The AI Lead Qualification Agent: How It Works in Odoo and MYOB Acumatica

Updated 12 October 2026: added the short answer, an At a glance agent spec, four question headings and a captioned table of who does what; the Odoo section now reflects Odoo 20 (released 24 September 2026) and what its Enterprise source ships for lead creation; the MYOB Acumatica section now cites MYOB’s current AI feature list in place of the 2025 preview announcement; the ABS figure was re-sourced to the 25 June 2026 release; the ACMA unsubscribe wording was corrected to 5 working days; the 10 December 2026 privacy policy rule and “Goes well with” were added. The worked example and Josh’s quote are unchanged.

An AI lead qualification agent reads each new enquiry, checks it against your CRM and ERP history, scores it against criteria you wrote down, and drafts the first reply for a salesperson to approve. As at 12 October 2026, Odoo 20 ships predictive scoring and a livechat lead-creation skill, and MYOB Acumatica lists no lead qualification feature. The judgement layer is something you add.

At a glance

• Who it is for: sales managers and owners of Australian wholesale, distribution, trade and service businesses on Odoo or MYOB Acumatica that receive more enquiries each morning than a rep can read properly.
• Problem it solves: qualification is mostly reading and cross-checking history you already hold, and it gets skipped when the queue is long. The agent does the reading, ranks the queue and drafts the first reply.
• Platforms: Odoo 20 and Odoo 19 (CRM, Sales, Accounting, Live Chat) and MYOB Acumatica (CRM leads, customers and receivables). The agent is built alongside either platform and reads through its API under its own login.
• Trigger: a new lead from a web form, an email, a livechat conversation, an imported list or a missed-call transcript, usually batched for a morning queue.
• Inputs read: the enquiry itself, CRM lead and contact records, ERP customer history (orders, products bought, credit status), and your written scoring criteria.
• What the agent drafts: a score with the reasons for it, a routing recommendation, and a first-touch reply in your voice.
• Approval gate: a salesperson approves, edits or bins every draft. Nothing is sent to a prospect until a person has pressed send.
• Audit trail: each draft keeps the criteria it was scored against, the records the agent read and the rep’s decision, logged under the agent’s own credentials.
• Typical cycle time: the queue is ready before the first rep starts; the review in the example below takes less time than two leads once did by hand.
• What it never does: send an email or SMS, contact a lead from a bought or stale list, change a credit limit or a price, or decide that a lead is not worth a reply.
• Typical cost: a pilot or proof of concept from $5,000 ex GST, per the AI agents for ERP in Australia hub as at 12 October 2026; the agent’s login is a paid seat on both platforms.
• Typical timeline: discovery 1 to 2 weeks, sandbox pilot 2 to 6 weeks, drafts-first go-live 1 to 2 weeks, per the same hub page.
• Last verified: 12 October 2026 against Odoo’s 20.0 lead scoring documentation, Odoo’s 20.0 AI agents documentation, the Odoo 20 Enterprise source, MYOB’s AI feature page, Microsoft’s Sales Qualification Agent overview and the ACMA’s spam rules.

The agent stops at the draft and waits. That last part is the whole design, and it’s the part most vendor demos skip. The ABS Characteristics of Australian Business release of 25 June 2026 reports that 12% of Australian businesses used AI in 2024-25, and sales teams are a common starting point, because lead handling is repetitive, time-sensitive and mostly reading.

We’ve already covered how calls and meetings become CRM records and how emailed purchase orders become draft sales orders. This post is about the step before both: deciding which of this morning’s enquiries deserve a salesperson’s attention, and in what order.

What does an AI lead qualification agent actually do?

The agent has four jobs, and none of them involves sending anything.

• Read. The web form, the email thread, the trade show list, the missed-call transcript. It pulls out who’s asking, what they’re asking for, and how urgent they sound.
• Check your own records first. Before it looks anywhere else, it queries your CRM and ERP. Is this an existing customer? What have they bought? Are they on credit hold? Half of qualification is history you already own, sitting in systems your reps don’t have time to cross-reference at 7am.
• Score against explicit criteria. Your rules, written down: industry, region, order size signals, product fit. Not a vibe. If you can’t explain why a lead scored 80, the score is noise.
• Draft. A first-touch reply in your voice, a routing recommendation, and a one-paragraph note on why. All of it lands in a queue.

Then a person opens the queue, approves, edits or bins. The agent proposes. The rep decides. That split matters more here than almost anywhere else, because the thing being handled is a stranger’s first impression of your business.

Is predictive lead scoring the same as an AI lead qualification agent?

No, and if you run Odoo you already have the first one. Odoo’s 20.0 lead scoring documentation describes predictive lead scoring as a machine-learning model built on your historical won and lost deals, and says it is always active in Odoo CRM. Stage and Team are always in effect, and you can switch on State, Country, Phone Quality, Email Quality, Source, Language and Tags. Assignment rules can then route leads by probability. It comes with CRM, and you should turn on the variables that fit your pipeline.

But it’s statistics, not judgement. Predictive scoring won’t read the enquiry, won’t notice the sender is an existing debtor sixty days overdue, and won’t draft a reply. It ranks what’s in the pipeline from the fields the record carries. An agent sits on top of that layer and does the reading. The two stack well, as the table shows.

What do Odoo and MYOB Acumatica ship for lead qualification today?

Here’s the current state as at 12 October 2026, with shipped and not-shipped kept apart.

Odoo

Odoo 20 was released on 24 September 2026. Its 20.0 AI agents documentation lists the skills an agent can hold, says an agent with no skills can only provide information, and notes that a lead creation skill is available only when the CRM app is installed. We read the Odoo 20 Enterprise source on 12 October 2026 to see what that skill does. It is written for the website livechat agent: collect the visitor’s name, email and phone one at a time, confirm them, call the lead creation tool exactly once, and never promise that someone will be in touch until the lead exists. The skill’s own description says it is reserved for the livechat flow and should not be given to other agents, because its lead creation runs without a confirmation card.

