The AI Credit Management Agent: How It Works Inside Odoo and MYOB Acumatica

Chasing late payments is nobody’s favourite job, and it’s one of the first places Australian businesses ask us about AI agents. The work is repetitive, the data already sits in the ERP, and doing it badly shows up as a cash flow problem within a quarter.

An AI credit management agent is software that reads your accounts receivable data, watches how each customer’s payment behaviour changes over time, and drafts the responses: reminder emails, credit limit recommendations, hold requests and call lists for the credit team. The word drafts matters. In every design we build through our AI agent services, the agent proposes and a person approves. Nothing reaches a customer, and no account goes on hold, until a human has looked at it.

This post covers what a credit management agent actually does, what your ERP already handles without any AI, and the guardrails that keep automated collections on the right side of Australian rules.

What an AI credit management agent actually does

Think of the agent as a credit officer’s assistant that never gets bored of the aged receivables report. Five jobs, roughly in order of value:

• It watches behaviour, not just balances. An aged trial balance tells you who is overdue today. The agent tracks the trend underneath: a customer who has drifted from paying in 30 days to paying in 55 over six months is a bigger risk signal than a single invoice a week late.
• It flags risk early. Part payments where full payments used to arrive, a broken payment promise, orders growing faster than the payment record justifies. These patterns are visible in ERP data long before an account lands in the 90 day column.
• It drafts the follow-ups. Instead of a template blast, the agent writes each reminder with context: the invoices involved, the customer’s history, what was agreed last time. A person reads, edits and sends.
• It recommends credit decisions. When a customer trips a risk threshold, the agent drafts the recommendation: reduce the limit, ask for prepayment on the next order, or place the account on hold. The decision stays with a human, because a wrong hold blocks a real order from a possibly loyal customer.
• It prepares the credit call sheet. Each morning the credit controller gets a prioritised list: who to call, what they owe, what they said last time, and a suggested talking point. That’s an hour of report wrangling replaced with judgement work.

What your ERP already does without AI

Here’s the part most AI content skips: a good share of credit control is already built into the ERP you own, and it isn’t AI at all.

Odoo 19 ships follow-up levels that schedule payment reminders by days overdue, escalate through email, SMS or post, attach the overdue invoices, and can run automatically once configured. Each level can also schedule an activity for a responsible user, and the chatter keeps a record of every reminder sent. That’s shipped functionality today, no AI involved.

MYOB Acumatica inherits the Acumatica platform’s credit controls: credit limits enforced at order entry and invoicing with a warning or a hard block, dunning letters, overdue charge calculation, parent and child account credit relationships, and role-based control over who can see or change customer balances. As usual with MYOB Acumatica, newer platform features tend to reach ANZ releases 6 to 9 months after the global Acumatica release, so check what your version includes.

If none of that is switched on in your system, start there. A surprising share of “we need AI for collections” conversations turn out to be “nobody ever configured follow-up levels”. The agent belongs on top of a working process, not instead of one.

Where the agent earns its keep

Rule-based dunning treats every overdue invoice the same. Day 7 gets email one, day 14 gets email two, day 30 gets the stern letter. That works until it doesn’t: your biggest customer gets the same stern letter as a habitual non-payer, over a $180 invoice they’ve already disputed.

The agent’s job is the judgement layer. Here’s a worked example from the pattern we design for wholesale and distribution businesses.

Overnight, the agent reads the aged receivables, payment history and open orders from the ERP under its own read-only login. It notices a customer whose average days to pay has moved from 32 to 61 across the last five invoices, while their order volume has nearly doubled. Exposure is growing while payment performance is falling, which is the pattern that precedes most bad debts.

By 7am, the credit controller has three drafts waiting: a reminder email referencing the specific invoices and the customer’s usual payment pattern, a recommendation to ask for a part payment before releasing the two open orders, and a note for the account manager, since the growth story might justify a different conversation. The controller softens the email, approves the part payment condition, rejects the hold, and it’s all done before the first coffee. The agent wrote everything. It sent nothing and decided nothing.

“The aged receivables report never tells you the story. The agent’s job is to read five reports at once and say: this account has changed. A person still decides what to do about it, and that’s how it should be.”

Bill Alvarez, Practice Manager, Auboros

The rules that keep AI collections safe in Australia

Collections is regulated conduct, which is a concrete reason to keep humans in the loop rather than a nice-to-have.

• Debt collection conduct rules. The ACCC and ASIC publish a joint debt collection guideline covering how, and how often, debtors may be contacted. An unsupervised agent firing reminders on a loop could cross lines a person wouldn’t. Draft and approve keeps contact frequency and tone under human control.
• Privacy. Payment histories are personal information when your customers are sole traders or individuals, so the Australian Privacy Principles apply to how that data moves through any AI workflow. From 10 December 2026, privacy policies must also disclose the kinds of substantially automated decisions that significantly affect people, and a credit decision is a textbook example. One more reason the agent recommends and a person decides.
• Approval gates on anything that acts. Credit holds, limit changes and every outbound message pass through a person. A clumsy reminder is embarrassing; a wrong credit hold stops a genuine order and sours a relationship you spent years building.
• An audit trail. Every recommendation the agent makes, and every approval or rejection, gets logged. When someone asks in June why a customer went on hold in March, there’s an answer.

How to start without betting your ledger

The rollout pattern we use for reconciliation agents applies here almost unchanged.

Run the agent read-only for the first month, observing and reporting, so you can judge its risk flags against what your credit controller already knows. Test in a sandbox before it touches production data. Scope its ERP login to customer, invoice and payment data and nothing else, the same role-scoped permissions approach we recommend for every agent. Then switch on drafting, keep every approval human, and review a sample of its recommendations weekly.

None of this waits on vendor roadmaps. Our own practice runs the read, analyse, draft pattern with external agents working through each platform’s APIs under scoped logins. If you want the platform-by-platform picture of what ships natively, we’ve covered AI agents in Odoo and MYOB Acumatica’s AI features separately.

---

Thinking about a credit management agent for your business?

We design drafts-first credit control agents over Odoo and MYOB Acumatica from Brisbane, with role-scoped access and human approval on every send and every hold. If the aged receivables report keeps growing and the chasing isn’t keeping up, book a free consultation. We’ll tell you what’s worth automating and what your ERP already does out of the box.