Trigger: An employee submits a new purchase requisition in Coupa.
Context/Data Pulled: The AI agent, via a webhook or scheduled job, retrieves the requisition details (items, quantities, supplier, total cost) and the requester's department, cost center, and approval history from Coupa's requisitions and users APIs.
Agent Action: The agent performs a multi-step analysis:
- Policy Check: Compares items against the company's approved supplier catalog and contracted pricing.
- Substitution Suggestion: If an item is off-catalog, it searches the Coupa catalog or approved supplier lists for compliant alternatives, using embeddings for semantic similarity.
- Routing Logic: Analyzes the requisition amount, requester role, and item category against the company's approval matrix (stored externally or in a custom Coupa object).
- Risk Flagging: Checks the supplier against an integrated risk database for any new financial or compliance alerts.
System Update/Next Step: The agent updates the Coupa requisition via API:
- Adds an internal comment with its findings (e.g., "Item is off-catalog. Suggested alternative: SKU #ABC123 from contracted supplier XYZ.").
- If a substitution is suggested, it can create a linked "suggestion" record for the requester to review.
- Sets a custom field with the recommended approver list or a risk score.
- The requisition is then routed to the correct approval workflow, either via Coupa's native routing or with a pre-populated approval list.
Human Review Point: The requester can accept or reject substitution suggestions before submission. Approvers see the agent's summary comment for context, speeding up their review.