# AI agent checks every purchase requisition before it becomes a purchase order: quote, contract, price history and policy (Tier-1 industrial manufacturer, Bayer)

> AI purchase requisition review: a Tier-1 manufacturer's agents handle 99.6% of requisitions after a year; Bayer cut manual errors about 70%.

- Canonical URL: https://aiagentsformanufacturing.com/use-cases/ai-purchase-requisition-review-agent-policy-check-to-po-tier1-bayer/
- Department: Procurement (https://aiagentsformanufacturing.com/departments/procurement/)
- Evidence level: Verified. Real deployment at a named or precisely described company, with a measured result from a primary source.
- Added: 2026-11-11
- Last updated: 2026-11-11
- Publisher: AI Agents for Manufacturing (https://aiagentsformanufacturing.com)
- Manufacturers named: Bayer (https://aiagentsformanufacturing.com/companies/#bayer)
- AI tools deployed: Pactum Alignment Agent (https://aiagentsformanufacturing.com/tools/#pactum-alignment-agent), Lio (https://aiagentsformanufacturing.com/tools/#lio)

## Role

Purchasing operations (requisition review, purchase order creation and procurement compliance). Typical users: purchasing manager, buyer, procurement operations or shared-services team, plant controller

## The problem

Before a requisition becomes a purchase order, someone has to check it: is a valid quote attached, does a framework contract apply, is the price in line with previous orders, does the cost centre match the account, are the required forms there? When something is missing, the reviewer emails the requester, waits, follows up and reviews again. At a global Tier-1 industrial manufacturer described by Lio (more than $25 billion in revenue and 100,000 employees), every requisition went to a reviewer in the requester's region, about 170,000 times a year, work that took time away from negotiating and sourcing. At Bayer's Global Business Services, staff logged into SAP to validate purchase requests and check vendor details and material codes by hand; about 30% of purchase requests were not automated, and 36% of those were under EUR 5,000.

## How AI helps

The agent sits in the approval chain like a technical approver. At the Tier-1 manufacturer, Lio's agent runs about 60 checks against the company's global standard (procurement channel, previous orders, price threshold, contract type, payment terms and so on); when everything checks out the requisition becomes a purchase order without a buyer touching it, and when something is missing, such as an attachment, the agent chases it. When a price is above previous orders or a quote is missing, negotiation and sourcing agents benchmark the price and open a negotiation or sourcing event with the supplier. Cases the agents cannot close reach a person with a summary and a recommendation, and every check and decision is logged. At Bayer, a software robot logs into SAP, checks that each purchase request follows the rules, assigns vendor details and creates the purchase order with predefined rules; exceptions such as missing vendor details or incorrect pricing are resolved with generative AI. SUEZ in the UK (environmental services, not a manufacturer) uses Pactum's agent inside Coupa to check every requisition for category, quote and rate-card compliance and correct it on the spot.

## Step by step

1. Write down the checks a reviewer really makes today, in order, and count how many requisitions bounce and why (missing quote, wrong approver, no contract reference, wrong cost centre).
2. Turn your purchasing policy into explicit rules with thresholds: which items need a quote, which must use a framework contract, what price variance against past orders is acceptable.
3. Start with low-value, repeat purchases in one site or region, and let the agent check and recommend while a buyer still approves.
4. Have the agent ask requesters for missing items directly and log every check, so buyers only see exceptions with a short summary.
5. Move to automatic conversion into purchase orders only for categories where the agent's recommendations matched the buyers' decisions over several weeks.
6. Track cycle time, share of requisitions handled without a buyer, rework loops and price variance, and audit a sample of automated orders each month.

