AI Document Analysis

Document analysis is PartnerAlly's core AI capability. When you upload a policy, procedure, or control document, AI reads it, understands its content, and identifies how it relates to your compliance frameworks.

How It Works

The Analysis Pipeline

What AI Looks For

Policy Content

AI identifies:

• Policy statements and objectives
• Scope and applicability
• Roles and responsibilities
• Approval and review dates

Control Implementation

AI recognizes:

• Technical controls described
• Process controls defined
• Administrative controls stated
• Compensating controls mentioned

Compliance Alignment

AI maps to frameworks:

• Which requirements are addressed
• How thoroughly they're covered
• What evidence supports compliance
• Where gaps exist

Document Types

Best Results With

Supported Formats

• PDF: Best for formatted documents
• Word (.docx): Good text extraction
• Text (.txt): Clean text, no formatting
• Markdown (.md): Structured text

Analysis Results

Mapped Controls

After analysis, you'll see:

• Which controls the document addresses
• Coverage level (full, partial, none)
• Confidence in the mapping
• Relevant excerpts from the document

Identified Gaps

AI reports gaps including:

• Framework requirement not addressed
• Control mentioned but incomplete
• Process described but not implemented
• Evidence needed but not found

Confidence Levels

Each finding includes confidence:

• High: Strong textual evidence
• Medium: Reasonable inference
• Low: Possible but uncertain

Understanding AI Reasoning

Evidence Citations

AI shows why it reached conclusions:

• Quotes from your document
• Section references
• Key phrases identified
• Control language matched

Reasoning Explanations

AI explains its logic:

• "This document describes an access review process, which addresses SOC 2 CC6.1"
• "No encryption policy was found for data at rest, creating a gap in ISO 27001 A.10.1.1"

Improving Analysis Quality

Document Quality

Better documents = better analysis:

Tips for Best Results

1. Use text-based PDFs

   • Not scanned images
   • OCR documents first if needed

2. Include full documents

   • Not summaries or excerpts
   • Complete policy content

3. Use clear headings

   • Section titles help AI navigate
   • Standard section names work best

4. Standard terminology

   • Common compliance terms
   • Avoid excessive abbreviations

Re-Analysis

When to Re-Analyze

Consider re-analyzing when:

• Document has been updated
• New frameworks are enabled
• Initial analysis had issues
• You want fresh results

How to Re-Analyze

Analysis Failures

Common Causes

Analysis may fail due to:

• Corrupt document file
• Password-protected PDF
• Image-only document
• Unsupported format
• Document too large
• System errors

Handling Failures

When analysis fails:

1. Check the error message
2. Verify document format
3. Try re-uploading
4. Convert to different format
5. Contact support if persistent

Multi-Document Analysis

How Documents Work Together

AI considers all your documents:

• Cross-references related policies
• Identifies coverage across documents
• Reduces duplicate gap findings
• Builds comprehensive picture

Document Relationships

AI may recognize:

• "Policy X references Procedure Y"
• "Standard A implements Policy B"
• "This document supplements the security policy"

Framework-Specific Analysis

Targeted Analysis

Analysis focuses on your enabled frameworks:

• SOC 2 Trust Service Criteria
• ISO 27001 Annex A controls
• HIPAA safeguards
• Other enabled frameworks

Control Mapping Depth

AI maps to framework specifics:

• Control objectives
• Implementation guidance
• Points of focus (where applicable)
• Sub-requirements

Viewing Analysis Results

Document Page

After analysis, see:

• Overall analysis status
• Framework coverage summary
• Linked gaps (if any)
• Confidence overview

Gaps Page

Related gaps show:

• Source document link
• AI confidence level
• Reasoning explanation
• Evidence from document

Technical Details

AI Models

PartnerAlly uses:

• Primary: Google Gemini for document understanding
• Fallback: Groq for redundancy
• Models optimized for compliance content

Processing Time

Typical analysis times:

• Short documents (1-5 pages): 30-60 seconds
• Medium documents (5-20 pages): 1-3 minutes
• Large documents (20+ pages): 3-5 minutes

Rate Limits

To ensure quality:

• Documents processed sequentially
• Bulk uploads queued automatically
• No limit on total documents

Common Questions

How accurate is the analysis?

Accuracy varies by document quality and content. High-confidence findings are typically 90%+ accurate. Always review results, especially for critical controls.

Can AI read handwritten documents?

No. AI works with text-based documents only. Handwritten content must be typed or transcribed first.

Does AI understand diagrams?

Limited. AI focuses on text content. Diagrams and flowcharts may not be fully interpreted.

How long are results stored?

Results are stored indefinitely until you delete the document or re-analyze.

Can I dispute AI findings?

Yes. AI findings are suggestions. You can mark gaps as false positives, change status, or add notes explaining disagreement.

Next Steps

• Chat Assistant - Ask questions about your documents
• Confidence Scores - Understand certainty levels
• Human Review - Validating AI findings