The AI Customer Reactivation Agent: Winning Back Dormant Customers in Odoo and MYOB Acumatica

Updated 15 October 2026: rebuilt to the current post format with a short answer, an agent spec block, a who-does-what table and a fifth FAQ. The Odoo section now covers Odoo 20 (Marketing Campaign Builder agent, Mass Mailing prompt, MCP server), the MYOB section follows MYOB’s current AI page, the ACMA unsubscribe window is corrected to 5 working days, and the ABS figure is re-sourced to the 25 June 2026 release.

An AI customer reactivation agent reads sales and service history in Odoo or MYOB Acumatica, finds accounts that have gone quiet against their own ordering rhythm, and drafts a win-back email for each, reason attached. It sends nothing. A salesperson reviews the queue, removes the names the data cannot explain, and approves the rest. Under the Spam Act 2003, consent is checked before a word is drafted.

At a glance

• Who it is for: sales managers and owners of Australian wholesale, distribution and B2B service businesses on Odoo or MYOB Acumatica with a few hundred to a few thousand trading accounts, where “who stopped ordering?” is a question nobody has time to answer.
• Problem it solves: dormant accounts are cheaper to recover than new ones to win, but spotting them against each account’s own cycle and writing something specific is tedious enough that it never gets done.
• Platforms: Odoo 19 and Odoo 20 Enterprise (Sales, CRM, Marketing Automation; in Odoo 20 the built-in MCP server for read access) and MYOB Acumatica (CRM dynamic lists and campaigns, the REST API). Both ship the rule-driven half; the judgement layer is wiring you add.
• Trigger: a scheduled run, typically quarterly or monthly, plus an on-demand run a sales manager can start before a pipeline meeting.
• Inputs read: sales orders, invoices, unconverted quotes, support tickets, call notes filed by a call logging agent, the consent and unsubscribe records, credit holds and open disputes.
• Systems: the ERP’s read API under a role that can read sales, service and contacts and write nothing (in Odoo 20, an MCP key with the read-only tools; in MYOB Acumatica, an API user with a read-only role), and a model on your approved-tool list.
• What the agent drafts: a ranked queue of quiet accounts, two or three lines of reasoning per name citing the records by ID, and a draft email that refers to what the customer actually bought. No discount unless a person adds one.
• Approval gate: one, and it is the whole design. Nothing is sent, scheduled or logged to the CRM until a named salesperson approves each draft.
• Audit trail: each draft stores the records it cited and the consent check it passed; approvals, edits and rejections are logged against the contact in the ERP, with the rejections kept as notes for the next run.
• Typical cycle time: the run finishes in minutes; a 60-account shortlist takes a sales manager about an hour to work through, against the half day of report trawling it replaces.
• What it never does: send email, change a contact’s consent status, apply a discount, create or close an opportunity, or contact anyone on a credit hold or in an open dispute.
• Typical cost: a scoped pilot from $5,000 ex GST per our AI agents for ERP in Australia hub, as at 15 October 2026.
• Typical timeline: one read-only run compared against what the sales team already suspects, then approved sends for a small segment; the hub’s pattern is discovery 1 to 2 weeks, sandbox pilot 2 to 6 weeks, drafts-first go-live 1 to 2 weeks.
• Last verified: 15 October 2026 against the Odoo 20 Enterprise source (Marketing Automation templates, the Marketing Campaign Builder agent, AI Mass Mailing), MYOB’s AI page, Acumatica’s marketing lists documentation, the ACMA’s spam guidance and the ABS AI adoption release of 25 June 2026.

Every ERP holds a list nobody enjoys reading: the customers who ordered every month for years and then stopped. Winning one of them back usually costs less than landing a new account, but the work is tedious enough that it rarely gets done. Someone has to trawl the sales history, work out who has gone quiet, remember why, and write something worth replying to. An agent can do the trawling. The deciding and the sending still need a person, for reasons this post gets into.

We’ve covered the lead qualification agent that decides which new enquiries deserve attention. Reactivation is the same pattern pointed the other way, at customers you already won and then lost track of.

What does an AI customer reactivation agent actually do?

The agent has four jobs, and sending email is not one of them.

• Read the history you already own. Sales orders, invoices, quotes that never converted, support tickets, the meeting notes in the CRM. Dormancy only means something against context, and the context is sitting in your ERP.
• Measure silence against each account’s own rhythm. A cafe that ordered weekly and hasn’t for six weeks is dormant. A machinery buyer with an eighteen month cycle isn’t dormant at twelve. A fixed “no orders in 90 days” report treats both the same, which is why those reports get ignored.
• Explain each pick. Two or three lines per account: how long quiet, what they used to buy, anything in the service history that might explain the silence. If the agent can’t say why a name is on the list, the name shouldn’t be there.
• Draft the outreach. An email in your voice that refers to what the customer actually bought, not a generic “we miss you”. Every draft lands in a queue with the records it was built from, and waits.

Adoption is past the experiment stage. The ABS reported that 12% of Australian businesses used AI in 2024-25, released 25 June 2026, and dormant-customer lists are one of the requests we hear most from wholesale and distribution clients. The appeal is obvious: the data is already there, and the downside of a bad draft that a human catches is zero.

What do Odoo 20 and MYOB Acumatica already ship for win-back?

Both platforms ship the deterministic half of this pattern today, and it’s worth switching on before you add any AI to it.

