AI Automation ·
If AI takes purchase requests, will approvals move faster too?
Coupa reported on customers using AI to intake and route requests across procurement, legal and IT. Korean companies can measure time saved by requesters separately from any reduction in the approval queue itself. [1][2]
Problem definition and research question
If an employee requests new software, procurement, legal and IT need different information. A single entry point saves work for the employee, but incomplete submissions or incorrect routing can force several teams to review them again. Reorders can also follow different approval paths from first-time suppliers. The practical question is whose time the new intake process saves and how it changes the queue and ownership at each handoff.
Evidence and causal analysis
On November 11, Coupa reported that Semrush cut request processing from 19 days to 10, while an unnamed database software company cut time spent entering a procurement ticket from 30 minutes to five. Coupa says its design enforces required approvals and records actions. [1] The 19-to-10-day result was already reported in a 2024 CPOstrategy interview. Tonkean's Semrush case also describes linking Oracle and the Ironclad contract platform. [2][4] The cases measure different stages, and the combination of integration and process changes makes it hard to isolate generative AI's contribution. Korean supplier-management vendor Emro connects purchase requests, orders and receipts and uses item master and contract data. A local comparison should therefore include how accurately each method finds the same item in the existing system. [3]
| Dimension | International evidence | Korean evidence and application |
|---|---|---|
| Entry channel | Coupa clarifies and routes plain-language requests [1] | Item and budget fields and workplace vocabulary in existing groupware |
| Approval evidence | Semrush connected purchasing and contract systems [2] | Emro uses item master and contract data [3] |
Comparison of alternatives
One option is to add required fields and routing rules to existing approvals. Procurement operations can maintain them at relatively low cost, though employees still have to choose the right category for mixed requests. A second option is AI intake that fills fields and recommends reviewers. That can reduce the work of managing multiple inboxes, while procurement operations maintains item and department codes and corrections, and IT maintains the integrations. The existing system retains control of who can approve or release a purchase order.
Implementation and evaluation
A Korean business unit could test recurring software renewals for four weeks, assigning similar requests to rules-based and AI-assisted intake. Record first-pass completeness, misrouting, 90th-percentile time to approval and reviewers' rework minutes. Include cancelled and rejected requests. As illustrative stop rules, the procurement owner suspends AI routing after any skipped mandatory approval or if misrouting rises by two percentage points against the baseline.
Conclusion and open questions
If writing the request gets faster while the long legal-review queue stays the same, where should the next budget go, toward AI intake or the team and process doing the review?