## Tools

- Lio (multi-agent procurement system for requisition review, negotiation and sourcing): used at a Tier-1 industrial manufacturer
- Intelligent automation robots in SAP with generative AI for exceptions: used at Bayer GBS
- Pactum Alignment Agent (requisition checks inside Coupa): used at SUEZ UK
- Other tools to evaluate: the guided-buying and approval features of your ERP or procure-to-pay suite (SAP Ariba, Coupa, Zip and others), intake-to-procure tools such as Fairmarkit
- To test the idea without new software: a sample of recent requisitions and your purchasing policy, plus any LLM (ChatGPT, Claude, Copilot, Gemini) with the prompt below

## Result

Tier-1 industrial manufacturer (not named), Lio: agents reviewing and approving three out of four requisitions on their own within six months, 99.6% after twelve months; review cycle time from weeks to seconds (the headline box says hours); roll-out region by region (Americas, Western Europe, Eastern Europe, China and APAC, India), with the other regions onboarded in six weeks instead of hiring new reviewers (Lio customer story, a vendor publication). Bayer GBS: manual errors down about 70%, 60% fewer mistakes on vendor and pricing details, and time from purchase request to purchase order down from 10 minutes to 3, with a target of 1 minute (IRPA AI case study, February 2026). SUEZ UK, Pactum: cycle time down 90% from days to minutes, 80% of manual reviews eliminated and more than 10% of buyer time freed (Pactum testimonial, a vendor publication).

As reported by the source; most figures are vendor or self-reported and not independently verified.

## Human review: cautions

- The Tier-1 manufacturer is not named and its results come from the vendor; SUEZ is an environmental services company, not a manufacturer; Bayer's figures come from an industry association case study.
- An agent that turns requisitions into orders can also spread a policy mistake fast: keep thresholds, an audit sample and a way to stop it.
- Automation only works on clean master data (suppliers, contracts, price history, cost centres); many projects spend most of their time there.
- Requesters should still know who to call: keep a named buyer for each category.

## Try it yourself (prompt)

```text
You are a purchasing operations analyst at a manufacturer. I want to see which requisition checks could be done by an AI agent before a buyer looks at them.

I will paste: (1) our purchasing policy [thresholds, quote rules, framework contracts, approval limits]; (2) 20 to 50 recent requisitions [item, quantity, price, supplier, attachments, cost centre, requester, what happened to it]; (3) the reasons requisitions were sent back last quarter, if we track them.

Do this:
1. Turn the policy into a numbered checklist of yes/no checks an agent could run, and mark which ones need data we may not have (price history, contract list).
2. Run the checklist on each requisition and give a table: pass, fail with the reason, or cannot tell.
3. Group the failures and tell me which three causes create the most rework.
4. Draft short, polite messages to requesters for the three most common missing items.
5. Suggest which purchase categories are safe to automate first and which should always stay with a buyer.

Do not approve or reject anything: this is a recommendation for my review. Use only the data I paste in; if something is missing, write 'not found' or list it as a question instead of guessing. Do not send, order, post or change anything in any system: produce drafts for my review and approval.
```

## Related use cases

- [AI agents negotiate with long-tail suppliers that buyers never had time to call (Honeywell, and separately an anonymous German consumer and industrial goods maker)](https://aiagentsformanufacturing.com/use-cases/ai-agent-tail-spend-supplier-negotiation-honeywell-pactum/)
- [Agent checks supplier invoices against contract terms to stop leakage](https://aiagentsformanufacturing.com/use-cases/invoice-contract-compliance-agent/)
- [AI supplier contract review and renewal tracking](https://aiagentsformanufacturing.com/use-cases/ai-supplier-contract-review-renewal-tracking/)
- [AI collects supplier emissions data and builds product carbon footprints for Scope 3 reporting (Western Digital, PALFINGER, Knorr-Bremse, Vishay)](https://aiagentsformanufacturing.com/use-cases/ai-supplier-emissions-data-product-carbon-footprint-western-digital-palfinger/)

## Sources

- https://www.lio.ai/customers/t1-global-manufacturer
- https://content.irpaai.org/p/new-post-deab
- https://pactum.com/testimonial/suez

---
Please cite the canonical URL: https://aiagentsformanufacturing.com/use-cases/ai-purchase-requisition-review-agent-policy-check-to-po-tier1-bayer/
