The AI Supplier PO Chasing Agent: How It Keeps Deliveries Honest in Odoo and MYOB Acumatica

Somewhere in your ERP right now there’s a purchase order (PO) with a promised date that quietly passed last Tuesday. Nobody changed it, the supplier never confirmed it, and the first person to find out will be whoever picks the sales order that needed that stock. Chasing suppliers is the least glamorous job in purchasing, which is exactly why it gets done late, in a rush, at the end of the week.

The appetite for handing this work to software is real. 78% of wholesale distribution businesses told MYOB they plan to use AI in their ERP, and the ABS found 12% of Australian businesses were already using AI in 2024-25, rising to 22% of medium-sized businesses. Supplier chasing comes up in those conversations constantly, partly because everyone hates the job and partly because the cost of skipping it is a stockout with a customer’s name on it.

What an AI supplier PO chasing agent actually does

An AI supplier PO chasing agent is a program that reads the open purchase orders in your ERP, works out which lines are unconfirmed, overdue or at risk, drafts the follow-up emails to those suppliers, reads the replies, and proposes the date changes so your system reflects what the supplier actually said. In a drafts-first design, that’s where its authority ends. A buyer approves every email before it sends and every date change before it lands, the same way we build every agent pattern we run.

The reading is the point. Plenty of tools can fire a reminder on a schedule. The chasing job is only half done when the email goes out; the other half is understanding “should be on the truck end of next week, the 40-footers are held up” and turning it into a proposed new receipt date, a flagged short shipment and a note on the record. That’s judgement work, and it’s the part language models are now good enough to draft, though not to own.

What Odoo and MYOB Acumatica already do without AI

Before you buy an agent, switch on what you already pay for.

Odoo 19 ships a receipt reminder on the vendor record: tick it and Odoo emails the supplier a set number of days before the expected receipt date, asking them to confirm. It’s part of the standard purchase module (the field is right there in Odoo’s source code), and together with the purchasing module’s RFQ and 3-way matching workflow it covers a decent share of the routine chasing. MYOB Acumatica comes at it through business events: Acumatica’s own worked example is a scheduled email listing every purchase order more than five days late against its promised date, built on a generic inquiry with no code.

Both are useful, both shipped, and both deterministic. They send on a schedule and they don’t read the answer. If the supplier replies “we can do half of it Thursday”, that email sits in someone’s inbox until a person deals with it. Neither vendor ships an AI layer for purchasing today. We checked the Odoo 19 Enterprise source again this week and none of its fifteen AI modules touch the purchase app, and MYOB Acumatica’s AI Studio is a tech preview that is user-initiated and can’t create records, with general availability planned for the 26 R1 release and no committed ANZ date. Anything that reads supplier replies and proposes date changes is wiring you add, not a feature you switch on.

Where the agent layer starts: reading the replies

The agent’s job begins where the reminders stop. Each morning it reads the open PO list and the purchasing inbox, and prepares three kinds of drafts.

• Chaser drafts. For unconfirmed or overdue lines, it drafts a follow-up in your tone, specific to the order: line numbers, quantities, the date you were promised. Suppliers with a reply already sitting in the thread don’t get chased again, which is the mistake a dumb schedule makes every time.
• Date change drafts. Where a supplier has replied, it extracts the new commitment and proposes the updated receipt date against the right lines, with the sentence it relied on quoted underneath so the buyer can check its reading in five seconds.
• Risk flags. It ranks what’s left by consequence, not by age: which late orders have sales orders or production waiting on them, which suppliers have a habit of optimistic promises, and which delays are worth escalating from a polite chaser to a phone call.

Those proposed dates matter well beyond purchasing. Your reordering logic runs on lead times, and stale receipt dates quietly poison it. A replenishment agent drafting purchase orders from bad dates just automates the shortage. Chasing is the pattern that keeps the rest of the data honest.

“The email is the easy bit. The judgement is knowing which three of your forty late orders will actually hurt you this week, and reading what the supplier really committed to, which is usually less than the sentence sounds like. That’s what the agent drafts. What it never does is decide on its own that a date moved.”

Bill Alvarez, Practice Manager, Auboros

A Thursday 4pm run, approval gates included

Here’s the shape of it in a wholesale business running about forty open POs. The buyer opens the queue the agent prepared. First, eleven chaser drafts: nine approved as written, one reworded because the supplier relationship is delicate, one deleted because that supplier phoned this morning. That’s gate one, and it took about eight minutes.

Next, four date change proposals from replies received overnight. Three quote the supplier’s sentence cleanly and get approved. On the fourth the supplier wrote “next week” on a Friday, and the agent guessed the earlier Monday; the buyer corrects it to the following Wednesday and approves. Gate two. Nothing touched the ERP until she did.

Last, one escalation flag: a supplier twelve days late on the order feeding your biggest customer’s backorder, with a drafted note to the account manager referencing the past three late deliveries. The buyer approves the note and decides to ring the supplier herself as well. Gate three. Half an hour, ten supplier emails, four date changes, one phone call. The agent did the reading; the buyer did the deciding.

The Australian rules, and the guardrails worth writing down

Emails to suppliers are business correspondence, but the accountability rules still shape the design. The ACMA’s position on electronic messages is blunt: your obligations can’t be outsourced, so if the agent sends it, your business sent it. And if a chaser ever drifts into promoting your products, you’re into Spam Act territory with its consent and unsubscribe rules, so keep the drafts factual and about the order. Your supplier contacts are also people: names, direct emails and mobile numbers are personal information under the Australian Privacy Principles, whether a person collected them or an agent did.

Inside the system, the guardrails are the usual ones and they’re not optional. The agent runs under its own named account with read access to purchasing and draft rights only, every draft and decision is logged, and the approval gates sit on both sends and date changes. We’ve written up the full control set, permissions, audit trails and the December 2026 privacy disclosure deadline in our guide to AI agent governance.

Failure modes, and how to start

The failure mode that keeps us disciplined is the hallucinated commitment: the agent records a delivery date the supplier never actually gave. It’s why every proposed change carries the quoted sentence it was based on, and why a buyer reads it before it lands. The second is over-chasing, because a reminder sent to a supplier who already answered damages the relationship at machine speed; suppression rules and the approval gate catch it. The third is optimism laundering, where “should be fine” becomes a confirmed date. Language models make mistakes, and the design has to assume it.

Start read-only. Run the agent for two weeks drafting into a queue nobody sends from, and compare its reading of the inbox against what your buyers actually did. If it earns trust, switch on approved sends and approved date changes, and nothing more. The same rule we apply to channel sync agents applies here: an agent with unattended write access isn’t faster, it’s just wrong in more places at once.

---

Late deliveries finding you before your system knows?

Auboros builds drafts-first supplier chasing agents over Odoo and MYOB Acumatica from Brisbane, working across Queensland and beyond. We start with the reminders your platform already ships and add the agent only where it earns its place.

If your buyers spend Friday afternoons chasing suppliers, book a free consultation. We’ll tell you honestly whether the native tools get you most of the way there.