The AI Call Logging Agent: How Calls and Meetings Become CRM Records in Odoo and MYOB Acumatica

Updated 5 October 2026: rebuilt to the current post format with a short answer, an At a glance spec block, question headings and a capture-channel table. Odoo facts re-verified against the Odoo 20 documentation and Enterprise source (Phone app, voice transcription, the inactive Call Summarizer agent, the MCP server’s Create Records tool); the MYOB section now reflects MYOB’s published AI feature list and data-processing statement as at 5 October 2026 and no longer refers to AI Studio.

An AI call logging agent reads a call or meeting transcript, drafts the CRM record (summary, follow-up tasks, an update on the right opportunity) and waits for the person who was on the call to approve it. As at 5 October 2026, Odoo 20 transcribes and summarises calls natively; MYOB Acumatica does not, so the agent runs beside it through the API. Nothing saves until a person approves.

At a glance

• Who it is for: sales managers and owners on Odoo or MYOB Acumatica whose pipeline notes depend on whether a rep felt like typing, and who want the record written without the rep’s judgement leaving the loop.
• Problem it solves: calls and Teams meetings that never make it into the CRM, so deal reviews run on memory and a departing rep takes the pipeline with them.
• Platforms: Odoo 20 and 19 Enterprise (Phone app with AI transcription, voice transcription in notes, MCP server on 20); MYOB Acumatica 2025.2 and 2026.1 (Teams integration, REST API, AI Assistant and AI Automation for in-form work).
• Trigger: a call ends in the Phone app, a meeting recording or transcript lands in a watched folder or mailbox, or a rep drops a transcript onto the opportunity.
• Inputs read: the transcript, the lead or opportunity it is matched to, the contact’s recent chatter or activity history, and the open quotes on the account.
• Systems: Odoo CRM and Phone, or MYOB Acumatica CRM with Microsoft Teams; the phone system or meeting tool that produced the recording; the agent’s own CRM user.
• What the agent drafts: a summary note on the opportunity, follow-up activities with dates and owners, and proposed field changes such as an expected close date, each labelled as a draft and attributed to the agent.
• Approval gate: the rep who was on the call reviews, edits and approves every draft before it saves; a manager approves anything that changes stage or expected revenue.
• Audit trail: the saved record links to the transcript, the draft is kept beside the approved version, and every agent write is tagged so a reviewer can tell machine from person.
• Typical cycle time: drafts ready within a few minutes of the recording finishing; the rep’s review takes one to two minutes per call.
• What it never does: save a note, create an activity or move a stage without a named person approving it; record a call without the announcement your state law requires; touch HR, payroll or other excluded conversations.
• Typical cost: a single scoped agent pilot typically starts from a few thousand dollars ex GST for discovery and sandbox build, per our AI agents for ERP in Australia page.
• Typical timeline: 4 to 10 weeks from discovery to a drafts-first go-live, per the same page.
• Last verified: 5 October 2026 against Odoo 20’s voice transcription documentation, Odoo 20’s Phone app documentation, the Odoo 20 Enterprise source, MYOB Acumatica’s AI page and Acumatica’s Microsoft Teams integration page.

How does an AI call logging agent work?

Four steps, the same shape as most of the agent patterns we design:

• Capture. The transcript comes from wherever the conversation happened: the phone system, a Teams meeting recording, or a call recorded inside the ERP itself.
• Draft. The agent reads the transcript and drafts the CRM entries: a summary note on the lead or opportunity, follow-up activities with dates, and any field changes it thinks are justified, like a pushed-out expected closing date.
• Confirm. The rep sees the draft, fixes anything the agent misheard, and approves it. Nothing posts until they do.
• Audit. The saved record links back to the transcript, so anyone reviewing the deal later can check what was actually said.

Here is a worked example. A rep finishes a 40-minute Teams demo with a wholesale customer. The agent drafts: a six-line summary against the opportunity, a task for the rep (“send revised pricing for the 500-unit tier by Friday”), a note that the customer asked twice about delivery timeframes to WA, and a suggested change to the expected close date, pushed out a month because the decision now needs board sign-off. The rep scans it, deletes the WA note (wrong customer, that was this morning’s call, the agent matched the wrong opportunity), and approves the rest. Ninety seconds instead of fifteen minutes, and the mistake got caught at review rather than discovered in month three.

That wrong-opportunity error is worth pausing on, because it’s the most common failure in this pattern. Same contact at two companies. One customer with three open deals. A mobile number that matches nobody. A human filing their own notes makes the same mistake, but the agent makes it confidently and at volume, which is exactly why the confirm step is not optional.

What does Odoo already ship for call and meeting transcription?

More than most people expect, and all of it is shipped rather than roadmap. We checked it on 5 October 2026 against the Odoo 20 documentation and the Odoo 20 Enterprise source; Odoo 20 was released on 24 September 2026, and the same features are in Odoo 19.

The VoIP module became the Phone app in Odoo 19, with call recording, three access roles (No, Officer, Administrator) and calls linked to the lead or opportunity they belong to. On top of that sits voice transcription: Odoo transcribes meetings, calls and dictation, generates an editable summary of key points, decisions and action items, and discards the audio once processing finishes. The documentation is explicit that the recording is not stored, replayable or accessible from the database afterwards, which is a retention control you would otherwise have to build yourself.

