s) the AI used to inform the decision.
- Reasoning — Bullet-point list of supporting factors.
- Concerns — Any red flags or items requiring attention.
Interactive Q&A chat
Approvers can ask the AI questions about the request:
- Click Ask Questions.
- Enter questions in natural language.
- AI responds using policy documents, request data, and attachments.
Example Q&A scenarios:
- “What are the key risk factors in this request?”
- “Why is the AI recommending rejection?”
- “What additional information would support approval?”
- “Does this request meet all policy requirements?”
Example use cases
The following scenarios show where agentic approval can support decision-making.
Purchase approval with spend limits
Configure policies for different spending tiers:
- Purchases under $500: Approve if in budget
- $500–$5,000: Require manager review
- Over $5,000: Multilevel approval required
The AI analyzes purchase requests against these thresholds and provides consistent recommendations.
Loan approval with compliance checks
Upload detailed lending policies with criteria for different loan types:
- Residential vs. commercial requirements
- Credit score thresholds
- Debt-to-income ratios
- Documentation requirements
The AI evaluates loan requests against these criteria and highlights any policy violations or concerns.
Travel request approval
Define travel policies with category-specific rules:
- Domestic vs. international travel
- Advance booking requirements
- Cost thresholds by travel type
- Exceptions for conferences or client meetings
The AI checks travel requests against these policies and identifies any exceptions or special circumstances.
Contract approval workflow
Configure approval criteria based on contract value and risk:
- Standard contracts under $10,000
- Non-standard terms requiring legal review
- Multiyear commitments
- Vendor compliance requirements
The AI evaluates contracts against these criteria and flags any non-standard terms or high-risk elements.
Best practices
The following practices help you get the most from agentic approval.
Policy document management
- Start simple — Begin with basic policies and expand over time.
- Clear criteria — Define specific thresholds and requirements.
- Examples — Include sample scenarios and expected decisions to help the AI understand context.
- Edge cases — Document exceptions and special circumstances.
- Structure — Organize policies by category or approval type.
- Concise language — Write policies in straightforward, unambiguous terms.
- Regular updates — Review and revise policies as business needs change.
- Version control — Keep historical versions of policy documents.
Using Q&A chat
- Ask clarifying questions — Use chat to understand AI reasoning.
- Verify understanding — Confirm AI interpretation of complex policies.
- Explore edge cases — Ask about unusual scenarios before making decisions.
- Request additional context — Get more details about specific concerns.
Troubleshooting
If you have issues with agentic approval, try the following:
- AI recommendation not appearing — Verify the AI connection is active and AI is enabled in the Agent Configuration tab of the Approval task. Check that your administrator enabled the AgenticApproval license module.
- Policy analysis seems incorrect — Review your uploaded policy documents for clarity and completeness. Ensure your policies define specific criteria and thresholds. Check logs to verify the system is loading policy files.
- Q&A chat not responding — Verify you configured the AI connection correctly and uploaded policy documents. Check that the system can access request data and attachments.
- Reasoning unclear or insufficient — Add more detailed policy documentation with specific criteria and examples. Ensure your policy files include clear guidelines for different scenarios.
While configuring this feature, refer to the Workflow Assistant guide for help answering questions about approval requests, tasks, files, or related workflow data.
Related guides
For additional approval task features, refer to:
- Approval task configuration — Basic approval task setup
- Approval task settings for user comments — Enable comment requirements
- Approval by email — Enable approvals via email links
- Enhancing workflow automation with forms — Display forms in approval tasks