Buying something routine in JD Edwards involves a surprising amount of research for a decision that is usually obvious. Which supplier did we use last time, what did we pay, did they deliver on time, is there an agreement covering this, what account does it code to.

All of that is in your system and all of it is reconstructed by hand every time. It is the clearest candidate for a drafting agent in the whole procurement function.

Supplier Recommendation
Purchase History
Lead Time Reality
Requisition Drafts
Order Drafts
Approval Routing
Audit Record

Recommending From Your Own Experience

The useful supplier recommendation is not which supplier is best in general. It is which supplier has served you best for this item, at this site, recently.

That is answerable from data you already hold, and it beats any external benchmark because it reflects how a supplier treats your business rather than how they present in aggregate.

Delivery record: Reliability for this item at this site, measured against original promised dates.
Price position: What you have paid, including the variance between order and invoice.
Real lead time: Order to receipt in practice rather than the quoted lead time, which is often optimistic.
Quality and exceptions: Return and receipt exception history where you record it.
ChatFin connects to JD Edwards and your wider ERP stack as the AI layer on top

Recommendations you can reason about

Taking a suggested purchase order at face value is the easy path. The value comes from seeing why the system suggested it: which history, which supplier record, which lead time.

AI is often the path of least resistance. There are times when there's something to be said for thinking through it and struggling a little bit.

Paul Barnhurst, on the AgentCFO podcast

Recommendations built from your own JDE history can be questioned line by line. A buyer who can see the reasoning makes a better call than one who just clicks approve.

Purchase History Analysis Before Buying

The questions a buyer asks before placing an order are consistent and each currently requires a lookup. Answering them all at once, before the order is drafted, is most of the value.

Last purchase: When, from whom, at what price and in what quantity.
Price trajectory: How the price has moved over time and whether the current quote is in line.
Alternative suppliers: Who else has supplied this item and how they compared.
Consumption: How much this site uses, which informs quantity rather than repeating the last order size.

Drafting a Requisition

A requisition draft is the safest place to start, because a requisition is a request rather than a commitment. Nothing binds anybody until it is approved and converted.

The draft should arrive complete: item, quantity, recommended supplier with the reasoning, expected price, account coding derived from how similar purchases have been coded, and the delivery location. The requester reviews rather than assembles.

Complete draft: Every field populated with a proposed value and the basis for it shown.
Coding proposed: Account and business unit from how this category has been coded at this site.
Supplier reasoning: Why this supplier, with the delivery and price evidence attached.
Nothing committed: A requisition is a request. Approval and conversion remain deliberate acts.
ChatFin automation agents executing controlled workflows

Purchase Order Drafts and the Non Negotiable Condition

Creating an order is a commitment to spend money, so the control requirement is absolute: the agent proposes and a named person with the right authority approves before anything is created.

When approved, creation must happen through a published orchestration that drives the application, so pricing, tax, account derivation, commitment tracking and approval routing all behave exactly as they would for an order keyed by a buyer. Writing order records directly would skip all of it.

Proposal first: The full order visible before creation, with every derived value shown.
Named approval: Within your existing authority matrix, by value and category.
Supported creation: Through an orchestration so business function logic and commitment tracking apply.
Reference returned: Order number returned on success so the action is traceable.

Where This Helps

The best candidates are the high volume, low judgement purchases that consume buyer time without benefiting from buyer expertise. Replenishment of routine items, repeat purchases from agreed suppliers, and low value requisitions that currently take longer to raise than they are worth.

Strategic sourcing is the opposite case. A new category, a significant contract or a supplier relationship change needs commercial judgement and negotiation, and an agent contributes analysis rather than a recommendation.

Routine replenishment: Known item, known supplier, known site. High volume, low judgement.
Repeat purchases: Buying the same thing again on established terms.
Low value requisitions: Where the raising effort currently exceeds the value of the purchase.
Not strategic sourcing: New categories and major contracts need people. The agent supplies evidence, not the decision.

"The agent should draft the order. It should never be the thing that decides your company is spending money."

ChatFin compared with generic AI agents

Controls Procurement Will Ask About

Purchasing controls exist because purchasing is a common route to loss, and any automation has to strengthen rather than route around them.

The design that achieves this is the same as elsewhere in finance. The agent operates within the requesting user's role, cannot approve what that person could not approve, and every draft and decision is recorded with its evidence.

Authority preserved: Existing approval thresholds apply unchanged, by value and category.
Segregation intact: The agent cannot both raise and approve, because it inherits one user's permissions.
Whitelisted actions: Only the specific orchestrations you permit can be called.
Complete record: The recommendation, its evidence, the draft, the approver and the resulting order number.

How ChatFin Does It

ChatFin recommends suppliers from your own JD Edwards delivery, price and lead time history, answers the purchase history questions before the order is drafted, and assembles a complete requisition or order draft with every derived value visible.

Creation happens through a published orchestration after a named approver confirms, so pricing, tax, account derivation and commitment tracking behave exactly as they do for a keyed order, and the full chain is retained.

ChatFin as the finance system of intelligence

Draft the Order, Approve the Spend

ChatFin recommends suppliers from your own performance history, answers the purchase history questions up front, drafts requisitions and purchase orders with every value shown, and creates them through a supported orchestration once approved.

Automate the research and the drafting. Keep the commitment decision with a person.

Book a Demo

Frequently Asked Questions

Can an AI agent create purchase orders in JD Edwards?

It can draft them, and creation should follow a named approval. When approved, the order must be created through a published orchestration that drives the application, so pricing, tax, account derivation, commitment tracking and approval routing all behave exactly as they would for an order keyed by a buyer. Writing order records directly would bypass all of that and produce commitments your reporting cannot rely on.

How does the agent decide which supplier to recommend?

From your own history rather than an external benchmark. Delivery reliability for that item at that site measured against original promised dates, what you have paid including order to invoice variance, real order to receipt lead time rather than the quoted figure, and any recorded return or receipt exception history. That reflects how a supplier treats your business specifically.

Where is this appropriate and where is it not?

It suits high volume, low judgement purchasing: routine replenishment, repeat buys from established suppliers, and low value requisitions that currently take longer to raise than they are worth. It is not appropriate for strategic sourcing. A new category, a significant contract or a supplier relationship change needs commercial judgement, and there the agent should supply analysis rather than a recommendation.

Related Articles