Odoo Enterprise includes the Marketing Automation app with a gallery of campaign templates. In the Odoo 20 Enterprise source, Create Repeat Customers still targets confirmed online orders above a set value, thanks the buyer and follows up after the first open, and Double Opt-in exists purely to collect provable consent. They are rule-driven: fixed triggers, fixed timers, no reading of context. Odoo 20 adds two AI pieces around them. A Marketing Campaign Builder agent turns a plain-language goal into the steps of a campaign you have already opened, and Mass Mailing gains an AI prompt that explains why people opt out of a list. Neither reads an account’s order history to decide who has gone quiet, and neither writes a per-customer email. In the source as at 30 September 2026 that judgement layer is still wiring you add, now more easily, because the Odoo 20 MCP server gives an outside agent scoped read access without custom code. If you’re weighing up the platform side first, our guide to what Odoo CRM covers for Australian businesses is the place to start.

On the Acumatica platform that MYOB Acumatica is built on, dynamic marketing lists rebuild themselves from criteria you define, including purchase history, and campaigns track revenue against the members of each list. These are longstanding CRM features. On the AI side, MYOB’s AI page as at 15 October 2026 lists eight features, including an AI Assistant for plain-language questions about customers and revenue, AI Automation for approvals, document processing and pricing, and a Cross-Sell Assistant that suggests products on a sale. None of them is described as finding dormant accounts or drafting outreach, and MYOB says automations “are best configured by power users or administrators”. A reactivation agent over MYOB Acumatica today is external tooling reading through the API, with the CRM’s own lists and campaigns doing the deterministic work.

The split matters. The list of quiet accounts is a query. Deciding which silences are worth breaking, and how, is the part that needs either a salesperson’s time or an agent drafting for a salesperson’s approval.

What does a quarterly win-back run look like?

A Queensland wholesale distributor runs the agent on the first Friday of each quarter. It pulls every account whose last order is more than twice its usual reorder gap: 61 accounts. It removes anyone with a credit hold, an open complaint in the service queue, or an unsubscribed contact, which leaves 38. For each one it writes its two lines of reasoning and a draft email that names the products the customer used to buy and asks one specific question, with no discount attached.

The sales manager works through the queue before the Friday sales meeting. She rejects six: one company went into administration in March, two are mid-dispute over a freight claim the agent could see but had no business weighing, and three are accounts she knows moved to a competitor on price, which makes them phone calls, not emails. She edits four drafts and approves 28. The agent schedules the sends, logs who approved what, and keeps her six rejections as notes for next quarter’s run. The agent found the silence. She decided which of it to break.

“Any report writer can list the customers who stopped ordering. The hard part is knowing which silences are recoverable and which ones you earned. The agent drafts that distinction for you. It never gets to decide it.”

Bill Alvarez, Practice Manager, Auboros

Who decides who gets contacted?

The table below is the design we scope, and the column that matters is the last one.

Is a win-back email legal without fresh consent?

The part no reactivation vendor leads with: in Australia, a win-back email is a commercial electronic message under the Spam Act 2003, and the ACMA’s guidance is blunt about what that requires. You need consent, express or inferred, before you send. Inferred consent depends on an ongoing relationship with your business and marketing related to it, the ACMA calls it less reliable than express consent, and it says plainly that it does not cover messages sent just because someone bought something from you. An account that hasn’t ordered in three years is the exact case where inferred consent is at its weakest, which makes reactivation the riskiest email in your marketing calendar, not the safest.

The practical consequences for agent design fall out of the rules. The burden of proving consent sits with you, so the agent’s first check on every account is the consent record, before a word gets drafted. Every message needs a working unsubscribe that is honoured within 5 working days, stays functional for at least 30 days, and demands no fee, login or extra personal information. And responsibility can’t be outsourced: the ACMA says you must still hold consent even when someone else sends your messages for you, and an AI agent is the strongest version of “someone else” yet invented. If personal information feeds the profiling, the Australian Privacy Principles apply to that too. Our AI agent governance guide covers the approval gates and audit trails that make all of this defensible, including the automated decision-making disclosure that applies from 10 December 2026.

What goes wrong, and where do you start?

The failure modes here are specific, and each one shapes the design.

• Invented history. Language models make mistakes, and a draft that references an order the customer never placed is worse than no email at all. The guardrail is structural: drafts may only cite records by ID, and the approver sees those records beside the draft.
• The complaint you forgot. Some accounts went quiet because of you: a botched delivery, a pricing dispute. The agent reads the service history, but tickets don’t capture everything, and a warm “it’s been a while” to that customer reads as not listening. This is the single best argument for the human gate, because the sales manager recognises the name even when the data doesn’t.
• The contact who left. B2B contacts churn fast, and a personal note addressed to a buyer who resigned two years ago lands on their replacement’s desk as noise. Contact hygiene is a prerequisite, not an afterthought.
• Training customers to wait. Blanket win-back discounts teach your regulars that going quiet earns 15% off. Keep offers out of the first round entirely and let the approver decide, account by account, whether one is warranted.

Starting is simpler than the list above suggests. Run the agent read-only for a fortnight and compare its shortlist with what your sales team already suspects. If the reasons hold up, switch on approved sends for a small segment and watch two numbers: replies and unsubscribes. A reply rate without an unsubscribe spike means the consent judgements are sound. And if a vendor offers to skip the approval step in the name of scale, the ACMA’s position on outsourced responsibility is worth re-reading before you sign. Our AI agents for ERP in Australia hub covers how we scope this pattern: drafts-first, consent checked before drafting, role-scoped access, every send approved.

Goes well with

• The AI lead qualification agent: the same pattern pointed at new enquiries instead of lost accounts.
• The AI rep call prep agent: what happens when a reactivated account says yes to a call.
• AI usage policy for Australian businesses: the staff-facing rules that say who may approve an outbound draft.

Wondering which of your quiet accounts are worth waking?

Auboros builds drafts-first reactivation agents over Odoo and MYOB Acumatica from Brisbane, working across Queensland and beyond, with a person approving every send. If you’d like to see your dormant list with reasons attached, book a free consultation. If your list turns out to be too small to justify an agent, that’s exactly what we’ll say.