C
Docs

Requirement Extraction

How Cothon's AI extracts and categorizes requirements from procurement documents

Updated 2025-01-153 min read

Requirement Extraction

Cothon uses advanced AI to automatically extract, categorize, and prioritize requirements from RFPs and procurement documents. This page explains how the extraction works and how to get the best results.

How It Works

When you upload a document, Cothon's AI pipeline:

  1. Parses the document - Extracts text while preserving structure and formatting
  2. Identifies requirements - Distinguishes actual requirements from background information
  3. Categorizes each requirement - Groups by type (technical, compliance, experience, etc.)
  4. Assigns priority - Determines mandatory vs. nice-to-have requirements
  5. Extracts key details - Captures quantities, dates, certifications, and specifications

Note

The AI is trained on thousands of government procurement documents and understands common RFP structures and terminology.

Requirement Categories

Extracted requirements are organized into these categories:

CategoryDescriptionExamples
TechnicalProduct/service specificationsPerformance specs, features, standards
ComplianceRegulatory and legal requirementsCertifications, licenses, security clearances
ExperiencePast performance requirementsYears in business, similar contracts, references
PersonnelStaffing requirementsKey personnel, qualifications, certifications
FinancialCost and pricing requirementsBudget limits, pricing format, payment terms
DeliveryTimeline and logisticsDeadlines, milestones, delivery locations
AdministrativeProcess requirementsSubmission format, page limits, required forms

Priority Levels

Each requirement is assigned a priority:

  • Mandatory - Must be met to be considered compliant (indicated by "shall", "must", "required")
  • Evaluated - Scored during evaluation (indicated by "should", "preferably", "desirable")
  • Informational - Context or background (no compliance impact)

Warning

Pay special attention to mandatory requirements. Missing even one can disqualify your proposal.

Improving Extraction Quality

Get better results with these tips:

Document Quality

  • Use searchable PDFs - Scanned images require OCR which reduces accuracy
  • Avoid password protection - Remove passwords before uploading
  • Include all appendices - Requirements often hide in attachments

Review and Refine

After extraction, you can:

  • Split compound requirements - Click to break multi-part requirements into separate items
  • Merge duplicates - Combine requirements that appear multiple times
  • Adjust categories - Drag requirements to different categories if miscategorized
  • Change priority - Override AI-assigned priorities when needed

Handling Complex Documents

For large or complex RFPs:

Extraction Confidence

Each requirement shows a confidence score indicating how certain the AI is:

  • High (90%+) - Clear requirement language, unambiguous
  • Medium (70-89%) - Likely a requirement but may need verification
  • Low (<70%) - Uncertain, review the source text carefully

Tip

Focus your review time on medium and low confidence items. High confidence extractions are usually accurate.

Next Steps

Was this page helpful?

Requirement Extraction | Cothon Docs | Cothon