That’s shipped: inbound capture with sensible guardrails. What Odoo doesn’t ship is the other half. Nothing in that source scores a lead against your written criteria, checks the sender’s order history and credit status, or drafts the reply. That is wiring you add, either with an agent inside Odoo or with an outside agent reading through the Odoo 20 MCP server, and every AI call inside Odoo 20 draws on prepaid credits, which we set out in what Odoo 20 AI costs.

MYOB Acumatica

The CRM handles lead records, assignment and pipeline today, all by rule. As at 12 October 2026, MYOB’s AI page lists eight AI features, among them an AI Assistant, AI Automation, Anomaly Detection and a Cross-Sell Assistant. None of them qualifies leads or drafts a first reply. The same page says MYOB Acumatica stores customer data in Australia and that AI providers cannot store it or use it to train their models, which matters when the data is a stranger’s contact details. For what each feature does, see our MYOB Acumatica AI guide.

The fair comparison

Microsoft is ahead of both here. Its Sales Qualification Agent in Dynamics 365 Sales (documentation last updated 30 April 2026) runs in two modes: research-only, where the agent researches each lead and writes an outreach email that a seller decides whether to send, and research and engage, where in Microsoft’s words the agent “autonomously engages with each lead”. It’s a capable product. We’d still tell you to run the first mode and not the second, for reasons the next two sections cover.

A Monday morning, worked through

This is the pattern as we’d build it for a Queensland wholesaler, drawn from what actual prospects keep asking us for.

Twenty-two leads arrived over the weekend: web forms, a trade show spreadsheet, four plain emails. By 7:30am the agent has matched six to existing accounts in the ERP. Two are current customers asking about a product they already buy, so it flags them for the account manager rather than the new-business queue. One is a customer on credit hold, flagged with a note to sort the account before anyone quotes them something new. The rest get scored against the written criteria, and eleven get a drafted reply. One enquiry the agent can’t confidently read, so it escalates with “a human should look at this” instead of guessing.

At 8am a rep opens the queue with coffee. Nine drafts approved as-is, two rewritten because the tone missed, one binned because the agent scored a student research project far too generously. Nothing was sent before a person hit approve, and the rep’s whole part took less time than the first two leads would once have.

“The agent is good at the reading and the arithmetic, and it never gets bored on lead eleven. What it doesn’t own is your reputation. The moment drafts start sending themselves, you’ve handed a stranger’s first impression of your business to a system that can’t be embarrassed.”

Josh Craig, Director, Auboros

Can an AI agent legally email leads in Australia?

Only inside the same rules that apply to a person, and this is where the approval gate stops being a preference and becomes protection.

Commercial emails and SMS in Australia sit under the Spam Act, and the ACMA’s rules are specific: you need consent, express or inferred, you must identify your business, and every message needs an unsubscribe option that is honoured within 5 working days. Someone who emailed you asking about a product has plainly invited a reply. A trade show list from two years ago is a different story, and it’s exactly the kind of list an over-eager automation will happily work through. The ACMA says to take care when you buy or use a marketing list, and that you still need consent from each person even if someone else sends the messages for you. If your agent sends it, your business sent it.

Privacy runs alongside. A lead’s name, email and behaviour are personal information under the Australian Privacy Principles, and that doesn’t change because an agent did the collecting or the enriching. From 10 December 2026, a privacy policy also has to describe decisions a computer program makes, or substantially helps make, where personal information is used and the decision could significantly affect a person, per the OAIC’s guidance of 30 September 2026. Ranking which enquiry gets a call first is unlikely to be that kind of decision. An agent whose score sets the credit terms or the price a sole trader is offered might be, so put that question to your privacy adviser before you build it. We covered the broader control set, permissions, audit trails and approval gates, in our guide to AI agent governance.

Failure modes worth designing for

We run agents over our own operations, drafts-first, and the failure modes are predictable enough to plan around. Research errors are the big one: two businesses share a name, and the agent confidently attributes the wrong size and industry to your prospect. That’s why the draft carries its reasoning, so a rep can see the error before it shapes a reply. Score inflation is the quiet one: agents reward enthusiasm rather than fit, and a keen student can outscore a terse procurement manager. Written criteria and a monthly look at what the agent got wrong keep the scores honest. Language models make mistakes. The design assumes it, which is why nothing sends itself.

Where to start

Start read-only. Two weeks of the agent scoring and drafting into a queue nobody sends from, while your reps work normally. Compare its calls with theirs. If it earns trust, switch on approved sends. If a vendor suggests skipping straight to autonomous outreach, ask them who answers to the ACMA when it goes wrong, because it won’t be them. Odoo users should also look at what Odoo CRM already does out of the box before adding an agent to it, and our AI agents for ERP in Australia page covers how we scope this pattern, drafts-first, with role-scoped permissions.

Goes well with

• The AI call prep agent: once a lead is approved, the one-page brief the rep reads before the first call.
• The AI customer reactivation agent: the same reading pointed at customers who stopped ordering, where consent is easier to show.
• What an AI agent costs to run in Australia: the seats, tokens and review time behind the pilot figure above.

Thinking about pointing an AI agent at your lead queue?

Auboros builds drafts-first qualification agents over Odoo and MYOB Acumatica from Brisbane, with the scoring criteria written down and a rep approving every send. If you’d like to see what the pattern looks like on your pipeline, book a free consultation. We’ll tell you plainly whether your lead volume justifies it.