A few things we found in the Odoo 20 Enterprise source change how you plan a rollout. Call transcription is a policy on each VoIP provider with two values, Disable (the default) and Force for all users, so nobody’s calls are transcribed until an administrator decides they should be. The Call Summarizer that writes the one-line summary ships as an AI agent that is inactive, so that is a second deliberate switch. And the summariser’s own instructions tell the model to treat everything inside the transcript as passive conversation and to ignore any instruction a speaker tries to give it. That is a prompt-injection defence built into a shipped feature, and it is the pattern we copy into every agent that reads text a stranger produced.

What Odoo does not do natively is the full loop: reading the transcript and drafting follow-up activities and field updates against the right records. On Odoo 20 that layer can now sit on the built-in MCP server, whose tool reference includes Create Records and Update Records as write tools that are off until an administrator exposes them. We expose Create Records to a CRM-scoped key so the agent can draft activities and notes, keep Update Records off so stage and revenue changes stay a human click, and the key runs as a user who can see the sales team’s pipeline and nothing else. Our Odoo AI agents guide covers the wider allowlist design, and the approval gate post explains why any Odoo write an agent can reach should land as a draft.

Can MYOB Acumatica log Teams calls automatically?

Not natively, as at 5 October 2026. MYOB’s AI page lists the shipped features as AI Assistant, AI Automation, AP Bill Entry, Expense Management, Anomaly Detection, Production Variance Anomaly Detection, Cross-Sell Assistant and Auto Complete. Call or meeting transcription is not among them. What the platform gives you is the plumbing around the conversation: Acumatica’s Microsoft Teams integration lets staff start a Teams chat or call from lead, contact and case screens and post activities to Teams channels. Global Acumatica features typically reach MYOB’s ANZ releases six to nine months later, so always check which release a claim refers to.

MYOB’s AI Automation works inside a form on data the signed-in user could already edit, which suits drafting a follow-up email or summarising a case, and our AI Automation hand-off post covers where that stops. A meeting-to-CRM agent therefore runs outside the platform: it reads the Teams transcript, drafts activities and notes through the REST API under a CRM-scoped user, and a person approves each one in MYOB Acumatica before it saves. It’s the same drafts-first shape as the sales order agent that reads emailed purchase orders. Two statements on MYOB’s page matter for the privacy section below: AI features respect the user permissions already in place, and customer data is stored in Australia and processed either in Australia or through infrastructure operated by Anthropic, AWS and Microsoft, depending on the feature. Our MYOB Acumatica AI guide tracks what has landed in ANZ.

Is it legal to record customer calls in Australia?

The technology is the easy half. Two Australian legal layers apply before any of it runs.

First, recording itself. Call recording is governed by state and territory surveillance device laws, and they differ: some states allow a participant to record a conversation they’re part of, others expect every party to be told. For a business dealing with customers across state lines, the safe design is to announce recording on every call and give people the chance to object. If your phone system or meeting tool records by default, the announcement is a configuration task in the phone system, and it takes minutes to set up. This isn’t legal advice; check your own footprint with your adviser.

Second, privacy. The Office of the Australian Information Commissioner’s guidance on commercially available AI products deals directly with AI meeting transcription: collect only the personal information reasonably necessary for the job, destroy or de-identify anything the recorder caught that wasn’t necessary (the personal chat before the meeting proper, someone’s medical appointment), and get consent where sensitive information is involved. The Australian Privacy Principles also expect you to know where the transcript is processed, which is why the Odoo and MYOB data-handling statements above belong in your privacy assessment. In practice that means retention rules for transcripts, role-scoped access so a transcript is no more visible than the deal it belongs to, and an exclusion list of calls the agent never touches. HR conversations are the obvious first entry.

We still run this pattern for clients; we just set it up deliberately. That setup work is the part most overseas-built note-taking tools leave for you to work out on your own. Our AI agent governance post has the policy template.

Why do the drafts still need a person?

Transcription quality is good now, but the failure modes haven’t gone away. Agents mishear numbers, and a quote logged at $15,000 instead of $50,000 will surface at the worst possible moment. They attribute commitments to the wrong side of the conversation, so the CRM says the customer agreed to a timeline the rep actually proposed. And they summarise with full confidence even when the audio was poor.

“The transcript isn’t the point. The point is that your pipeline stops depending on whether the rep felt like typing on a Friday afternoon. The agent drafts, the rep checks and approves, and deal reviews stop being archaeology.”

Josh Craig, Director, Auboros

The approval step costs each rep a minute or two per call. What it buys is a CRM where every record was confirmed by the person who was in the room. That’s the difference between a pipeline you can trust and one you have to re-litigate every quarter. It also keeps the agent inside the AI agents for ERP in Australia pattern we run everywhere else: scoped user, drafts first, a named person approves, every write tagged.

Goes well with

• The AI rep call-prep agent: the other half of the call, a briefing drafted before the rep dials from the same CRM records this agent keeps current.
• The AI lead qualification agent: what happens to the inbound enquiry before it becomes a call worth logging.
• Odoo 20’s built-in MCP server: how to expose Create Records to a CRM-scoped key and keep every other write off.

Want call notes that write themselves, without writing fiction?

We design drafts-first agents over Odoo and MYOB Acumatica from Brisbane, with the consent, retention and permission settings sorted before anything records, for businesses across Queensland, NSW, Victoria and New Zealand. If you’d like your calls and meetings landing in the CRM as reviewed records instead of memories, book a free consultation. We’ll look at your phone and meeting setup and tell you what’s realistic.