Purchase Requisition Agent

The purchase requisition is the most expensive decision in the entire purchase-to-pay process - not because of its own value, but because it determines whether negotiated conditions ever take effect. Anyone who selects the wrong vendor here, orders outside the framework agreement or overlooks budget limits destroys savings before the first invoice enters the system. The Purchase Requisition Agent anchors framework agreements, budget authority and approval matrices directly in the ordering process - and releases human judgement only where it makes a difference: above the threshold.

Maverick Buying Costs Finance Organisations Up to 16 Percent of Negotiated Savings

The data is clear: according to APQC, organisations with high maverick buying rates need a median of 16 hours longer to issue a purchase order, and lose between 10 and 16 percent of savings between the negotiated framework and actual procurement. Hackett Group reports in its Digital World Class Benchmark 2025 that top performers lose 60 percent fewer savings through consistent contract compliance and reduced maverick buying compared to the average. The lever is not in procurement - it is at the moment of the purchase requisition. Whoever decides here decides on the margin.

In mid-sized Finance organisations, the reality looks like this: hundreds of requisitions per day flow through departments that neither know the framework agreements nor the remaining cost-centre budgets. The result: orders placed with the wrong vendor, call-offs without framework agreement reference, overruns that only surface at month-end. Each one is a lost negotiation.

The Decision Layer Breaks the Requisition into Seven Traceable Steps

The Purchase Requisition Agent cleanly separates the seven decisions in every requisition by responsibility. Demand detection from consumption data runs as ML-supported forecasting, for instance when minimum stock levels for consumables are breached. Budget check, vendor suggestion, framework contract conditions, approval workflow and order generation are fully rule-based - numerical comparisons and master data logic, no discretion.

Concretely: a department triggers a requirement for EUR 8,400 (USD 9,100) in maintenance material. The agent checks in under one second whether the cost centre has sufficient remaining budget, identifies the preferred vendor from master data, pulls framework contract conditions (price, minimum quantity, lead time), determines the approver via the approval matrix and generates the order. Contract compliance here is not the result of good intentions but the default.

The effect on cycle time: APQC benchmarks show that top performers achieve a requisition-to-order time of five hours while weak processes take around 48 hours. A rule-based Decision Layer brings routine orders into the minutes range - without sacrificing documentation.

Human Judgement Stays Where It Matters Commercially

Not every order can be processed rule-based. Individual orders above a defined threshold - say EUR 25,000 (USD 27,000) or a cost-centre-specific limit - require deliberate judgement: market conditions, strategic supplier relationships, extraordinary urgency. Here the agent steps back. It prepares the decision completely (budget status, contract availability, alternative vendors, approval chain), presents it to the authorised decision-maker and documents the human approval as an explicit step in the audit trail.

This principle is not a safety net - it is governance architecture: the human approval is an internal control element and part of the audit-compliant documentation obligation. Every purchase requisition - whether rule-based or human-approved - is fully traceable as a business transaction.

Framework Agreements Move from Paper to Infrastructure

The real gain lies beyond the individual order: the framework agreement engine the agent uses becomes shared infrastructure in the Decision Layer. The Contract Compliance Agent accesses the same contract logic. The approval matrix is reused by the Invoice Approval Agent and Payment Run Agent. The generated order becomes the reference for three-way matching between order, goods receipt and invoice. What begins as automation of a single task progressively builds the decision infrastructure on which the entire P2P process rests.