The AI Call Prep Agent: How It Briefs Sales Reps from Odoo and MYOB Acumatica

A sales rep sits in the car park outside a customer’s office in Toowoomba with eleven minutes to spare. She scrolls her inbox for the last email thread, finds a quote from May, and walks in. What she doesn’t know: the account went on credit hold on Friday, two invoices are 47 days overdue, and a return has been sitting in the service queue since mid August. The ERP knew all of it. Nobody asked.

An AI call prep agent asks. It reads your ERP and CRM before the meeting and hands the rep one page: trading history, money owed, open quotes, unresolved problems, and a few suggested talking points clearly labelled as suggestions. It sends nothing to the customer and changes nothing in the system. The rep walks in knowing what the business knows.

What an AI call prep agent does

An AI call prep agent is a read-only agent that assembles a briefing before a customer meeting. It pulls sales history, receivables status, open quotes, delivery and service issues, and recent correspondence from your ERP and CRM, then drafts a one-page brief with suggested talking points. It proposes. The rep decides what to use.

Most call prep tools on the market work from the outside in. They build briefs from meeting transcripts, email threads and public company research, and Oracle now ships a meeting preparation agent in its sales cloud that summarises CRM context the same way. Useful tools. If your business lives entirely inside a CRM, one of them may be enough.

For a business that runs on an ERP, the outside-in brief misses the part that decides the meeting. A transcript tool can tell you what was said last time. It can’t tell you that margin on the account has slipped since January, that the last two orders shipped short, or that finance put the account on hold. That information lives in Odoo’s CRM and accounting records or in MYOB Acumatica, and it changes what a sensible rep says in the first five minutes.

A useful ERP-drawn brief covers five things:

• Trading position: revenue over the past 12 months, order frequency, and whether volume is up or down on the year before
• Money: overdue invoices with amounts and days outstanding, the credit limit, any hold, and how the account usually pays
• Open threads: quotes waiting on a response and when they expire, backorders, lines ordered but not yet delivered
• Service history: open tickets, recent returns, complaints that never got a closing note
• Last contact: who spoke to the account, when, and what was promised

On top of that sit the suggested talking points. They’re the least reliable part of the brief and should be labelled that way. The facts above them are what earn the agent its place.

What Odoo and MYOB Acumatica give you today

Odoo 19 shipped real building blocks for this. You can configure AI agents with your own instructions and let staff query them, and AI fields can fill a summary field on a record from a prompt over that record’s data. Both work today on a current Odoo 19 Enterprise database.

What doesn’t exist is a packaged call prep feature. We checked the v19 Enterprise source code again this run: 14 AI modules ship with the platform, and none of them joins CRM, receivables and logistics into a briefing. That layer is wiring you add, an agent that reads across modules through the API under the rep’s own login. Odoo Experience runs 24 to 26 September and the v20 announcements may move this, but treat anything said on stage as roadmap until it ships.

MYOB Acumatica holds the data side well. Customer screens carry receivables, credit terms and sales history, which is exactly what a brief needs. On the AI side, AI Studio is a technology preview in the ANZ product: user-initiated, unable to create records, with general availability planned for the 26 R1 release. Planned means planned, not shipped. Global Acumatica is further ahead, with an AI assistant under managed availability in its 2026 R1 release, and global features typically reach the ANZ product 6 to 9 months later. For call prep specifically, both platforms are in the same position: the records are there, the briefing layer is something you build.

A worked example, with the human in the right place

Here’s the pattern as we’d build it. Each weekday at 6:30am, the agent reads the day’s calendar for each rep. For every customer meeting it finds, it pulls that customer’s records under that rep’s permissions, drafts a one-page brief, and posts it to the rep’s chat.

One brief might read: credit hold applied 29 August. Invoices 10412 and 10547 total $9,180, both past 45 days. Open quote Q-2214 for $18,400 expires Friday. Return RMA-388 waiting on inspection since 18 August. Suggested opener: the delivery problem, not the upsell.

The rep reads it in three minutes and decides what to use. Maybe she opens with the return. Maybe she knows something the system doesn’t and ignores the suggestion entirely. That’s the design working. After the meeting, the only write-back on offer is a drafted activity note, and she approves it before it touches the CRM.

Three rules make this pattern safe to run:

• The agent uses the rep’s own data scope. Never a shared admin login. A super-user briefing agent means every rep’s morning chat can contain every customer’s financials, which is a data breach waiting for a screenshot.
• Every line links to its source record. If the brief says an invoice is overdue, one click shows the invoice. The rep can check anything in seconds.
• Write-backs are drafted, even trivial ones. A logged note seems harmless until the agent logs the wrong meeting against the wrong account.

“The most expensive meeting is the one where the customer knows more about the relationship than your rep does. The ledger already has the answers. The agent’s job is making sure someone reads them before the handshake, not after.”

Josh Craig, Director, Auboros

Where it goes wrong

Language models make mistakes, and a briefing is a soft place to hide them. We’ve seen summaries that read a credit note as an unpaid invoice, and a complaint closed in March described as ongoing. That’s why every line links back to a record. The brief is a draft, and the links are what make checking it quick.

The second failure is staleness. The brief is only as good as the CRM hygiene underneath it. If nobody has logged a call since last year, the agent will confidently brief your rep on last year. Fixing data habits comes before wiring an agent, which is the same order we recommend in our lead qualification and call logging work. The call logging agent is actually the natural partner here: one writes the record, the other reads it back out.

The third is privacy. A brief contains personal information about the people your rep is meeting, so the Australian Privacy Principles apply to how it’s produced, stored and shared. Keep briefs inside your own systems, don’t paste them into public AI tools outside your policy, and resist the temptation to have the agent scrape social profiles into your CRM. Our AI agent governance guide covers the permission and audit patterns in detail.

None of this should put you off. A read-only agent is the lowest-risk way to start with AI over your ERP, and the field is emptier than the vendor noise suggests: the ABS found 12% of Australian businesses used AI in 2024-25, rising to 22% of medium-sized businesses. A rep who walks in briefed is still unusual. That’s the opportunity.

---

A better brief than an inbox search

We design read-only briefing agents over Odoo and MYOB Acumatica from Brisbane, for sales teams across Queensland and beyond, as part of our AI agent services. Everything is drafts-first with the rep in charge of the call.

If your reps walk into meetings without knowing what finance knows, book a free consultation. We’ll tell you whether an agent earns its keep here, or whether a saved report would do the same job for less.