Every Odoo partner now talks about AI. Some promise large savings; some promise an ERP that "thinks". Business owners are left asking a simple question: what does it actually do on a normal Tuesday?
This article answers that plainly. It covers the AI features in Odoo 19, what Odoo 20 adds, and the limits you should plan for.
The AI features in Odoo 19
Ask AI. An assistant you can call from anywhere in Odoo. It answers questions about your records, summarises them and suggests next steps. It reads data; it does not change it.
Automations from plain language. You describe what should happen (for example "when an invoice is overdue by 15 days, send a reminder and notify the account manager") and Odoo drafts the automation rule. Someone still needs to check the rule before switching it on.
AI agents trained on your documents. You can point an agent at your own FAQs, policies or product information. It can then answer questions internally or on your website.
Website chat agent. A chatbot that answers visitor questions from your content and can create a CRM lead when someone shows buying interest.
Reading invoices and bills. Uploaded supplier bills and documents are read automatically, and the main fields are filled in for you to check.
AI reports and AI fields. You can ask for a report in plain English, and Studio can add fields that generate text such as summaries.
What Odoo 20 adds
Odoo 20 was unveiled in September 2026. The headline change is that AI agents can create and update records, not just read them. Odoo has also shown voice commands and AI help for routine bookkeeping and help desk replies.
That is a real step forward, and it raises the stakes. An assistant that only reads data can give a wrong answer. An agent that writes data can create a wrong invoice. (Check Odoo's official release notes for the final feature list for your edition.)
What AI can't do for you
It can't fix messy data. If customers are duplicated, products are coded three ways and half the vendor records are incomplete, AI summarises the mess faster. Clean data first.
It can't decide your process. AI can draft an approval rule. It can't tell you who should approve a discount above 10%. That is a business decision.
It can't replace testing. Every automation and every AI agent that writes data needs to be tested like any other configuration change, with real scenarios and sign-off.
It isn't free of cost or risk. Some features use credits or need an AI provider set up. Some send data to an outside AI service. Check what your plan includes and what your data policy allows.
Where to start
- Pick one painful, repetitive task, such as entering supplier bills or answering the same customer questions.
- Check the data behind it is clean enough to trust.
- Switch on one feature, test it with real examples for two weeks, and measure the time saved.
- Write down the rules: who can use it, what it may change, and who reviews its output.
- Only then extend to the next task.
An example from our own work. Before a version upgrade, we asked an AI tool to compare our test plan with past incident tickets. It pointed to areas where real problems had happened before but no test case existed yet. It didn't write the tests for us, and a person checked every suggestion before anything changed.
The same lesson applies to Odoo's own AI: it is most useful when you give it a narrow job, and someone checks what it produces.
How we can help
RNXOur Odoo AI Readiness review checks your data, picks the AI features that fit your business, and sets the governance rules before anything goes live.