VendorIQ

AI agents that analyze vendor contracts, research companies in real-time, score risk against 10 compliance policies, and log decisions to Slack and Airtable - weeks of procurement review in minutes.

Comment

Inspiration

Enterprise procurement teams spend weeks reviewing vendor contracts manually. A single missed clause - like an uncapped liability or a missing data protection agreement - can expose a company to millions in risk. We built VendorIQ to automate this entire process using a coordinated team of AI agents.

What it does

VendorIQ is a multi-agent due diligence pipeline deployed as a Slack bot. A procurement team member sends a vendor contract to VendorIQ in Slack, and within minutes receives a complete risk assessment covering:

  • Contract Analysis - Extracts key terms, obligations, financial commitments, and red flags from the full contract text
  • Vendor Research - Searches the web in real-time for the vendor's reputation, lawsuits, financial health, and public track record
  • Compliance Check - Evaluates the contract against 10 enterprise procurement policies (data protection, liability caps, termination rights, IP ownership, insurance, and more)
  • Risk Scoring - Synthesizes all findings into a CRITICAL/HIGH/MEDIUM/LOW risk score with a clear APPROVE/REJECT recommendation

After the reviewer approves or rejects, VendorIQ automatically posts the decision to a #vendor-approvals Slack channel and logs a full audit trail in Airtable.

How we built it

VendorIQ runs on the Airia platform as 6 coordinated agents:

  1. Orchestrator (Claude Sonnet 4.6) - Coordinates the full pipeline and manages the conversational approval flow
  2. Document Analysis Agent (Claude Sonnet 4.6) - Deep contract parsing and clause extraction
  3. Vendor Research Agent (Claude Sonnet 4.6 + Brave Search) - Real-time web research on the vendor company
  4. Compliance Check Agent (Claude Haiku 4.5) - 10-point policy compliance evaluation
  5. Risk Scoring Agent (Claude Haiku 4.5) - Multi-factor risk synthesis and recommendation
  6. Action Agent (Claude Haiku 4.5) - Executes Slack notifications and Airtable logging

The orchestrator calls all four analysis agents sequentially, presents the summary, then waits for a human decision before triggering post-approval actions via Slack and Airtable integrations.

Challenges we ran into

  • Agent-calling-agent via Airia's "Airia Agent" tool in post-approval nodes was unreliable - models would ask for confirmation instead of executing. We solved this by giving the orchestrator direct access to Slack and Airtable tools.
  • Getting Slack message formatting right took iteration - the model kept outputting markdown that doesn't render in Slack.
  • Balancing model selection between cost and reliability: Sonnet for complex reasoning tasks, Haiku for structured evaluation tasks.

What we learned

  • Multi-agent orchestration works best when sub-agents have narrow, well-defined responsibilities
  • Real-time web search adds massive value to vendor research - it caught CrowdStrike's July 2024 global outage and active Delta Air Lines lawsuit as risk factors
  • Human-in-the-loop approval is critical for enterprise trust - fully automated decisions would never get adoption in procurement

What's next for VendorIQ

  • Email notifications for stakeholders outside Slack
  • PDF report generation for audit archives
  • Integration with procurement systems like SAP Ariba and Coupa
  • Configurable compliance policies per organization
  • Historical vendor scoring to track risk over time

Built With

  • airia
  • airtable-api
  • brave
  • claude-haiku-4.5
  • claude-sonnet-4.6
  • slack-api
Share this project:

Updates